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 "lsm/lsm.h"
  74
  75 lxc_log_define(lxc_start, lxc);
  76
  77 const struct ns_info ns_info[LXC_NS_MAX] = {
  78         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  79         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  80         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  81         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  82         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  83         [LXC_NS_NET] = {"net", CLONE_NEWNET}
  84 };
  85
  86 static void close_ns(int ns_fd[LXC_NS_MAX]) {
  87         int i;
  88
  89         for (i = 0; i < LXC_NS_MAX; i++) {
  90                 if (ns_fd[i] > -1) {
  91                         close(ns_fd[i]);
  92                         ns_fd[i] = -1;
  93                 }
  94         }
  95 }
  96
  97 static int preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags) {
  98         int i, saved_errno;
  99         char path[MAXPATHLEN];
 100
 101         for (i = 0; i < LXC_NS_MAX; i++)
 102                 ns_fd[i] = -1;
 103
 104         if (access("/proc/self/ns", X_OK)) {
 105                 WARN("Kernel does not support attach; preserve_ns ignored");
 106                 return 0;
 107         }
 108
 109         for (i = 0; i < LXC_NS_MAX; i++) {
 110                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 111                         continue;
 112                 snprintf(path, MAXPATHLEN, "/proc/self/ns/%s", ns_info[i].proc_name);
 113                 ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC);
 114                 if (ns_fd[i] < 0)
 115                         goto error;
 116         }
 117
 118         return 0;
 119
 120 error:
 121         saved_errno = errno;
 122         close_ns(ns_fd);
 123         errno = saved_errno;
 124         SYSERROR("failed to open '%s'", path);
 125         return -1;
 126 }
 127
 128 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 129         int i;
 130
 131         for (i = 0; i < LXC_NS_MAX; i++) {
 132                 if (ns_fd[i] < 0)
 133                         continue;
 134
 135                 if (setns(ns_fd[i], 0) != 0)
 136                         goto error;
 137         }
 138         return 0;
 139
 140 error:
 141         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 142         return -1;
 143 }
 144
 145 static int match_fd(int fd)
 146 {
 147         return (fd == 0 || fd == 1 || fd == 2);
 148 }
 149
 150 int lxc_check_inherited(struct lxc_conf *conf, int fd_to_ignore)
 151 {
 152         struct dirent dirent, *direntp;
 153         int fd, fddir;
 154         DIR *dir;
 155
 156 restart:
 157         dir = opendir("/proc/self/fd");
 158         if (!dir) {
 159                 WARN("failed to open directory: %m");
 160                 return -1;
 161         }
 162
 163         fddir = dirfd(dir);
 164
 165         while (!readdir_r(dir, &dirent, &direntp)) {
 166                 if (!direntp)
 167                         break;
 168
 169                 if (!strcmp(direntp->d_name, "."))
 170                         continue;
 171
 172                 if (!strcmp(direntp->d_name, ".."))
 173                         continue;
 174
 175                 fd = atoi(direntp->d_name);
 176
 177                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 178                         continue;
 179
 180                 if (match_fd(fd))
 181                         continue;
 182
 183                 if (conf->close_all_fds) {
 184                         close(fd);
 185                         closedir(dir);
 186                         INFO("closed inherited fd %d", fd);
 187                         goto restart;
 188                 }
 189                 WARN("inherited fd %d", fd);
 190         }
 191
 192         closedir(dir); /* cannot fail */
 193         return 0;
 194 }
 195
 196 static int setup_signal_fd(sigset_t *oldmask)
 197 {
 198         sigset_t mask;
 199         int fd;
 200
 201         /* Block everything except serious error signals */
 202         if (sigfillset(&mask) ||
 203             sigdelset(&mask, SIGILL) ||
 204             sigdelset(&mask, SIGSEGV) ||
 205             sigdelset(&mask, SIGBUS) ||
 206             sigdelset(&mask, SIGWINCH) ||
 207             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 208                 SYSERROR("failed to set signal mask");
 209                 return -1;
 210         }
 211
 212         fd = signalfd(-1, &mask, 0);
 213         if (fd < 0) {
 214                 SYSERROR("failed to create the signal fd");
 215                 return -1;
 216         }
