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 "lxcseccomp.h"
  72 #include "caps.h"
  73 #include "lxclock.h"
  74 #include "lsm/lsm.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         lsm_init();
 289
 290         handler->name = strdup(name);
 291         if (!handler->name) {
 292                 ERROR("failed to allocate memory");
 293                 goto out_free;
 294         }
 295
 296         if (lxc_cmd_init(name, handler, lxcpath))
 297                 goto out_free_name;
 298
 299         if (lxc_read_seccomp_config(conf) != 0) {
 300                 ERROR("failed loading seccomp policy");
 301                 goto out_close_maincmd_fd;
 302         }
 303
 304         /* Begin by setting the state to STARTING */
 305         if (lxc_set_state(name, handler, STARTING)) {
 306                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 307                 goto out_close_maincmd_fd;
 308         }
 309
 310         /* Start of environment variable setup for hooks */
 311         if (setenv("LXC_NAME", name, 1)) {
 312                 SYSERROR("failed to set environment variable for container name");
 313         }
 314         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 315                 SYSERROR("failed to set environment variable for config path");
 316         }
 317         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 318                 SYSERROR("failed to set environment variable for rootfs mount");
 319         }
 320         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 321                 SYSERROR("failed to set environment variable for rootfs mount");
 322         }
 323         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 324                 SYSERROR("failed to set environment variable for console path");
 325         }
 326         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 327                 SYSERROR("failed to set environment variable for console log");
 328         }
 329         /* End of environment variable setup for hooks */
 330
 331         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 332                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 333                 goto out_aborting;
 334         }
 335
 336         if (lxc_create_tty(name, conf)) {
 337                 ERROR("failed to create the ttys");
 338                 goto out_aborting;
 339         }
 340
 341         /* the signal fd has to be created before forking otherwise
 342          * if the child process exits before we setup the signal fd,
 343          * the event will be lost and the command will be stuck */
 344         handler->sigfd = setup_signal_fd(&handler->oldmask);
 345         if (handler->sigfd < 0) {
 346                 ERROR("failed to set sigchild fd handler");
 347                 goto out_delete_tty;
 348         }
 349
 350         /* do this after setting up signals since it might unblock SIGWINCH */
 351         if (lxc_console_create(conf)) {
 352                 ERROR("failed to create console");
 353                 goto out_restore_sigmask;
 354         }
 355
 356         INFO("'%s' is initialized", name);
 357         return handler;
 358
 359 out_restore_sigmask:
 360         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 361 out_delete_tty:
 362         lxc_delete_tty(&conf->tty_info);
 363 out_aborting:
 364         lxc_set_state(name, handler, ABORTING);
 365 out_close_maincmd_fd:
 366         process_lock();
 367         close(conf->maincmd_fd);
 368         process_unlock();
 369         conf->maincmd_fd = -1;
 370 out_free_name:
 371         free(handler->name);
 372         handler->name = NULL;
 373 out_free:
 374         free(handler);
 375         return NULL;
 376 }
 377
 378 static void lxc_fini(const char *name, struct lxc_handler *handler)
 379 {
 380         /* The STOPPING state is there for future cleanup code
 381          * which can take awhile
 382          */
 383         lxc_set_state(name, handler, STOPPING);
 384         lxc_set_state(name, handler, STOPPED);
 385
 386         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 387                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 388
 389         /* reset mask set by setup_signal_fd */
 390         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 391                 WARN("failed to restore sigprocmask");
 392
 393         lxc_console_delete(&handler->conf->console);
 394         lxc_delete_tty(&handler->conf->tty_info);
 395         process_lock();
 396         close(handler->conf->maincmd_fd);
 397         process_unlock();
 398         handler->conf->maincmd_fd = -1;
 399         free(handler->name);
 400         if (handler->cgroup) {
 401                 lxc_cgroup_process_info_free_and_remove(handler->cgroup);
 402                 handler->cgroup = NULL;
 403         }
 404         free(handler);
 405 }
 406
 407 void lxc_abort(const char *name, struct lxc_handler *handler)
 408 {
 409         int ret, status;
 410
 411         lxc_set_state(name, handler, ABORTING);
 412         if (handler->pid > 0)
 413                 kill(handler->pid, SIGKILL);
 414         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 415 }
 416
 417 #include <sys/reboot.h>
 418 #include <linux/reboot.h>
 419
 420 /*
 421  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 422  * in a child pid namespace if container reboot support exists.
