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 #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 #include "lxclock.h"
  75
  76 lxc_log_define(lxc_start, lxc);
  77
  78 static int match_fd(int fd)
  79 {
  80         return (fd == 0 || fd == 1 || fd == 2);
  81 }
  82
  83 int lxc_check_inherited(struct lxc_conf *conf, int fd_to_ignore)
  84 {
  85         struct dirent dirent, *direntp;
  86         int fd, fddir;
  87         DIR *dir;
  88
  89 restart:
  90         process_lock();
  91         dir = opendir("/proc/self/fd");
  92         process_unlock();
  93         if (!dir) {
  94                 WARN("failed to open directory: %m");
  95                 return -1;
  96         }
  97
  98         fddir = dirfd(dir);
  99
 100         while (!readdir_r(dir, &dirent, &direntp)) {
 101                 if (!direntp)
 102                         break;
 103
 104                 if (!strcmp(direntp->d_name, "."))
 105                         continue;
 106
 107                 if (!strcmp(direntp->d_name, ".."))
 108                         continue;
 109
 110                 fd = atoi(direntp->d_name);
 111
 112                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 113                         continue;
 114
 115                 if (match_fd(fd))
 116                         continue;
 117
 118                 if (conf->close_all_fds) {
 119                         process_lock();
 120                         close(fd);
 121                         closedir(dir);
 122                         process_unlock();
 123                         INFO("closed inherited fd %d", fd);
 124                         goto restart;
 125                 }
 126                 WARN("inherited fd %d", fd);
 127         }
 128
 129         process_lock();
 130         closedir(dir); /* cannot fail */
 131         process_unlock();
 132         return 0;
 133 }
 134
 135 static int setup_signal_fd(sigset_t *oldmask)
 136 {
 137         sigset_t mask;
 138         int fd;
 139
 140         /* Block everything except serious error signals */
 141         if (sigfillset(&mask) ||
 142             sigdelset(&mask, SIGILL) ||
 143             sigdelset(&mask, SIGSEGV) ||
 144             sigdelset(&mask, SIGBUS) ||
 145             sigdelset(&mask, SIGWINCH) ||
 146             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 147                 SYSERROR("failed to set signal mask");
 148                 return -1;
 149         }
 150
 151         fd = signalfd(-1, &mask, 0);
 152         if (fd < 0) {
 153                 SYSERROR("failed to create the signal fd");
 154                 return -1;
 155         }
 156
 157         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 158                 SYSERROR("failed to set sigfd to close-on-exec");
 159                 close(fd);
 160                 return -1;
 161         }
 162
 163         DEBUG("sigchild handler set");
 164
 165         return fd;
 166 }
 167
 168 static int signal_handler(int fd, void *data,
 169                            struct lxc_epoll_descr *descr)
 170 {
 171         struct signalfd_siginfo siginfo;
 172         siginfo_t info;
 173         int ret;
 174         pid_t *pid = data;
 175         bool init_died = false;
 176
 177         ret = read(fd, &siginfo, sizeof(siginfo));
 178         if (ret < 0) {
 179                 ERROR("failed to read signal info");
 180                 return -1;
 181         }
 182
 183         if (ret != sizeof(siginfo)) {
 184                 ERROR("unexpected siginfo size");
 185                 return -1;
 186         }
 187
 188         // check whether init is running
 189         info.si_pid = 0;
 190         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 191         if (ret == 0 && info.si_pid == *pid) {
 192                 init_died = true;
 193         }
 194
 195         if (siginfo.ssi_signo != SIGCHLD) {
 196                 kill(*pid, siginfo.ssi_signo);
 197                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 198                 return init_died ? 1 : 0;
 199         }
 200
 201         if (siginfo.ssi_code == CLD_STOPPED ||
