src/lxc/start.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <dlezcano at fr.ibm.com>
   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 #ifdef HAVE_SYS_SIGNALFD_H
  54 #  include <sys/signalfd.h>
  55 #else
  56 /* assume kernel headers are too old */
  57 #include <stdint.h>
  58 struct signalfd_siginfo
  59 {
  60         uint32_t ssi_signo;
  61         int32_t ssi_errno;
  62         int32_t ssi_code;
  63         uint32_t ssi_pid;
  64         uint32_t ssi_uid;
  65         int32_t ssi_fd;
  66         uint32_t ssi_tid;
  67         uint32_t ssi_band;
  68         uint32_t ssi_overrun;
  69         uint32_t ssi_trapno;
  70         int32_t ssi_status;
  71         int32_t ssi_int;
  72         uint64_t ssi_ptr;
  73         uint64_t ssi_utime;
  74         uint64_t ssi_stime;
  75         uint64_t ssi_addr;
  76         uint8_t __pad[48];
  77 };
  78
  79 #  ifndef __NR_signalfd4
  80 /* assume kernel headers are too old */
  81 #    if __i386__
  82 #      define __NR_signalfd4 327
  83 #    elif __x86_64__
  84 #      define __NR_signalfd4 289
  85 #    elif __powerpc__
  86 #      define __NR_signalfd4 313
  87 #    elif __s390x__
  88 #      define __NR_signalfd4 322
  89 #    endif
  90 #endif
  91
  92 #  ifndef __NR_signalfd
  93 /* assume kernel headers are too old */
  94 #    if __i386__
  95 #      define __NR_signalfd 321
  96 #    elif __x86_64__
  97 #      define __NR_signalfd 282
  98 #    elif __powerpc__
  99 #      define __NR_signalfd 305
 100 #    elif __s390x__
 101 #      define __NR_signalfd 316
 102 #    endif
 103 #endif
 104
 105 int signalfd(int fd, const sigset_t *mask, int flags)
 106 {
 107         int retval;
 108
 109         retval = syscall (__NR_signalfd4, fd, mask, _NSIG / 8, flags);
 110         if (errno == ENOSYS && flags == 0)
 111                 retval = syscall (__NR_signalfd, fd, mask, _NSIG / 8);
 112         return retval;
 113 }
 114 #endif
 115
 116 #if !HAVE_DECL_PR_CAPBSET_DROP
 117 #define PR_CAPBSET_DROP 24
 118 #endif
 119
 120 #include "start.h"
 121 #include "conf.h"
 122 #include "log.h"
 123 #include "cgroup.h"
 124 #include "error.h"
 125 #include "af_unix.h"
 126 #include "mainloop.h"
 127 #include "utils.h"
 128 #include "lxcutmp.h"
 129 #include "monitor.h"
 130 #include "commands.h"
 131 #include "console.h"
 132 #include "sync.h"
 133 #include "namespace.h"
 134 #include "apparmor.h"
 135 #include "lxcseccomp.h"
 136 #include "caps.h"
 137
 138 lxc_log_define(lxc_start, lxc);
 139
 140 static int match_fd(int fd)
 141 {
 142         return (fd == 0 || fd == 1 || fd == 2);
 143 }
 144
 145 int lxc_check_inherited(struct lxc_conf *conf, int fd_to_ignore)
 146 {
 147         struct dirent dirent, *direntp;
 148         int fd, fddir;
 149         DIR *dir;
 150
 151 restart:
 152         dir = opendir("/proc/self/fd");
 153         if (!dir) {
 154                 WARN("failed to open directory: %m");
 155                 return -1;
 156         }
 157
 158         fddir = dirfd(dir);
 159
 160         while (!readdir_r(dir, &dirent, &direntp)) {
 161                 if (!direntp)
 162                         break;
 163
 164                 if (!strcmp(direntp->d_name, "."))
 165                         continue;
 166
 167                 if (!strcmp(direntp->d_name, ".."))