 423  * Otherwise, it will either succeed or return -EPERM.
 424  */
 425 static int container_reboot_supported(void *arg)
 426 {
 427         int *cmd = arg;
 428         int ret;
 429
 430         ret = reboot(*cmd);
 431         if (ret == -1 && errno == EINVAL)
 432                 return 1;
 433         return 0;
 434 }
 435
 436 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 437 {
 438         FILE *f;
 439         int ret, cmd, v, flags;
 440         long stack_size = 4096;
 441         void *stack = alloca(stack_size);
 442         int status;
 443         pid_t pid;
 444
 445         process_lock();
 446         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 447         process_unlock();
 448         if (!f) {
 449                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 450                 return 1;
 451         }
 452
 453         ret = fscanf(f, "%d", &v);
 454         process_lock();
 455         fclose(f);
 456         process_unlock();
 457         if (ret != 1) {
 458                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 459                 return 1;
 460         }
 461         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 462
 463         flags = CLONE_NEWPID | SIGCHLD;
 464         if (!lxc_list_empty(&conf->id_map))
 465                 flags |= CLONE_NEWUSER;
 466
 467 #ifdef __ia64__
 468         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 469 #else
 470         stack += stack_size;
 471         pid = clone(container_reboot_supported, stack, flags, &cmd);
 472 #endif
 473         if (pid < 0) {
 474                 SYSERROR("failed to clone\n");
 475                 return -1;
 476         }
 477         if (wait(&status) < 0) {
 478                 SYSERROR("unexpected wait error: %m\n");
 479                 return -1;
 480         }
 481
 482         if (WEXITSTATUS(status) != 1)
 483                 return 1;
 484
 485         return 0;
 486 }
 487
 488 static int do_start(void *data)
 489 {
 490         struct lxc_handler *handler = data;
 491
 492         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 493                 SYSERROR("failed to set sigprocmask");
 494                 return -1;
 495         }
 496
 497         /* This prctl must be before the synchro, so if the parent
 498          * dies before we set the parent death signal, we will detect
 499          * its death with the synchro right after, otherwise we have
 500          * a window where the parent can exit before we set the pdeath
 501          * signal leading to a unsupervized container.
 502          */
 503         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 504                 SYSERROR("failed to set pdeath signal");
 505                 return -1;
 506         }
 507
 508         lxc_sync_fini_parent(handler);
 509
 510         /* don't leak the pinfd to the container */
 511         if (handler->pinfd >= 0) {
 512                 process_lock();
 513                 close(handler->pinfd);
 514                 process_unlock();
 515         }
 516
 517         /* Tell the parent task it can begin to configure the
 518          * container and wait for it to finish
 519          */
 520         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 521                 return -1;
 522
 523         /*
 524          * if we are in a new user namespace, become root there to have
 525          * privilege over our namespace
 526          */
 527         if (!lxc_list_empty(&handler->conf->id_map)) {
 528                 NOTICE("switching to gid/uid 0 in new user namespace");
 529                 if (setgid(0)) {
 530                         SYSERROR("setgid");
 531                         goto out_warn_father;
 532                 }
 533                 if (setuid(0)) {
 534                         SYSERROR("setuid");
 535                         goto out_warn_father;
 536                 }
 537         }
 538
 539         #if HAVE_SYS_CAPABILITY_H
 540         if (handler->conf->need_utmp_watch) {
 541                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 542                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 543                         goto out_warn_father;
 544                 }
 545                 DEBUG("Dropped cap_sys_boot\n");
 546         }
 547         #endif
 548
 549         /* Setup the container, ip, names, utsname, ... */
 550         if (lxc_setup(handler->name, handler->conf, handler->lxcpath, handler->cgroup)) {
 551                 ERROR("failed to setup the container");
 552                 goto out_warn_father;
 553         }
 554
 555         /* ask father to setup cgroups and wait for him to finish */
 556         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 557                 return -1;
 558
 559         /* XXX: hmm apparmor switches right away since it uses
 560          * aa_change_profile() and not aa_change_onexec(). SELinux on the other
 561          * hand is going to transition on exec(). Is it bad to run the stuff
 562          * between here and exec() in the more privileged context?