 202             siginfo.ssi_code == CLD_CONTINUED) {
 203                 INFO("container init process was stopped/continued");
 204                 return init_died ? 1 : 0;
 205         }
 206
 207         /* more robustness, protect ourself from a SIGCHLD sent
 208          * by a process different from the container init
 209          */
 210         if (siginfo.ssi_pid != *pid) {
 211                 WARN("invalid pid for SIGCHLD");
 212                 return init_died ? 1 : 0;
 213         }
 214
 215         DEBUG("container init process exited");
 216         return 1;
 217 }
 218
 219 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 220 {
 221         handler->state = state;
 222         lxc_monitor_send_state(name, state, handler->lxcpath);
 223         return 0;
 224 }
 225
 226 int lxc_poll(const char *name, struct lxc_handler *handler)
 227 {
 228         int sigfd = handler->sigfd;
 229         int pid = handler->pid;
 230         struct lxc_epoll_descr descr;
 231
 232         if (lxc_mainloop_open(&descr)) {
 233                 ERROR("failed to create mainloop");
 234                 goto out_sigfd;
 235         }
 236
 237         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 238                 ERROR("failed to add handler for the signal");
 239                 goto out_mainloop_open;
 240         }
 241
 242         if (lxc_console_mainloop_add(&descr, handler)) {
 243                 ERROR("failed to add console handler to mainloop");
 244                 goto out_mainloop_open;
 245         }
 246
 247         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 248                 ERROR("failed to add command handler to mainloop");
 249                 goto out_mainloop_open;
 250         }
 251
 252         if (handler->conf->need_utmp_watch) {
 253                 #if HAVE_SYS_CAPABILITY_H
 254                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 255                         ERROR("failed to add utmp handler to mainloop");
 256                         goto out_mainloop_open;
 257                 }
 258                 #else
 259                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support\n");
 260                 #endif
 261         }
 262
 263         return lxc_mainloop(&descr, -1);
 264
 265 out_mainloop_open:
 266         lxc_mainloop_close(&descr);
 267 out_sigfd:
 268         process_lock();
 269         close(sigfd);
 270         process_unlock();
 271         return -1;
 272 }
 273
 274 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 275 {
 276         struct lxc_handler *handler;
 277
 278         handler = malloc(sizeof(*handler));
 279         if (!handler)
 280                 return NULL;
 281
 282         memset(handler, 0, sizeof(*handler));
 283
 284         handler->conf = conf;
 285         handler->lxcpath = lxcpath;
 286         handler->pinfd = -1;
 287
 288         apparmor_handler_init(handler);
 289         handler->name = strdup(name);
 290         if (!handler->name) {
 291                 ERROR("failed to allocate memory");
 292                 goto out_free;
 293         }
 294
 295         if (lxc_cmd_init(name, handler, lxcpath))
 296                 goto out_free_name;
 297
 298         if (lxc_read_seccomp_config(conf) != 0) {
 299                 ERROR("failed loading seccomp policy");
 300                 goto out_close_maincmd_fd;
 301         }
 302
 303         /* Begin by setting the state to STARTING */
 304         if (lxc_set_state(name, handler, STARTING)) {
 305                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 306                 goto out_close_maincmd_fd;
 307         }
 308
 309         /* Start of environment variable setup for hooks */
 310         if (setenv("LXC_NAME", name, 1)) {
 311                 SYSERROR("failed to set environment variable for container name");
 312         }
 313         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 314                 SYSERROR("failed to set environment variable for config path");
 315         }