 168                         continue;
 169
 170                 fd = atoi(direntp->d_name);
 171
 172                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 173                         continue;
 174
 175                 if (match_fd(fd))
 176                         continue;
 177
 178                 if (conf->close_all_fds) {
 179                         close(fd);
 180                         closedir(dir);
 181                         INFO("closed inherited fd %d", fd);
 182                         goto restart;
 183                 }
 184                 WARN("inherited fd %d", fd);
 185         }
 186
 187         closedir(dir); /* cannot fail */
 188         return 0;
 189 }
 190
 191 static int setup_signal_fd(sigset_t *oldmask)
 192 {
 193         sigset_t mask;
 194         int fd;
 195
 196         /* Block everything except serious error signals */
 197         if (sigfillset(&mask) ||
 198             sigdelset(&mask, SIGILL) ||
 199             sigdelset(&mask, SIGSEGV) ||
 200             sigdelset(&mask, SIGBUS) ||
 201             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 202                 SYSERROR("failed to set signal mask");
 203                 return -1;
 204         }
 205
 206         fd = signalfd(-1, &mask, 0);
 207         if (fd < 0) {
 208                 SYSERROR("failed to create the signal fd");
 209                 return -1;
 210         }
 211
 212         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 213                 SYSERROR("failed to set sigfd to close-on-exec");
 214                 close(fd);
 215                 return -1;
 216         }
 217
 218         DEBUG("sigchild handler set");
 219
 220         return fd;
 221 }
 222
 223 static int signal_handler(int fd, void *data,
 224                            struct lxc_epoll_descr *descr)
 225 {
 226         struct signalfd_siginfo siginfo;
 227         int ret;
 228         pid_t *pid = data;
 229
 230         ret = read(fd, &siginfo, sizeof(siginfo));
 231         if (ret < 0) {
 232                 ERROR("failed to read signal info");
 233                 return -1;
 234         }
 235
 236         if (ret != sizeof(siginfo)) {
 237                 ERROR("unexpected siginfo size");
 238                 return -1;
 239         }
 240
 241         if (siginfo.ssi_signo != SIGCHLD) {
 242                 kill(*pid, siginfo.ssi_signo);
 243                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 244                 return 0;
 245         }
 246
 247         if (siginfo.ssi_code == CLD_STOPPED ||
 248             siginfo.ssi_code == CLD_CONTINUED) {
 249                 INFO("container init process was stopped/continued");
 250                 return 0;
 251         }
 252
 253         /* more robustness, protect ourself from a SIGCHLD sent
 254          * by a process different from the container init
 255          */
 256         if (siginfo.ssi_pid != *pid) {
 257                 WARN("invalid pid for SIGCHLD");
 258                 return 0;
 259         }
 260
 261         DEBUG("container init process exited");
 262         return 1;
 263 }
 264
 265 int lxc_pid_callback(int fd, struct lxc_request *request,
 266                      struct lxc_handler *handler)
 267 {
 268         struct lxc_answer answer;
 269         int ret;
 270
 271         answer.pid = handler->pid;
 272         answer.ret = 0;
 273
 274         ret = send(fd, &answer, sizeof(answer), 0);
 275         if (ret < 0) {
 276                 WARN("failed to send answer to the peer");
 277                 return -1;
 278         }
 279
 280         if (ret != sizeof(answer)) {
 281                 ERROR("partial answer sent");
 282                 return -1;
 283         }
 284
 285         return 0;
 286 }
 287
 288 int lxc_clone_flags_callback(int fd, struct lxc_request *request,
 289                              struct lxc_handler *handler)
 290 {
 291         struct lxc_answer answer;
 292         int ret;
 293
 294         answer.pid = 0;
 295         answer.ret = handler->clone_flags;
 296
 297         ret = send(fd, &answer, sizeof(answer), 0);
 298         if (ret < 0) {
 299                 WARN("failed to send answer to the peer");
 300                 return -1;
 301         }
 302
 303         if (ret != sizeof(answer)) {
 304                 ERROR("partial answer sent");
 305                 return -1;
 306         }
 307
 308         return 0;
 309 }
 310
 311 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 312 {
 313         handler->state = state;
 314         lxc_monitor_send_state(name, state);
 315         return 0;
 316 }
 317
 318 int lxc_poll(const char *name, struct lxc_handler *handler)
 319 {
 320         int sigfd = handler->sigfd;
 321         int pid = handler->pid;
 322         struct lxc_epoll_descr descr;
 323
 324         if (lxc_mainloop_open(&descr)) {
 325                 ERROR("failed to create mainloop");
 326                 goto out_sigfd;
 327         }
 328
 329         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 330                 ERROR("failed to add handler for the signal");
 331                 goto out_mainloop_open;
 332         }
 333
 334         if (lxc_console_mainloop_add(&descr, handler)) {
 335                 ERROR("failed to add console handler to mainloop");
 336                 goto out_mainloop_open;
 337         }
 338
 339         if (lxc_command_mainloop_add(name, &descr, handler)) {
 340                 ERROR("failed to add command handler to mainloop");
 341                 goto out_mainloop_open;
 342         }
 343
 344         if (handler->conf->need_utmp_watch) {
 345                 #if HAVE_SYS_CAPABILITY_H
 346                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 347                         ERROR("failed to add utmp handler to mainloop");