 563          */
 564         if (lsm_process_label_set(handler->conf->lsm_aa_profile ?
 565                                   handler->conf->lsm_aa_profile :
 566                                   handler->conf->lsm_se_context, 1) < 0)
 567                 goto out_warn_father;
 568         lsm_proc_unmount(handler->conf);
 569
 570         if (lxc_seccomp_load(handler->conf) != 0)
 571                 goto out_warn_father;
 572
 573         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 574                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 575                 goto out_warn_father;
 576         }
 577
 578         /* The clearenv() and putenv() calls have been moved here
 579          * to allow us to use enviroment variables passed to the various
 580          * hooks, such as the start hook above.  Not all of the
 581          * variables like CONFIG_PATH or ROOTFS are valid in this
 582          * context but others are. */
 583         if (clearenv()) {
 584                 SYSERROR("failed to clear environment");
 585                 /* don't error out though */
 586         }
 587
 588         if (putenv("container=lxc")) {
 589                 SYSERROR("failed to set environment variable");
 590                 goto out_warn_father;
 591         }
 592
 593         process_lock();
 594         close(handler->sigfd);
 595         process_unlock();
 596
 597         /* after this call, we are in error because this
 598          * ops should not return as it execs */
 599         handler->ops->start(handler, handler->data);
 600
 601 out_warn_father:
 602         /* we want the parent to know something went wrong, so any
 603          * value other than what it expects is ok. */
 604         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 605         return -1;
 606 }
 607
 608 int save_phys_nics(struct lxc_conf *conf)
 609 {
 610         struct lxc_list *iterator;
 611
 612         lxc_list_for_each(iterator, &conf->network) {
 613                 struct lxc_netdev *netdev = iterator->elem;
 614
 615                 if (netdev->type != LXC_NET_PHYS)
 616                         continue;
 617                 conf->saved_nics = realloc(conf->saved_nics,
 618                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 619                 if (!conf->saved_nics) {
 620                         SYSERROR("failed to allocate memory");
 621                         return -1;
 622                 }
 623                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 624                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 625                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 626                         SYSERROR("failed to allocate memory");
 627                         return -1;
 628                 }
 629                 INFO("stored saved_nic #%d idx %d name %s\n", conf->num_savednics,
 630                         conf->saved_nics[conf->num_savednics].ifindex,
 631                         conf->saved_nics[conf->num_savednics].orig_name);
 632                 conf->num_savednics++;
 633         }
 634
 635         return 0;
 636 }
 637
 638 int lxc_spawn(struct lxc_handler *handler)
 639 {
 640         int failed_before_rename = 0;
 641         const char *name = handler->name;
 642         struct cgroup_meta_data *cgroup_meta = NULL;
 643         const char *cgroup_pattern = NULL;
 644
 645         if (lxc_sync_init(handler))
 646                 return -1;
 647
 648         handler->clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 649         if (!lxc_list_empty(&handler->conf->id_map)) {
 650                 INFO("Cloning a new user namespace");
 651                 handler->clone_flags |= CLONE_NEWUSER;
 652         }
 653         if (!lxc_list_empty(&handler->conf->network)) {
 654
 655                 handler->clone_flags |= CLONE_NEWNET;
 656
 657                 /* Find gateway addresses from the link device, which is
 658                  * no longer accessible inside the container. Do this
 659                  * before creating network interfaces, since goto
 660                  * out_delete_net does not work before lxc_clone. */
 661                 if (lxc_find_gateway_addresses(handler)) {
 662                         ERROR("failed to find gateway addresses");
 663                         lxc_sync_fini(handler);
 664                         return -1;
 665                 }
 666
 667                 /* that should be done before the clone because we will
 668                  * fill the netdev index and use them in the child
 669                  */
 670                 if (lxc_create_network(handler)) {
 671                         ERROR("failed to create the network");
 672                         lxc_sync_fini(handler);
 673                         return -1;
 674                 }
 675         }
 676
 677         if (save_phys_nics(handler->conf)) {
 678                 ERROR("failed to save physical nic info");
 679                 goto out_abort;
 680         }
 681
 682         cgroup_meta = lxc_cgroup_load_meta();
 683         if (!cgroup_meta) {
 684                 ERROR("failed to detect cgroup metadata");
 685                 goto out_delete_net;
 686         }
 687
 688         /* if we are running as root, use system cgroup pattern, otherwise
 689          * just create a cgroup under the current one. But also fall back to
 690          * that if for some reason reading the configuration fails and no
 691          * default value is available
 692          */
 693         if (getuid() == 0)
 694                 cgroup_pattern = lxc_global_config_value("cgroup.pattern");
 695         if (!cgroup_pattern)
 696                 cgroup_pattern = "%n";
 697
 698         /* Create cgroup before doing clone(), so the child will know from
 699          * handler which cgroup it is going to be put in later.