 316         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 317                 SYSERROR("failed to set environment variable for rootfs mount");
 318         }
 319         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 320                 SYSERROR("failed to set environment variable for rootfs mount");
 321         }
 322         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 323                 SYSERROR("failed to set environment variable for console path");
 324         }
 325         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 326                 SYSERROR("failed to set environment variable for console log");
 327         }
 328         /* End of environment variable setup for hooks */
 329
 330         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 331                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 332                 goto out_aborting;
 333         }
 334
 335         if (lxc_create_tty(name, conf)) {
 336                 ERROR("failed to create the ttys");
 337                 goto out_aborting;
 338         }
 339
 340         /* the signal fd has to be created before forking otherwise
 341          * if the child process exits before we setup the signal fd,
 342          * the event will be lost and the command will be stuck */
 343         handler->sigfd = setup_signal_fd(&handler->oldmask);
 344         if (handler->sigfd < 0) {
 345                 ERROR("failed to set sigchild fd handler");
 346                 goto out_delete_tty;
 347         }
 348
 349         /* do this after setting up signals since it might unblock SIGWINCH */
 350         if (lxc_console_create(conf)) {
 351                 ERROR("failed to create console");
 352                 goto out_restore_sigmask;
 353         }
 354
 355         INFO("'%s' is initialized", name);
 356         return handler;
 357
 358 out_restore_sigmask:
 359         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 360 out_delete_tty:
 361         lxc_delete_tty(&conf->tty_info);
 362 out_aborting:
 363         lxc_set_state(name, handler, ABORTING);
 364 out_close_maincmd_fd:
 365         process_lock();
 366         close(conf->maincmd_fd);
 367         process_unlock();
 368         conf->maincmd_fd = -1;
 369 out_free_name:
 370         free(handler->name);
 371         handler->name = NULL;
 372 out_free:
 373         free(handler);
 374         return NULL;
 375 }
 376
 377 static void lxc_fini(const char *name, struct lxc_handler *handler)
 378 {
 379         /* The STOPPING state is there for future cleanup code
 380          * which can take awhile
 381          */
 382         lxc_set_state(name, handler, STOPPING);
 383         lxc_set_state(name, handler, STOPPED);
 384
 385         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 386                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 387
 388         /* reset mask set by setup_signal_fd */
 389         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 390                 WARN("failed to restore sigprocmask");
 391
 392         lxc_console_delete(&handler->conf->console);
 393         lxc_delete_tty(&handler->conf->tty_info);
 394         process_lock();
 395         close(handler->conf->maincmd_fd);
 396         process_unlock();
 397         handler->conf->maincmd_fd = -1;
 398         free(handler->name);
 399         if (handler->cgroup) {
 400                 lxc_cgroup_process_info_free_and_remove(handler->cgroup);
 401                 handler->cgroup = NULL;
 402         }
 403         free(handler);
 404 }
 405
 406 void lxc_abort(const char *name, struct lxc_handler *handler)
 407 {
 408         int ret, status;
 409
 410         lxc_set_state(name, handler, ABORTING);
 411         if (handler->pid > 0)
 412                 kill(handler->pid, SIGKILL);
 413         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 414 }
 415
 416 #include <sys/reboot.h>
 417 #include <linux/reboot.h>
 418
 419 /*
 420  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 421  * in a child pid namespace if container reboot support exists.
 422  * Otherwise, it will either succeed or return -EPERM.
