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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  22  */
  23
  24 #include "config.h"
  25
  26 #include <stdio.h>
  27 #undef _GNU_SOURCE
  28 #include <string.h>
  29 #include <stdlib.h>
  30 #include <dirent.h>
  31 #include <errno.h>
  32 #include <unistd.h>
  33 #include <signal.h>
  34 #include <fcntl.h>
  35 #include <termios.h>
  36 #include <sys/param.h>
  37 #include <sys/file.h>
  38 #include <sys/mount.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/socket.h>
  42 #include <sys/prctl.h>
  43 #include <sys/types.h>
  44 #include <sys/wait.h>
  45 #include <sys/un.h>
  46 #include <sys/poll.h>
  47 #include <sys/syscall.h>
  48
  49 #if HAVE_SYS_CAPABILITY_H
  50 #include <sys/capability.h>
  51 #endif
  52
  53 #if !HAVE_DECL_PR_CAPBSET_DROP
  54 #define PR_CAPBSET_DROP 24
  55 #endif
  56
  57 #include "start.h"
  58 #include "conf.h"
  59 #include "log.h"
  60 #include "cgroup.h"
  61 #include "error.h"
  62 #include "af_unix.h"
  63 #include "mainloop.h"
  64 #include "utils.h"
  65 #include "lxcutmp.h"
  66 #include "monitor.h"
  67 #include "commands.h"
  68 #include "console.h"
  69 #include "sync.h"
  70 #include "namespace.h"
  71 #include "apparmor.h"
  72 #include "lxcseccomp.h"
  73 #include "caps.h"
  74
  75 lxc_log_define(lxc_start, lxc);
  76
  77 static int match_fd(int fd)
  78 {
  79         return (fd == 0 || fd == 1 || fd == 2);
  80 }
  81
  82 int lxc_check_inherited(struct lxc_conf *conf, int fd_to_ignore)
  83 {
  84         struct dirent dirent, *direntp;
  85         int fd, fddir;
  86         DIR *dir;
  87
  88 restart:
  89         dir = opendir("/proc/self/fd");
  90         if (!dir) {
  91                 WARN("failed to open directory: %m");
  92                 return -1;
  93         }
  94
  95         fddir = dirfd(dir);
  96
  97         while (!readdir_r(dir, &dirent, &direntp)) {
  98                 if (!direntp)
  99                         break;
 100
 101                 if (!strcmp(direntp->d_name, "."))
 102                         continue;
 103
 104                 if (!strcmp(direntp->d_name, ".."))
 105                         continue;
 106
 107                 fd = atoi(direntp->d_name);
 108
 109                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 110                         continue;
 111
 112                 if (match_fd(fd))
 113                         continue;
 114
 115                 if (conf->close_all_fds) {
 116                         close(fd);
 117                         closedir(dir);
 118                         INFO("closed inherited fd %d", fd);
 119                         goto restart;
 120                 }
 121                 WARN("inherited fd %d", fd);
 122         }
 123
 124         closedir(dir); /* cannot fail */
 125         return 0;
 126 }
 127
 128 static int setup_signal_fd(sigset_t *oldmask)
 129 {
 130         sigset_t mask;
 131         int fd;
 132
 133         /* Block everything except serious error signals */
 134         if (sigfillset(&mask) ||
 135             sigdelset(&mask, SIGILL) ||
 136             sigdelset(&mask, SIGSEGV) ||
 137             sigdelset(&mask, SIGBUS) ||
 138             sigdelset(&mask, SIGWINCH) ||
 139             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 140                 SYSERROR("failed to set signal mask");
 141                 return -1;
 142         }
 143
 144         fd = signalfd(-1, &mask, 0);
 145         if (fd < 0) {
 146                 SYSERROR("failed to create the signal fd");
 147                 return -1;
 148         }
 149
 150         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 151                 SYSERROR("failed to set sigfd to close-on-exec");
 152                 close(fd);
 153                 return -1;
 154         }
 155
 156         DEBUG("sigchild handler set");
 157
 158         return fd;
 159 }
 160
 161 static int signal_handler(int fd, void *data,
 162                            struct lxc_epoll_descr *descr)
 163 {
 164         struct signalfd_siginfo siginfo;
 165         siginfo_t info;
 166         int ret;
 167         pid_t *pid = data;
 168         bool init_died = false;
 169
 170         ret = read(fd, &siginfo, sizeof(siginfo));
 171         if (ret < 0) {
 172                 ERROR("failed to read signal info");
 173                 return -1;
 174         }
 175
 176         if (ret != sizeof(siginfo)) {
 177                 ERROR("unexpected siginfo size");
 178                 return -1;
 179         }
 180
 181         // check whether init is running
 182         info.si_pid = 0;
 183         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 184         if (ret == 0 && info.si_pid == *pid) {
 185                 init_died = true;
 186         }
 187
 188         if (siginfo.ssi_signo != SIGCHLD) {
 189                 kill(*pid, siginfo.ssi_signo);
 190                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 191                 return init_died ? 1 : 0;
 192         }
 193
 194         if (siginfo.ssi_code == CLD_STOPPED ||
 195             siginfo.ssi_code == CLD_CONTINUED) {