 700          */
 701         if ((handler->cgroup = lxc_cgroup_create(name, cgroup_pattern, cgroup_meta, NULL)) == NULL) {
 702                 ERROR("failed to create cgroups for '%s'", name);
 703                 goto out_delete_net;
 704         }
 705
 706         /*
 707          * if the rootfs is not a blockdev, prevent the container from
 708          * marking it readonly.
 709          */
 710
 711         handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 712         if (handler->pinfd == -1)
 713                 INFO("failed to pin the container's rootfs");
 714
 715         /* Create a process in a new set of namespaces */
 716         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 717         if (handler->pid < 0) {
 718                 SYSERROR("failed to fork into a new namespace");
 719                 goto out_delete_net;
 720         }
 721
 722         lxc_sync_fini_child(handler);
 723
 724         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 725                 failed_before_rename = 1;
 726
 727         /* In case there is still legacy ns cgroup support in the kernel.
 728          * Should be removed at some later point in time.
 729          */
 730         if (lxc_cgroup_create_legacy(handler->cgroup, name, handler->pid) < 0) {
 731                 ERROR("failed to create legacy ns cgroups for '%s'", name);
 732                 goto out_delete_net;
 733         }
 734
 735         if (lxc_setup_cgroup_without_devices(handler, &handler->conf->cgroup)) {
 736                 ERROR("failed to setup the cgroups for '%s'", name);
 737                 goto out_delete_net;
 738         }
 739
 740         if (lxc_cgroup_enter(handler->cgroup, handler->pid, false) < 0)
 741                 goto out_delete_net;
 742
 743         if (failed_before_rename)
 744                 goto out_delete_net;
 745
 746         /* Create the network configuration */
 747         if (handler->clone_flags & CLONE_NEWNET) {
 748                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 749                         ERROR("failed to create the configured network");
 750                         goto out_delete_net;
 751                 }
 752         }
 753
 754         /* map the container uids - the container became an invalid
 755          * userid the moment it was cloned with CLONE_NEWUSER - this
 756          * call doesn't change anything immediately, but allows the
 757          * container to setuid(0) (0 being mapped to something else on
 758          * the host) later to become a valid uid again */
 759         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
 760                 ERROR("failed to set up id mapping");
 761                 goto out_delete_net;
 762         }
 763
 764         /* Tell the child to continue its initialization.  we'll get
 765          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
 766          */
 767         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 768                 goto out_delete_net;
 769
 770         if (lxc_setup_cgroup_devices(handler, &handler->conf->cgroup)) {
 771                 ERROR("failed to setup the devices cgroup for '%s'", name);
 772                 goto out_delete_net;
 773         }
 774
 775         /* Tell the child to complete its initialization and wait for
 776          * it to exec or return an error.  (the child will never
 777          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
 778          * sync pipe, causing lxc_sync_barrier_child to return
 779          * success, or return a different value, causing us to error
 780          * out).