 423  */
 424 static int container_reboot_supported(void *arg)
 425 {
 426         int *cmd = arg;
 427         int ret;
 428
 429         ret = reboot(*cmd);
 430         if (ret == -1 && errno == EINVAL)
 431                 return 1;
 432         return 0;
 433 }
 434
 435 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 436 {
 437         FILE *f;
 438         int ret, cmd, v, flags;
 439         long stack_size = 4096;
 440         void *stack = alloca(stack_size);
 441         int status;
 442         pid_t pid;
 443
 444         process_lock();
 445         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 446         process_unlock();
 447         if (!f) {
 448                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 449                 return 1;
 450         }
 451
 452         ret = fscanf(f, "%d", &v);
 453         process_lock();
 454         fclose(f);
 455         process_unlock();
 456         if (ret != 1) {
 457                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 458                 return 1;
 459         }
 460         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 461
 462         flags = CLONE_NEWPID | SIGCHLD;
 463         if (!lxc_list_empty(&conf->id_map))
 464                 flags |= CLONE_NEWUSER;
 465
 466 #ifdef __ia64__
 467         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 468 #else
 469         stack += stack_size;
 470         pid = clone(container_reboot_supported, stack, flags, &cmd);
 471 #endif
 472         if (pid < 0) {
 473                 SYSERROR("failed to clone\n");
 474                 return -1;
 475         }
 476         if (wait(&status) < 0) {
 477                 SYSERROR("unexpected wait error: %m\n");
 478                 return -1;
 479         }
 480
 481         if (WEXITSTATUS(status) != 1)
 482                 return 1;
 483
 484         return 0;
 485 }
 486
 487 static int do_start(void *data)
 488 {
 489         struct lxc_handler *handler = data;
 490
 491         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 492                 SYSERROR("failed to set sigprocmask");
 493                 return -1;
 494         }
 495
 496         /* This prctl must be before the synchro, so if the parent
 497          * dies before we set the parent death signal, we will detect
 498          * its death with the synchro right after, otherwise we have
 499          * a window where the parent can exit before we set the pdeath
 500          * signal leading to a unsupervized container.
 501          */
 502         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 503                 SYSERROR("failed to set pdeath signal");
 504                 return -1;
 505         }
 506
 507         lxc_sync_fini_parent(handler);
 508
 509         /* don't leak the pinfd to the container */
 510         if (handler->pinfd >= 0) {
 511                 process_lock();
 512                 close(handler->pinfd);
 513                 process_unlock();
 514         }
 515
 516         /* Tell the parent task it can begin to configure the
 517          * container and wait for it to finish
 518          */
 519         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 520                 return -1;
 521
 522         /*
 523          * if we are in a new user namespace, become root there to have
 524          * privilege over our namespace
 525          */
 526         if (!lxc_list_empty(&handler->conf->id_map)) {
 527                 NOTICE("switching to gid/uid 0 in new user namespace");
 528                 if (setgid(0)) {
 529                         SYSERROR("setgid");
 530                         goto out_warn_father;
 531                 }
 532                 if (setuid(0)) {
 533                         SYSERROR("setuid");
 534                         goto out_warn_father;
 535                 }
 536         }
 537
 538         #if HAVE_SYS_CAPABILITY_H
 539         if (handler->conf->need_utmp_watch) {
 540                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 541                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 542                         goto out_warn_father;
 543                 }
 544                 DEBUG("Dropped cap_sys_boot\n");
 545         }
 546         #endif
 547
 548         /* Setup the container, ip, names, utsname, ... */
 549         if (lxc_setup(handler->name, handler->conf, handler->lxcpath, handler->cgroup)) {
 550                 ERROR("failed to setup the container");
 551                 goto out_warn_father;
 552         }
 553
 554         /* ask father to setup cgroups and wait for him to finish */
 555         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 556                 return -1;
 557
 558         if (apparmor_load(handler) < 0)
 559                 goto out_warn_father;
 560
 561         if (lxc_seccomp_load(handler->conf) != 0)
 562                 goto out_warn_father;
 563
 564         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 565                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 566                 goto out_warn_father;
 567         }
 568
 569         /* The clearenv() and putenv() calls have been moved here
 570          * to allow us to use enviroment variables passed to the various
 571          * hooks, such as the start hook above.  Not all of the
 572          * variables like CONFIG_PATH or ROOTFS are valid in this