 196                 INFO("container init process was stopped/continued");
 197                 return init_died ? 1 : 0;
 198         }
 199
 200         /* more robustness, protect ourself from a SIGCHLD sent
 201          * by a process different from the container init
 202          */
 203         if (siginfo.ssi_pid != *pid) {
 204                 WARN("invalid pid for SIGCHLD");
 205                 return init_died ? 1 : 0;
 206         }
 207
 208         DEBUG("container init process exited");
 209         return 1;
 210 }
 211
 212 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 213 {
 214         handler->state = state;
 215         lxc_monitor_send_state(name, state, handler->lxcpath);
 216         return 0;
 217 }
 218
 219 int lxc_poll(const char *name, struct lxc_handler *handler)
 220 {
 221         int sigfd = handler->sigfd;
 222         int pid = handler->pid;
 223         struct lxc_epoll_descr descr;
 224
 225         if (lxc_mainloop_open(&descr)) {
 226                 ERROR("failed to create mainloop");
 227                 goto out_sigfd;
 228         }
 229
 230         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 231                 ERROR("failed to add handler for the signal");
 232                 goto out_mainloop_open;
 233         }
 234
 235         if (lxc_console_mainloop_add(&descr, handler)) {
 236                 ERROR("failed to add console handler to mainloop");
 237                 goto out_mainloop_open;
 238         }
 239
 240         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 241                 ERROR("failed to add command handler to mainloop");
 242                 goto out_mainloop_open;
 243         }
 244
 245         if (handler->conf->need_utmp_watch) {
 246                 #if HAVE_SYS_CAPABILITY_H
 247                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 248                         ERROR("failed to add utmp handler to mainloop");
 249                         goto out_mainloop_open;
 250                 }
 251                 #else
 252                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support\n");
 253                 #endif
 254         }
 255
 256         return lxc_mainloop(&descr, -1);
 257
 258 out_mainloop_open:
 259         lxc_mainloop_close(&descr);
 260 out_sigfd:
 261         close(sigfd);
 262         return -1;
 263 }
 264
 265 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 266 {
 267         struct lxc_handler *handler;
 268
 269         handler = malloc(sizeof(*handler));
 270         if (!handler)
 271                 return NULL;
 272
 273         memset(handler, 0, sizeof(*handler));
 274
 275         handler->conf = conf;
 276         handler->lxcpath = lxcpath;
 277
 278         apparmor_handler_init(handler);
 279         handler->name = strdup(name);
 280         if (!handler->name) {
 281                 ERROR("failed to allocate memory");
 282                 goto out_free;
 283         }
 284
 285         if (lxc_cmd_init(name, handler, lxcpath))
 286                 goto out_free_name;
 287
 288         if (lxc_read_seccomp_config(conf) != 0) {
 289                 ERROR("failed loading seccomp policy");
 290                 goto out_close_maincmd_fd;
 291         }
 292
 293         /* Begin by setting the state to STARTING */
 294         if (lxc_set_state(name, handler, STARTING)) {
 295                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 296                 goto out_close_maincmd_fd;
 297         }
 298
 299         /* Start of environment variable setup for hooks */
 300         if (setenv("LXC_NAME", name, 1)) {
 301                 SYSERROR("failed to set environment variable for container name");
 302         }
 303         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 304                 SYSERROR("failed to set environment variable for config path");
 305         }
 306         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 307                 SYSERROR("failed to set environment variable for rootfs mount");
 308         }
 309         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 310                 SYSERROR("failed to set environment variable for rootfs mount");
 311         }
 312         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 313                 SYSERROR("failed to set environment variable for console path");
 314         }
 315         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 316                 SYSERROR("failed to set environment variable for console log");
 317         }
 318         /* End of environment variable setup for hooks */
 319
 320         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 321                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 322                 goto out_aborting;
 323         }
 324
 325         if (lxc_create_tty(name, conf)) {
 326                 ERROR("failed to create the ttys");
 327                 goto out_aborting;
 328         }
 329
 330         /* the signal fd has to be created before forking otherwise
 331          * if the child process exits before we setup the signal fd,