 781          */
 782         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
 783                 return -1;
 784
 785         if (detect_shared_rootfs())
 786                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
 787
 788         /* If child is in a fresh user namespace, chown his ptys for
 789          * it */
 790         if (uid_shift_ttys(handler->pid, handler->conf))
 791                 DEBUG("Failed to chown ptys.\n");
 792
 793         if (handler->ops->post_start(handler, handler->data))
 794                 goto out_abort;
 795
 796         if (lxc_set_state(name, handler, RUNNING)) {
 797                 ERROR("failed to set state to %s",
 798                               lxc_state2str(RUNNING));
 799                 goto out_abort;
 800         }
 801
 802         lxc_cgroup_put_meta(cgroup_meta);
 803         lxc_sync_fini(handler);
 804
 805         return 0;
 806
 807 out_delete_net:
 808         if (handler->clone_flags & CLONE_NEWNET)
 809                 lxc_delete_network(handler);
 810 out_abort:
 811         lxc_cgroup_put_meta(cgroup_meta);
 812         lxc_abort(name, handler);
 813         lxc_sync_fini(handler);
 814         if (handler->pinfd >= 0) {
 815                 process_lock();
 816                 close(handler->pinfd);
 817                 process_unlock();
 818                 handler->pinfd = -1;
 819         }
 820
 821         return -1;
 822 }
 823
 824 int __lxc_start(const char *name, struct lxc_conf *conf,
 825                 struct lxc_operations* ops, void *data, const char *lxcpath)
 826 {
 827         struct lxc_handler *handler;
 828         int err = -1;
 829         int status;
 830
 831         handler = lxc_init(name, conf, lxcpath);
 832         if (!handler) {
 833                 ERROR("failed to initialize the container");
 834                 return -1;
 835         }
 836         handler->ops = ops;
 837         handler->data = data;
 838
 839         if (must_drop_cap_sys_boot(handler->conf)) {
 840                 #if HAVE_SYS_CAPABILITY_H
 841                 DEBUG("Dropping cap_sys_boot\n");
 842                 #else
 843                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported\n");
 844                 #endif
 845         } else {
 846                 DEBUG("Not dropping cap_sys_boot or watching utmp\n");
 847                 handler->conf->need_utmp_watch = 0;
 848         }
 849
 850         err = lxc_spawn(handler);
 851         if (err) {
 852                 ERROR("failed to spawn '%s'", name);
 853                 goto out_fini_nonet;
 854         }
 855
 856         err = lxc_poll(name, handler);
 857         if (err) {
 858                 ERROR("mainloop exited with an error");
 859                 goto out_abort;
 860         }
 861
 862         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 863                 continue;
 864
 865         /*
 866          * If the child process exited but was not signaled,
 867          * it didn't call reboot.  This should mean it was an
 868          * lxc-execute which simply exited.  In any case, treat
 869          * it as a 'halt'
 870          */
 871         if (WIFSIGNALED(status)) {
 872                 switch(WTERMSIG(status)) {
 873                 case SIGINT: /* halt */
 874                         DEBUG("Container halting");
 875                         break;
 876                 case SIGHUP: /* reboot */
 877                         DEBUG("Container rebooting");
 878                         handler->conf->reboot = 1;
 879                         break;
 880                 default:
 881                         DEBUG("unknown exit status for init: %d\n", WTERMSIG(status));
 882                         break;
 883                 }
 884         }
 885
 886         lxc_rename_phys_nics_on_shutdown(handler->conf);
 887
 888         if (handler->pinfd >= 0) {
 889                 process_lock();
 890                 close(handler->pinfd);
 891                 process_unlock();
 892                 handler->pinfd = -1;
 893         }
 894
 895         err =  lxc_error_set_and_log(handler->pid, status);
 896 out_fini:
 897         lxc_delete_network(handler);
 898
 899 out_fini_nonet:
 900         lxc_fini(name, handler);
 901         return err;
 902
 903 out_abort:
 904         lxc_abort(name, handler);
 905         goto out_fini;
 906 }
 907
 908 struct start_args {
 909         char *const *argv;
 910 };
 911
 912 static int start(struct lxc_handler *handler, void* data)
 913 {
 914         struct start_args *arg = data;
 915
 916         NOTICE("exec'ing '%s'", arg->argv[0]);
 917
 918         execvp(arg->argv[0], arg->argv);
 919         SYSERROR("failed to exec %s", arg->argv[0]);
 920         return 0;
 921 }
 922
 923 static int post_start(struct lxc_handler *handler, void* data)
 924 {
 925         struct start_args *arg = data;
 926
 927         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
 928         return 0;
 929 }
 930
 931 static struct lxc_operations start_ops = {
 932         .start = start,
 933         .post_start = post_start
 934 };
 935
 936 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
 937               const char *lxcpath)
 938 {
 939         struct start_args start_arg = {
 940                 .argv = argv,
 941         };
 942
 943         if (lxc_check_inherited(conf, -1))
 944                 return -1;
 945
 946         conf->need_utmp_watch = 1;
 947         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath);
 948 }