 573          * context but others are. */
 574         if (clearenv()) {
 575                 SYSERROR("failed to clear environment");
 576                 /* don't error out though */
 577         }
 578
 579         if (putenv("container=lxc")) {
 580                 SYSERROR("failed to set environment variable");
 581                 goto out_warn_father;
 582         }
 583
 584         process_lock();
 585         close(handler->sigfd);
 586         process_unlock();
 587
 588         /* after this call, we are in error because this
 589          * ops should not return as it execs */
 590         handler->ops->start(handler, handler->data);
 591
 592 out_warn_father:
 593         /* we want the parent to know something went wrong, so any
 594          * value other than what it expects is ok. */
 595         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 596         return -1;
 597 }
 598
 599 int save_phys_nics(struct lxc_conf *conf)
 600 {
 601         struct lxc_list *iterator;
 602
 603         lxc_list_for_each(iterator, &conf->network) {
 604                 struct lxc_netdev *netdev = iterator->elem;
 605
 606                 if (netdev->type != LXC_NET_PHYS)
 607                         continue;
 608                 conf->saved_nics = realloc(conf->saved_nics,
 609                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 610                 if (!conf->saved_nics) {
 611                         SYSERROR("failed to allocate memory");
 612                         return -1;
 613                 }
 614                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 615                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 616                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 617                         SYSERROR("failed to allocate memory");
 618                         return -1;
 619                 }
 620                 INFO("stored saved_nic #%d idx %d name %s\n", conf->num_savednics,
 621                         conf->saved_nics[conf->num_savednics].ifindex,
 622                         conf->saved_nics[conf->num_savednics].orig_name);
 623                 conf->num_savednics++;
 624         }
 625
 626         return 0;
 627 }
 628
 629 int lxc_spawn(struct lxc_handler *handler)
 630 {
 631         int failed_before_rename = 0;
 632         const char *name = handler->name;
 633         struct cgroup_meta_data *cgroup_meta = NULL;
 634         const char *cgroup_pattern = NULL;
 635
 636         if (lxc_sync_init(handler))
 637                 return -1;
 638
 639         handler->clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 640         if (!lxc_list_empty(&handler->conf->id_map)) {
 641                 INFO("Cloning a new user namespace");
 642                 handler->clone_flags |= CLONE_NEWUSER;
 643         }
 644         if (!lxc_list_empty(&handler->conf->network)) {
 645
 646                 handler->clone_flags |= CLONE_NEWNET;
 647
 648                 /* Find gateway addresses from the link device, which is
 649                  * no longer accessible inside the container. Do this
 650                  * before creating network interfaces, since goto
 651                  * out_delete_net does not work before lxc_clone. */
 652                 if (lxc_find_gateway_addresses(handler)) {
 653                         ERROR("failed to find gateway addresses");
 654                         lxc_sync_fini(handler);
 655                         return -1;
 656                 }
 657
 658                 /* that should be done before the clone because we will
 659                  * fill the netdev index and use them in the child
 660                  */
 661                 if (lxc_create_network(handler)) {
 662                         ERROR("failed to create the network");
 663                         lxc_sync_fini(handler);
 664                         return -1;
 665                 }
 666         }
 667
 668         if (save_phys_nics(handler->conf)) {
 669                 ERROR("failed to save physical nic info");
 670                 goto out_abort;
 671         }
 672
 673         cgroup_meta = lxc_cgroup_load_meta();
 674         if (!cgroup_meta) {
 675                 ERROR("failed to detect cgroup metadata");
 676                 goto out_delete_net;
 677         }
 678
 679         /* if we are running as root, use system cgroup pattern, otherwise
 680          * just create a cgroup under the current one. But also fall back to
 681          * that if for some reason reading the configuration fails and no
 682          * default value is available
 683          */
 684         if (getuid() == 0)
 685                 cgroup_pattern = lxc_global_config_value("cgroup.pattern");
 686         if (!cgroup_pattern)
 687                 cgroup_pattern = "%n";
 688
 689         /* Create cgroup before doing clone(), so the child will know from
 690          * handler which cgroup it is going to be put in later.
 691          */
 692         if ((handler->cgroup = lxc_cgroup_create(name, cgroup_pattern, cgroup_meta, NULL)) == NULL) {
 693                 ERROR("failed to create cgroups for '%s'", name);
 694                 goto out_delete_net;
 695         }
 696
 697         /*
 698          * if the rootfs is not a blockdev, prevent the container from
 699          * marking it readonly.