 348                         goto out_mainloop_open;
 349                 }
 350                 #else
 351                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support\n");
 352                 #endif
 353         }
 354
 355         return lxc_mainloop(&descr);
 356
 357 out_mainloop_open:
 358         lxc_mainloop_close(&descr);
 359 out_sigfd:
 360         close(sigfd);
 361         return -1;
 362 }
 363
 364 extern int lxc_caps_check(void);
 365
 366 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf)
 367 {
 368         struct lxc_handler *handler;
 369
 370         if (!lxc_caps_check()) {
 371                 ERROR("Not running with sufficient privilege");
 372                 return NULL;
 373         }
 374
 375         handler = malloc(sizeof(*handler));
 376         if (!handler)
 377                 return NULL;
 378
 379         memset(handler, 0, sizeof(*handler));
 380
 381         handler->conf = conf;
 382
 383         apparmor_handler_init(handler);
 384         handler->name = strdup(name);
 385         if (!handler->name) {
 386                 ERROR("failed to allocate memory");
 387                 goto out_free;
 388         }
 389
 390         if (lxc_command_init(name, handler))
 391                 goto out_free_name;
 392
 393         if (lxc_read_seccomp_config(conf) != 0) {
 394                 ERROR("failed loading seccomp policy");
 395                 goto out_close_maincmd_fd;
 396         }
 397
 398         /* Begin the set the state to STARTING*/
 399         if (lxc_set_state(name, handler, STARTING)) {
 400                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 401                 goto out_free_name;
 402         }
 403
 404         /* Start of environment variable setup for hooks */
 405         if (setenv("LXC_NAME", name, 1)) {
 406                 SYSERROR("failed to set environment variable for container name");
 407         }
 408         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 409                 SYSERROR("failed to set environment variable for config path");
 410         }
 411         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 412                 SYSERROR("failed to set environment variable for rootfs mount");
 413         }
 414         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 415                 SYSERROR("failed to set environment variable for rootfs mount");
 416         }
 417         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 418                 SYSERROR("failed to set environment variable for console path");
 419         }
 420         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 421                 SYSERROR("failed to set environment variable for console log");
 422         }
 423         /* End of environment variable setup for hooks */
 424
 425         if (run_lxc_hooks(name, "pre-start", conf)) {
 426                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 427                 goto out_aborting;
 428         }
 429
 430         if (lxc_create_tty(name, conf)) {
 431                 ERROR("failed to create the ttys");
 432                 goto out_aborting;
 433         }
 434
 435         if (lxc_create_console(conf)) {
 436                 ERROR("failed to create console");
 437                 goto out_delete_tty;
 438         }
 439
 440         /* the signal fd has to be created before forking otherwise
 441          * if the child process exits before we setup the signal fd,
 442          * the event will be lost and the command will be stuck */
 443         handler->sigfd = setup_signal_fd(&handler->oldmask);
 444         if (handler->sigfd < 0) {
 445                 ERROR("failed to set sigchild fd handler");
 446                 goto out_delete_console;
 447         }
 448
 449         INFO("'%s' is initialized", name);
 450         return handler;
 451
 452 out_delete_console:
 453         lxc_delete_console(&conf->console);
 454 out_delete_tty:
 455         lxc_delete_tty(&conf->tty_info);
 456 out_aborting:
 457         lxc_set_state(name, handler, ABORTING);
 458 out_close_maincmd_fd:
 459         close(conf->maincmd_fd);
 460         conf->maincmd_fd = -1;
 461 out_free_name:
 462         free(handler->name);
 463         handler->name = NULL;
 464 out_free:
 465         free(handler);
 466         return NULL;
 467 }
 468
 469 void lxc_fini(const char *name, struct lxc_handler *handler)
 470 {
 471         /* The STOPPING state is there for future cleanup code
 472          * which can take awhile
 473          */
 474         lxc_set_state(name, handler, STOPPING);
 475         lxc_set_state(name, handler, STOPPED);
 476
 477         if (run_lxc_hooks(name, "post-stop", handler->conf))
 478                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 479
 480         /* reset mask set by setup_signal_fd */
 481         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 482                 WARN("failed to restore sigprocmask");
 483
 484         lxc_delete_console(&handler->conf->console);
 485         lxc_delete_tty(&handler->conf->tty_info);
 486         close(handler->conf->maincmd_fd);
 487         handler->conf->maincmd_fd = -1;
 488         free(handler->name);
 489         free(handler);
 490 }
 491
 492 void lxc_abort(const char *name, struct lxc_handler *handler)
 493 {
 494         lxc_set_state(name, handler, ABORTING);
 495         if (handler->pid > 0)
 496                 kill(handler->pid, SIGKILL);
 497 }
 498
 499 #include <sys/reboot.h>
 500 #include <linux/reboot.h>
 501
 502 /*
 503  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 504  * in a child pid namespace if container reboot support exists.