 332          * the event will be lost and the command will be stuck */
 333         handler->sigfd = setup_signal_fd(&handler->oldmask);
 334         if (handler->sigfd < 0) {
 335                 ERROR("failed to set sigchild fd handler");
 336                 goto out_delete_tty;
 337         }
 338
 339         /* do this after setting up signals since it might unblock SIGWINCH */
 340         if (lxc_console_create(conf)) {
 341                 ERROR("failed to create console");
 342                 goto out_restore_sigmask;
 343         }
 344
 345         INFO("'%s' is initialized", name);
 346         return handler;
 347
 348 out_restore_sigmask:
 349         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 350 out_delete_tty:
 351         lxc_delete_tty(&conf->tty_info);
 352 out_aborting:
 353         lxc_set_state(name, handler, ABORTING);
 354 out_close_maincmd_fd:
 355         close(conf->maincmd_fd);
 356         conf->maincmd_fd = -1;
 357 out_free_name:
 358         free(handler->name);
 359         handler->name = NULL;
 360 out_free:
 361         free(handler);
 362         return NULL;
 363 }
 364
 365 static void lxc_fini(const char *name, struct lxc_handler *handler)
 366 {
 367         /* The STOPPING state is there for future cleanup code
 368          * which can take awhile
 369          */
 370         lxc_set_state(name, handler, STOPPING);
 371         lxc_set_state(name, handler, STOPPED);
 372
 373         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 374                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 375
 376         /* reset mask set by setup_signal_fd */
 377         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 378                 WARN("failed to restore sigprocmask");
 379
 380         lxc_console_delete(&handler->conf->console);
 381         lxc_delete_tty(&handler->conf->tty_info);
 382         close(handler->conf->maincmd_fd);
 383         handler->conf->maincmd_fd = -1;
 384         free(handler->name);
 385         if (handler->cgroup) {
 386                 lxc_cgroup_destroy_desc(handler->cgroup);
 387                 handler->cgroup = NULL;
 388         }
 389         free(handler);
 390 }
 391
 392 void lxc_abort(const char *name, struct lxc_handler *handler)
 393 {
 394         int ret, status;
 395
 396         lxc_set_state(name, handler, ABORTING);
 397         if (handler->pid > 0)
 398                 kill(handler->pid, SIGKILL);
 399         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 400 }
 401
 402 #include <sys/reboot.h>
 403 #include <linux/reboot.h>
 404
 405 /*
 406  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 407  * in a child pid namespace if container reboot support exists.
 408  * Otherwise, it will either succeed or return -EPERM.
 409  */
 410 static int container_reboot_supported(void *arg)
 411 {
 412         int *cmd = arg;
 413         int ret;
 414
 415         ret = reboot(*cmd);
 416         if (ret == -1 && errno == EINVAL)
 417                 return 1;
 418         return 0;
 419 }
 420
 421 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 422 {
 423         FILE *f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 424         int ret, cmd, v, flags;
 425         long stack_size = 4096;
 426         void *stack = alloca(stack_size);
 427         int status;
 428         pid_t pid;
 429
 430         if (!f) {
 431                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 432                 return 1;
 433         }
 434
 435         ret = fscanf(f, "%d", &v);
 436         fclose(f);
 437         if (ret != 1) {
 438                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 439                 return 1;
 440         }
 441         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 442
 443         flags = CLONE_NEWPID | SIGCHLD;
 444         if (!lxc_list_empty(&conf->id_map))
 445                 flags |= CLONE_NEWUSER;
 446
 447 #ifdef __ia64__
 448         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 449 #else
 450         stack += stack_size;
 451         pid = clone(container_reboot_supported, stack, flags, &cmd);
 452 #endif
 453         if (pid < 0) {
 454                 SYSERROR("failed to clone\n");
 455                 return -1;
 456         }
 457         if (wait(&status) < 0) {
 458                 SYSERROR("unexpected wait error: %m\n");
 459                 return -1;
 460         }
 461
 462         if (WEXITSTATUS(status) != 1)
 463                 return 1;
 464
 465         return 0;
 466 }
 467
 468 static int do_start(void *data)
 469 {
 470         struct lxc_handler *handler = data;
 471
 472         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 473                 SYSERROR("failed to set sigprocmask");
 474                 return -1;
 475         }
 476
 477         /* This prctl must be before the synchro, so if the parent
 478          * dies before we set the parent death signal, we will detect
 479          * its death with the synchro right after, otherwise we have
 480          * a window where the parent can exit before we set the pdeath
 481          * signal leading to a unsupervized container.