 700          */
 701
 702         handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 703         if (handler->pinfd == -1)
 704                 INFO("failed to pin the container's rootfs");
 705
 706         /* Create a process in a new set of namespaces */
 707         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 708         if (handler->pid < 0) {
 709                 SYSERROR("failed to fork into a new namespace");
 710                 goto out_delete_net;
 711         }
 712
 713         lxc_sync_fini_child(handler);
 714
 715         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 716                 failed_before_rename = 1;
 717
 718         /* In case there is still legacy ns cgroup support in the kernel.
 719          * Should be removed at some later point in time.
 720          */
 721         if (lxc_cgroup_create_legacy(handler->cgroup, name, handler->pid) < 0) {
 722                 ERROR("failed to create legacy ns cgroups for '%s'", name);
 723                 goto out_delete_net;
 724         }
 725
 726         if (lxc_setup_cgroup_without_devices(handler, &handler->conf->cgroup)) {
 727                 ERROR("failed to setup the cgroups for '%s'", name);
 728                 goto out_delete_net;
 729         }
 730
 731         if (lxc_cgroup_enter(handler->cgroup, handler->pid, false) < 0)
 732                 goto out_delete_net;
 733
 734         if (failed_before_rename)
 735                 goto out_delete_net;
 736
 737         /* Create the network configuration */
 738         if (handler->clone_flags & CLONE_NEWNET) {
 739                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 740                         ERROR("failed to create the configured network");
 741                         goto out_delete_net;
 742                 }
 743         }
 744
 745         /* map the container uids - the container became an invalid
 746          * userid the moment it was cloned with CLONE_NEWUSER - this
 747          * call doesn't change anything immediately, but allows the
 748          * container to setuid(0) (0 being mapped to something else on
 749          * the host) later to become a valid uid again */
 750         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
 751                 ERROR("failed to set up id mapping");
 752                 goto out_delete_net;
 753         }
 754
 755         /* Tell the child to continue its initialization.  we'll get
 756          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
 757          */
 758         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 759                 goto out_delete_net;
 760
 761         if (lxc_setup_cgroup_devices(handler, &handler->conf->cgroup)) {
 762                 ERROR("failed to setup the devices cgroup for '%s'", name);
 763                 goto out_delete_net;
 764         }
 765
 766         /* Tell the child to complete its initialization and wait for
 767          * it to exec or return an error.  (the child will never
 768          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
 769          * sync pipe, causing lxc_sync_barrier_child to return
 770          * success, or return a different value, causing us to error
 771          * out).
 772          */
 773         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
 774                 return -1;
 775
 776         if (detect_shared_rootfs())
 777                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
 778
 779         /* If child is in a fresh user namespace, chown his ptys for
 780          * it */
 781         if (uid_shift_ttys(handler->pid, handler->conf))
 782                 DEBUG("Failed to chown ptys.\n");
 783
 784         if (handler->ops->post_start(handler, handler->data))
 785                 goto out_abort;
 786
 787         if (lxc_set_state(name, handler, RUNNING)) {
 788                 ERROR("failed to set state to %s",
 789                               lxc_state2str(RUNNING));
 790                 goto out_abort;
 791         }
 792
 793         lxc_cgroup_put_meta(cgroup_meta);
 794         lxc_sync_fini(handler);
 795
 796         return 0;
 797
 798 out_delete_net:
 799         if (handler->clone_flags & CLONE_NEWNET)
 800                 lxc_delete_network(handler);
 801 out_abort:
 802         lxc_cgroup_put_meta(cgroup_meta);
 803         lxc_abort(name, handler);