 505  * Otherwise, it will either succeed or return -EPERM.
 506  */
 507 static int container_reboot_supported(void *arg)
 508 {
 509         int *cmd = arg;
 510         int ret;
 511
 512         ret = reboot(*cmd);
 513         if (ret == -1 && errno == EINVAL)
 514                 return 1;
 515         return 0;
 516 }
 517
 518 static int must_drop_cap_sys_boot(void)
 519 {
 520         FILE *f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 521         int ret, cmd, v;
 522         long stack_size = 4096;
 523         void *stack = alloca(stack_size) + stack_size;
 524         int status;
 525         pid_t pid;
 526
 527         if (!f) {
 528                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 529                 return 1;
 530         }
 531
 532         ret = fscanf(f, "%d", &v);
 533         fclose(f);
 534         if (ret != 1) {
 535                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 536                 return 1;
 537         }
 538         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 539
 540         pid = clone(container_reboot_supported, stack, CLONE_NEWPID | SIGCHLD, &cmd);
 541         if (pid < 0) {
 542                 SYSERROR("failed to clone\n");
 543                 return -1;
 544         }
 545         if (wait(&status) < 0) {
 546                 SYSERROR("unexpected wait error: %m\n");
 547                 return -1;
 548         }
 549
 550         if (WEXITSTATUS(status) != 1)
 551                 return 1;
 552
 553         return 0;
 554 }
 555
 556 static int do_start(void *data)
 557 {
 558         struct lxc_handler *handler = data;
 559
 560         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 561                 SYSERROR("failed to set sigprocmask");
 562                 return -1;
 563         }
 564
 565         /* This prctl must be before the synchro, so if the parent
 566          * dies before we set the parent death signal, we will detect
 567          * its death with the synchro right after, otherwise we have
 568          * a window where the parent can exit before we set the pdeath
 569          * signal leading to a unsupervized container.
 570          */
 571         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 572                 SYSERROR("failed to set pdeath signal");
 573                 return -1;
 574         }
 575
 576         lxc_sync_fini_parent(handler);
 577
 578         /* Tell the parent task it can begin to configure the
 579          * container and wait for it to finish
 580          */
 581         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 582                 return -1;
 583
 584         #if HAVE_SYS_CAPABILITY_H
 585         if (handler->conf->need_utmp_watch) {
 586                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 587                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 588                         goto out_warn_father;
 589                 }
 590                 DEBUG("Dropped cap_sys_boot\n");
 591         }
 592         #endif
 593
 594         /* Setup the container, ip, names, utsname, ... */
 595         if (lxc_setup(handler->name, handler->conf)) {
 596                 ERROR("failed to setup the container");
 597                 goto out_warn_father;
 598         }
 599
 600         /* ask father to setup cgroups and wait for him to finish */
 601         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 602                 return -1;
 603
 604         if (apparmor_load(handler) < 0)
 605                 goto out_warn_father;
 606
 607         if (lxc_seccomp_load(handler->conf) != 0)
 608                 goto out_warn_father;
 609
 610         if (run_lxc_hooks(handler->name, "start", handler->conf)) {
 611                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 612                 goto out_warn_father;
 613         }
 614
 615         /* The clearenv() and putenv() calls have been moved here
 616          * to allow us to use enviroment variables passed to the various
 617          * hooks, such as the start hook above.  Not all of the
 618          * variables like CONFIG_PATH or ROOTFS are valid in this
 619          * context but others are. */
 620         if (clearenv()) {
 621                 SYSERROR("failed to clear environment");
 622                 /* don't error out though */
 623         }
 624
 625         if (putenv("container=lxc")) {
 626                 SYSERROR("failed to set environment variable");
 627                 goto out_warn_father;
 628         }
 629
 630         close(handler->sigfd);
 631
 632         /* after this call, we are in error because this
 633          * ops should not return as it execs */
 634         handler->ops->start(handler, handler->data);
 635
 636 out_warn_father:
 637         /* we want the parent to know something went wrong, so any
 638          * value other than what it expects is ok. */
 639         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 640         return -1;
 641 }
 642
 643 int save_phys_nics(struct lxc_conf *conf)
 644 {