 482          */
 483         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 484                 SYSERROR("failed to set pdeath signal");
 485                 return -1;
 486         }
 487
 488         lxc_sync_fini_parent(handler);
 489
 490         /* don't leak the pinfd to the container */
 491         if (handler->pinfd >= 0)
 492                 close(handler->pinfd);
 493
 494         /* Tell the parent task it can begin to configure the
 495          * container and wait for it to finish
 496          */
 497         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 498                 return -1;
 499
 500         /*
 501          * if we are in a new user namespace, become root there to have
 502          * privilege over our namespace
 503          */
 504         if (!lxc_list_empty(&handler->conf->id_map)) {
 505                 NOTICE("switching to gid/uid 0 in new user namespace");
 506                 if (setgid(0)) {
 507                         SYSERROR("setgid");
 508                         goto out_warn_father;
 509                 }
 510                 if (setuid(0)) {
 511                         SYSERROR("setuid");
 512                         goto out_warn_father;
 513                 }
 514         }
 515
 516         #if HAVE_SYS_CAPABILITY_H
 517         if (handler->conf->need_utmp_watch) {
 518                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 519                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 520                         goto out_warn_father;
 521                 }
 522                 DEBUG("Dropped cap_sys_boot\n");
 523         }
 524         #endif
 525
 526         /* Setup the container, ip, names, utsname, ... */
 527         if (lxc_setup(handler->name, handler->conf, handler->lxcpath)) {
 528                 ERROR("failed to setup the container");
 529                 goto out_warn_father;
 530         }
 531
 532         /* ask father to setup cgroups and wait for him to finish */
 533         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 534                 return -1;
 535
 536         if (apparmor_load(handler) < 0)
 537                 goto out_warn_father;
 538
 539         if (lxc_seccomp_load(handler->conf) != 0)
 540                 goto out_warn_father;
 541
 542         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 543                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 544                 goto out_warn_father;
 545         }
 546
 547         /* The clearenv() and putenv() calls have been moved here
 548          * to allow us to use enviroment variables passed to the various
 549          * hooks, such as the start hook above.  Not all of the
 550          * variables like CONFIG_PATH or ROOTFS are valid in this
 551          * context but others are. */
 552         if (clearenv()) {
 553                 SYSERROR("failed to clear environment");
 554                 /* don't error out though */
 555         }
 556
 557         if (putenv("container=lxc")) {
 558                 SYSERROR("failed to set environment variable");
 559                 goto out_warn_father;
 560         }
 561
 562         close(handler->sigfd);
 563
 564         /* after this call, we are in error because this
 565          * ops should not return as it execs */
 566         handler->ops->start(handler, handler->data);
 567
 568 out_warn_father:
 569         /* we want the parent to know something went wrong, so any
 570          * value other than what it expects is ok. */
 571         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 572         return -1;
 573 }
 574
 575 int save_phys_nics(struct lxc_conf *conf)
 576 {
 577         struct lxc_list *iterator;
 578
 579         lxc_list_for_each(iterator, &conf->network) {
 580                 struct lxc_netdev *netdev = iterator->elem;
 581
 582                 if (netdev->type != LXC_NET_PHYS)
 583                         continue;
 584                 conf->saved_nics = realloc(conf->saved_nics,
 585                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 586                 if (!conf->saved_nics) {
 587                         SYSERROR("failed to allocate memory");
 588                         return -1;
 589                 }
 590                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 591                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 592                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 593                         SYSERROR("failed to allocate memory");
 594                         return -1;
 595                 }
 596                 INFO("stored saved_nic #%d idx %d name %s\n", conf->num_savednics,
 597                         conf->saved_nics[conf->num_savednics].ifindex,
 598                         conf->saved_nics[conf->num_savednics].orig_name);
 599                 conf->num_savednics++;
 600         }
 601
 602         return 0;
 603 }
 604