 804         lxc_sync_fini(handler);
 805         if (handler->pinfd >= 0) {
 806                 process_lock();
 807                 close(handler->pinfd);
 808                 process_unlock();
 809                 handler->pinfd = -1;
 810         }
 811
 812         return -1;
 813 }
 814
 815 int __lxc_start(const char *name, struct lxc_conf *conf,
 816                 struct lxc_operations* ops, void *data, const char *lxcpath)
 817 {
 818         struct lxc_handler *handler;
 819         int err = -1;
 820         int status;
 821
 822         handler = lxc_init(name, conf, lxcpath);
 823         if (!handler) {
 824                 ERROR("failed to initialize the container");
 825                 return -1;
 826         }
 827         handler->ops = ops;
 828         handler->data = data;
 829
 830         if (must_drop_cap_sys_boot(handler->conf)) {
 831                 #if HAVE_SYS_CAPABILITY_H
 832                 DEBUG("Dropping cap_sys_boot\n");
 833                 #else
 834                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported\n");
 835                 #endif
 836         } else {
 837                 DEBUG("Not dropping cap_sys_boot or watching utmp\n");
 838                 handler->conf->need_utmp_watch = 0;
 839         }
 840
 841         err = lxc_spawn(handler);
 842         if (err) {
 843                 ERROR("failed to spawn '%s'", name);
 844                 goto out_fini_nonet;
 845         }
 846
 847         err = lxc_poll(name, handler);
 848         if (err) {
 849                 ERROR("mainloop exited with an error");
 850                 goto out_abort;
 851         }
 852
 853         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 854                 continue;
 855
 856         /*
 857          * If the child process exited but was not signaled,
 858          * it didn't call reboot.  This should mean it was an
 859          * lxc-execute which simply exited.  In any case, treat
 860          * it as a 'halt'
 861          */
 862         if (WIFSIGNALED(status)) {
 863                 switch(WTERMSIG(status)) {
 864                 case SIGINT: /* halt */
 865                         DEBUG("Container halting");
 866                         break;
 867                 case SIGHUP: /* reboot */
 868                         DEBUG("Container rebooting");
 869                         handler->conf->reboot = 1;
 870                         break;
 871                 default:
 872                         DEBUG("unknown exit status for init: %d\n", WTERMSIG(status));
 873                         break;
 874                 }
 875         }
 876
 877         lxc_rename_phys_nics_on_shutdown(handler->conf);
 878
 879         if (handler->pinfd >= 0) {
 880                 process_lock();
 881                 close(handler->pinfd);
 882                 process_unlock();
 883                 handler->pinfd = -1;
 884         }
 885
 886         err =  lxc_error_set_and_log(handler->pid, status);
 887 out_fini:
 888         lxc_delete_network(handler);
 889
 890 out_fini_nonet:
 891         lxc_fini(name, handler);
 892         return err;
 893
 894 out_abort:
 895         lxc_abort(name, handler);
 896         goto out_fini;
 897 }
 898
 899 struct start_args {
 900         char *const *argv;
 901 };
 902
 903 static int start(struct lxc_handler *handler, void* data)
 904 {
 905         struct start_args *arg = data;
 906
 907         NOTICE("exec'ing '%s'", arg->argv[0]);
 908
 909         execvp(arg->argv[0], arg->argv);
 910         SYSERROR("failed to exec %s", arg->argv[0]);
 911         return 0;
 912 }
 913
 914 static int post_start(struct lxc_handler *handler, void* data)
 915 {
 916         struct start_args *arg = data;
 917
 918         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
 919         return 0;
 920 }
 921
 922 static struct lxc_operations start_ops = {
 923         .start = start,
 924         .post_start = post_start
 925 };
 926
 927 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
 928               const char *lxcpath)
 929 {
 930         struct start_args start_arg = {
 931                 .argv = argv,
 932         };
 933
 934         if (lxc_check_inherited(conf, -1))
 935                 return -1;
 936
 937         conf->need_utmp_watch = 1;
 938         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath);
 939 }