 645         struct lxc_list *iterator;
 646
 647         lxc_list_for_each(iterator, &conf->network) {
 648                 struct lxc_netdev *netdev = iterator->elem;
 649
 650                 if (netdev->type != LXC_NET_PHYS)
 651                         continue;
 652                 conf->saved_nics = realloc(conf->saved_nics,
 653                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 654                 if (!conf->saved_nics) {
 655                         SYSERROR("failed to allocate memory");
 656                         return -1;
 657                 }
 658                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 659                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 660                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 661                         SYSERROR("failed to allocate memory");
 662                         return -1;
 663                 }
 664                 INFO("stored saved_nic #%d idx %d name %s\n", conf->num_savednics,
 665                         conf->saved_nics[conf->num_savednics].ifindex,
 666                         conf->saved_nics[conf->num_savednics].orig_name);
 667                 conf->num_savednics++;
 668         }
 669
 670         return 0;
 671 }
 672
 673
 674 int lxc_spawn(struct lxc_handler *handler)
 675 {
 676         int failed_before_rename = 0;
 677         const char *name = handler->name;
 678         int pinfd;
 679
 680         if (lxc_sync_init(handler))
 681                 return -1;
 682
 683         handler->clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 684         if (!lxc_list_empty(&handler->conf->network)) {
 685
 686                 handler->clone_flags |= CLONE_NEWNET;
 687
 688                 /* Find gateway addresses from the link device, which is
 689                  * no longer accessible inside the container. Do this
 690                  * before creating network interfaces, since goto
 691                  * out_delete_net does not work before lxc_clone. */
 692                 if (lxc_find_gateway_addresses(handler)) {
 693                         ERROR("failed to find gateway addresses");
 694                         lxc_sync_fini(handler);
 695                         return -1;
 696                 }
 697
 698                 /* that should be done before the clone because we will
 699                  * fill the netdev index and use them in the child
 700                  */
 701                 if (lxc_create_network(handler)) {
 702                         ERROR("failed to create the network");
 703                         lxc_sync_fini(handler);
 704                         return -1;
 705                 }
 706         }
 707
 708         if (save_phys_nics(handler->conf)) {
 709                 ERROR("failed to save physical nic info");
 710                 goto out_abort;
 711         }
 712
 713         /*
 714          * if the rootfs is not a blockdev, prevent the container from
 715          * marking it readonly.
 716          */
 717
 718         pinfd = pin_rootfs(handler->conf->rootfs.path);
 719         if (pinfd == -1) {
 720                 ERROR("failed to pin the container's rootfs");
 721                 goto out_abort;
 722         }
 723
 724         /* Create a process in a new set of namespaces */
 725         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 726         if (handler->pid < 0) {
 727                 SYSERROR("failed to fork into a new namespace");
 728                 goto out_delete_net;
 729         }
 730
 731         lxc_sync_fini_child(handler);
 732
 733         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 734                 failed_before_rename = 1;
 735
 736         if (lxc_cgroup_create(name, handler->pid))
 737                 goto out_delete_net;
 738
 739         if (failed_before_rename)
 740                 goto out_delete_net;
 741
 742         /* Create the network configuration */
 743         if (handler->clone_flags & CLONE_NEWNET) {
 744                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 745                         ERROR("failed to create the configured network");
 746                         goto out_delete_net;
 747                 }
 748         }
 749
 750         /* Tell the child to continue its initialization.  we'll get
 751          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
 752          */
 753         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 754                 goto out_delete_net;
 755
 756         if (setup_cgroup(name, &handler->conf->cgroup)) {
 757                 ERROR("failed to setup the cgroups for '%s'", name);
 758                 goto out_delete_net;
 759         }
 760
 761
 762         /* Tell the child to complete its initialization and wait for
 763          * it to exec or return an error.  (the child will never
 764          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
 765          * sync pipe, causing lxc_sync_barrier_child to return
 766          * success, or return a different value, causing us to error
 767          * out).