 605 extern bool is_in_subcgroup(int pid, const char *subsystem, struct cgroup_desc *d);
 606 int lxc_spawn(struct lxc_handler *handler)
 607 {
 608         int failed_before_rename = 0;
 609         const char *name = handler->name;
 610
 611         if (lxc_sync_init(handler))
 612                 return -1;
 613
 614         handler->clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 615         if (!lxc_list_empty(&handler->conf->id_map)) {
 616                 INFO("Cloning a new user namespace");
 617                 handler->clone_flags |= CLONE_NEWUSER;
 618         }
 619         if (!lxc_list_empty(&handler->conf->network)) {
 620
 621                 handler->clone_flags |= CLONE_NEWNET;
 622
 623                 /* Find gateway addresses from the link device, which is
 624                  * no longer accessible inside the container. Do this
 625                  * before creating network interfaces, since goto
 626                  * out_delete_net does not work before lxc_clone. */
 627                 if (lxc_find_gateway_addresses(handler)) {
 628                         ERROR("failed to find gateway addresses");
 629                         lxc_sync_fini(handler);
 630                         return -1;
 631                 }
 632
 633                 /* that should be done before the clone because we will
 634                  * fill the netdev index and use them in the child
 635                  */
 636                 if (lxc_create_network(handler)) {
 637                         ERROR("failed to create the network");
 638                         lxc_sync_fini(handler);
 639                         return -1;
 640                 }
 641         }
 642
 643         if (save_phys_nics(handler->conf)) {
 644                 ERROR("failed to save physical nic info");
 645                 goto out_abort;
 646         }
 647
 648         /*
 649          * if the rootfs is not a blockdev, prevent the container from
 650          * marking it readonly.
 651          */
 652
 653         handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 654         if (handler->pinfd == -1) {
 655                 ERROR("failed to pin the container's rootfs");
 656                 goto out_delete_net;
 657         }
 658
 659         /* Create a process in a new set of namespaces */
 660         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 661         if (handler->pid < 0) {
 662                 SYSERROR("failed to fork into a new namespace");
 663                 goto out_delete_net;
 664         }
 665
 666         lxc_sync_fini_child(handler);
 667
 668         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 669                 failed_before_rename = 1;
 670
 671         if ((handler->cgroup = lxc_cgroup_path_create(name)) == NULL)
 672                 goto out_delete_net;
 673
 674         if (setup_cgroup(handler, &handler->conf->cgroup)) {
 675                 ERROR("failed to setup the cgroups for '%s'", name);
 676                 goto out_delete_net;
 677         }
 678
 679         if (lxc_cgroup_enter(handler->cgroup, handler->pid) < 0)
 680                 goto out_delete_net;
 681
 682         if (failed_before_rename)
 683                 goto out_delete_net;
 684
 685         /* Create the network configuration */
 686         if (handler->clone_flags & CLONE_NEWNET) {
 687                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 688                         ERROR("failed to create the configured network");
 689                         goto out_delete_net;
 690                 }
 691         }
 692
 693         /* map the container uids - the container became an invalid
 694          * userid the moment it was cloned with CLONE_NEWUSER - this
 695          * call doesn't change anything immediately, but allows the
 696          * container to setuid(0) (0 being mapped to something else on
 697          * the host) later to become a valid uid again */
 698         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
 699                 ERROR("failed to set up id mapping");
 700                 goto out_delete_net;
 701         }
 702
 703         /* Tell the child to continue its initialization.  we'll get
 704          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
 705          */
 706         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 707                 goto out_delete_net;
 708
 709         if (setup_cgroup_devices(handler, &handler->conf->cgroup)) {
 710                 ERROR("failed to setup the devices cgroup for '%s'", name);
 711                 goto out_delete_net;
 712         }
 713
 714         /* Tell the child to complete its initialization and wait for
 715          * it to exec or return an error.  (the child will never
 716          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
 717          * sync pipe, causing lxc_sync_barrier_child to return
 718          * success, or return a different value, causing us to error
 719          * out).