 768          */
 769         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
 770                 return -1;
 771
 772         if (detect_shared_rootfs())
 773                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
 774
 775         if (handler->ops->post_start(handler, handler->data))
 776                 goto out_abort;
 777
 778         if (lxc_set_state(name, handler, RUNNING)) {
 779                 ERROR("failed to set state to %s",
 780                               lxc_state2str(RUNNING));
 781                 goto out_abort;
 782         }
 783
 784         lxc_sync_fini(handler);
 785
 786         if (pinfd >= 0)
 787                 close(pinfd);
 788
 789         return 0;
 790
 791 out_delete_net:
 792         if (handler->clone_flags & CLONE_NEWNET)
 793                 lxc_delete_network(handler);
 794 out_abort:
 795         lxc_abort(name, handler);
 796         lxc_sync_fini(handler);
 797         return -1;
 798 }
 799
 800 int __lxc_start(const char *name, struct lxc_conf *conf,
 801                 struct lxc_operations* ops, void *data)
 802 {
 803         struct lxc_handler *handler;
 804         int err = -1;
 805         int status;
 806
 807         handler = lxc_init(name, conf);
 808         if (!handler) {
 809                 ERROR("failed to initialize the container");
 810                 return -1;
 811         }
 812         handler->ops = ops;
 813         handler->data = data;
 814
 815         if (must_drop_cap_sys_boot()) {
 816                 #if HAVE_SYS_CAPABILITY_H
 817                 DEBUG("Dropping cap_sys_boot\n");
 818                 #else
 819                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported\n");
 820                 #endif
 821         } else {
 822                 DEBUG("Not dropping cap_sys_boot or watching utmp\n");
 823                 handler->conf->need_utmp_watch = 0;
 824         }
 825
 826         err = lxc_spawn(handler);
 827         if (err) {
 828                 ERROR("failed to spawn '%s'", name);
 829                 goto out_fini_nonet;
 830         }
 831
 832         err = lxc_poll(name, handler);
 833         if (err) {
 834                 ERROR("mainloop exited with an error");
 835                 goto out_abort;
 836         }
 837
 838         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 839                 continue;
 840
 841         /*
 842          * If the child process exited but was not signaled,
 843          * it didn't call reboot.  This should mean it was an
 844          * lxc-execute which simply exited.  In any case, treat
 845          * it as a 'halt'
 846          */
 847         if (WIFSIGNALED(status)) {
 848                 switch(WTERMSIG(status)) {
 849                 case SIGINT: /* halt */
 850                         DEBUG("Container halting");
 851                         break;
 852                 case SIGHUP: /* reboot */
 853                         DEBUG("Container rebooting");
 854                         handler->conf->reboot = 1;
 855                         break;
 856                 default:
 857                         DEBUG("unknown exit status for init: %d\n", WTERMSIG(status));
 858                         break;
 859                 }
 860         }
 861
 862         lxc_rename_phys_nics_on_shutdown(handler->conf);
 863
 864         err =  lxc_error_set_and_log(handler->pid, status);
 865 out_fini:
 866         lxc_delete_network(handler);
 867
 868 out_fini_nonet:
 869         lxc_cgroup_destroy(name);
 870         lxc_fini(name, handler);
 871         return err;
 872
 873 out_abort:
 874         lxc_abort(name, handler);
 875         goto out_fini;
 876 }
 877
 878 struct start_args {
 879         char *const *argv;
 880 };
 881
 882 static int start(struct lxc_handler *handler, void* data)
 883 {
 884         struct start_args *arg = data;
 885
 886         NOTICE("exec'ing '%s'", arg->argv[0]);
 887
 888         execvp(arg->argv[0], arg->argv);
 889         SYSERROR("failed to exec %s", arg->argv[0]);
 890         return 0;
 891 }
 892
 893 static int post_start(struct lxc_handler *handler, void* data)
 894 {
 895         struct start_args *arg = data;
 896
 897         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
 898         return 0;
 899 }
 900
 901 static struct lxc_operations start_ops = {
 902         .start = start,
 903         .post_start = post_start
 904 };
 905
 906 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf)
 907 {
 908         struct start_args start_arg = {
 909                 .argv = argv,
 910         };
 911
 912         if (lxc_check_inherited(conf, -1))
 913                 return -1;
 914
 915         conf->need_utmp_watch = 1;
 916         return __lxc_start(name, conf, &start_ops, &start_arg);
 917 }