 720          */
 721         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
 722                 return -1;
 723
 724         if (detect_shared_rootfs())
 725                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
 726
 727         /* If child is in a fresh user namespace, chown his ptys for
 728          * it */
 729         if (uid_shift_ttys(handler->pid, handler->conf))
 730                 DEBUG("Failed to chown ptys.\n");
 731
 732         if (handler->ops->post_start(handler, handler->data))
 733                 goto out_abort;
 734
 735         if (lxc_set_state(name, handler, RUNNING)) {
 736                 ERROR("failed to set state to %s",
 737                               lxc_state2str(RUNNING));
 738                 goto out_abort;
 739         }
 740
 741         lxc_sync_fini(handler);
 742
 743         if (handler->pinfd >= 0)
 744                 close(handler->pinfd);
 745
 746         return 0;
 747
 748 out_delete_net:
 749         if (handler->clone_flags & CLONE_NEWNET)
 750                 lxc_delete_network(handler);
 751 out_abort:
 752         lxc_abort(name, handler);
 753         lxc_sync_fini(handler);
 754         return -1;
 755 }
 756
 757 int __lxc_start(const char *name, struct lxc_conf *conf,
 758                 struct lxc_operations* ops, void *data, const char *lxcpath)
 759 {
 760         struct lxc_handler *handler;
 761         int err = -1;
 762         int status;
 763
 764         handler = lxc_init(name, conf, lxcpath);
 765         if (!handler) {
 766                 ERROR("failed to initialize the container");
 767                 return -1;
 768         }
 769         handler->ops = ops;
 770         handler->data = data;
 771
 772         if (must_drop_cap_sys_boot(handler->conf)) {
 773                 #if HAVE_SYS_CAPABILITY_H
 774                 DEBUG("Dropping cap_sys_boot\n");
 775                 #else
 776                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported\n");
 777                 #endif
 778         } else {
 779                 DEBUG("Not dropping cap_sys_boot or watching utmp\n");
 780                 handler->conf->need_utmp_watch = 0;
 781         }
 782
 783         err = lxc_spawn(handler);
 784         if (err) {
 785                 ERROR("failed to spawn '%s'", name);
 786                 goto out_fini_nonet;
 787         }
 788
 789         err = lxc_poll(name, handler);
 790         if (err) {
 791                 ERROR("mainloop exited with an error");
 792                 goto out_abort;
 793         }
 794
 795         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 796                 continue;
 797
 798         /*
 799          * If the child process exited but was not signaled,
 800          * it didn't call reboot.  This should mean it was an
 801          * lxc-execute which simply exited.  In any case, treat
 802          * it as a 'halt'
 803          */
 804         if (WIFSIGNALED(status)) {
 805                 switch(WTERMSIG(status)) {
 806                 case SIGINT: /* halt */
 807                         DEBUG("Container halting");
 808                         break;
 809                 case SIGHUP: /* reboot */
 810                         DEBUG("Container rebooting");
 811                         handler->conf->reboot = 1;
 812                         break;
 813                 default:
 814                         DEBUG("unknown exit status for init: %d\n", WTERMSIG(status));
 815                         break;
 816                 }
 817         }
 818
 819         lxc_rename_phys_nics_on_shutdown(handler->conf);
 820
 821         err =  lxc_error_set_and_log(handler->pid, status);
 822 out_fini:
 823         lxc_delete_network(handler);
 824
 825 out_fini_nonet:
 826         lxc_fini(name, handler);
 827         return err;
 828
 829 out_abort:
 830         lxc_abort(name, handler);
 831         goto out_fini;
 832 }
 833
 834 struct start_args {
 835         char *const *argv;
 836 };
 837
 838 static int start(struct lxc_handler *handler, void* data)
 839 {
 840         struct start_args *arg = data;
 841
 842         NOTICE("exec'ing '%s'", arg->argv[0]);
 843
 844         execvp(arg->argv[0], arg->argv);
 845         SYSERROR("failed to exec %s", arg->argv[0]);
 846         return 0;
 847 }
 848
 849 static int post_start(struct lxc_handler *handler, void* data)
 850 {
 851         struct start_args *arg = data;
 852
 853         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
 854         return 0;
 855 }
 856
 857 static struct lxc_operations start_ops = {
 858         .start = start,
 859         .post_start = post_start
 860 };
 861
 862 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
 863               const char *lxcpath)
 864 {
 865         struct start_args start_arg = {
 866                 .argv = argv,
 867         };
 868
 869         if (lxc_check_inherited(conf, -1))
 870                 return -1;
 871
 872         conf->need_utmp_watch = 1;
 873         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath);
 874 }