 217
 218         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 219                 SYSERROR("failed to set sigfd to close-on-exec");
 220                 close(fd);
 221                 return -1;
 222         }
 223
 224         DEBUG("sigchild handler set");
 225
 226         return fd;
 227 }
 228
 229 static int signal_handler(int fd, uint32_t events, void *data,
 230                            struct lxc_epoll_descr *descr)
 231 {
 232         struct signalfd_siginfo siginfo;
 233         siginfo_t info;
 234         int ret;
 235         pid_t *pid = data;
 236         bool init_died = false;
 237
 238         ret = read(fd, &siginfo, sizeof(siginfo));
 239         if (ret < 0) {
 240                 ERROR("failed to read signal info");
 241                 return -1;
 242         }
 243
 244         if (ret != sizeof(siginfo)) {
 245                 ERROR("unexpected siginfo size");
 246                 return -1;
 247         }
 248
 249         // check whether init is running
 250         info.si_pid = 0;
 251         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 252         if (ret == 0 && info.si_pid == *pid) {
 253                 init_died = true;
 254         }
 255
 256         if (siginfo.ssi_signo != SIGCHLD) {
 257                 kill(*pid, siginfo.ssi_signo);
 258                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 259                 return init_died ? 1 : 0;
 260         }
 261
 262         if (siginfo.ssi_code == CLD_STOPPED ||
 263             siginfo.ssi_code == CLD_CONTINUED) {
 264                 INFO("container init process was stopped/continued");
 265                 return init_died ? 1 : 0;
 266         }
 267
 268         /* more robustness, protect ourself from a SIGCHLD sent
 269          * by a process different from the container init
 270          */
 271         if (siginfo.ssi_pid != *pid) {
 272                 WARN("invalid pid for SIGCHLD");
 273                 return init_died ? 1 : 0;
 274         }
 275
 276         DEBUG("container init process exited");
 277         return 1;
 278 }
 279
 280 static int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 281 {
 282         handler->state = state;
 283         lxc_monitor_send_state(name, state, handler->lxcpath);
 284         return 0;
 285 }
 286
 287 static int lxc_poll(const char *name, struct lxc_handler *handler)
 288 {
 289         int sigfd = handler->sigfd;
 290         int pid = handler->pid;
 291         struct lxc_epoll_descr descr;
 292
 293         if (lxc_mainloop_open(&descr)) {
 294                 ERROR("failed to create mainloop");
 295                 goto out_sigfd;
 296         }
 297
 298         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 299                 ERROR("failed to add handler for the signal");
 300                 goto out_mainloop_open;
 301         }
 302
 303         if (lxc_console_mainloop_add(&descr, handler)) {
 304                 ERROR("failed to add console handler to mainloop");
 305                 goto out_mainloop_open;
 306         }
 307
 308         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 309                 ERROR("failed to add command handler to mainloop");
 310                 goto out_mainloop_open;
 311         }
 312
 313         if (handler->conf->need_utmp_watch) {
 314                 #if HAVE_SYS_CAPABILITY_H
 315                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 316                         ERROR("failed to add utmp handler to mainloop");
 317                         goto out_mainloop_open;
 318                 }
 319                 #else
 320                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support\n");
 321                 #endif
 322         }
 323
 324         return lxc_mainloop(&descr, -1);
 325
 326 out_mainloop_open:
 327         lxc_mainloop_close(&descr);
 328 out_sigfd:
 329         close(sigfd);
 330         return -1;
 331 }
 332
 333 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 334 {
 335         struct lxc_handler *handler;
 336
 337         handler = malloc(sizeof(*handler));
 338         if (!handler)
 339                 return NULL;
 340
 341         memset(handler, 0, sizeof(*handler));
 342
 343         handler->conf = conf;
 344         handler->lxcpath = lxcpath;
 345         handler->pinfd = -1;
 346
 347         lsm_init();
 348
 349         handler->name = strdup(name);
 350         if (!handler->name) {
 351                 ERROR("failed to allocate memory");
 352                 goto out_free;
 353         }
 354
 355         if (lxc_cmd_init(name, handler, lxcpath))
 356                 goto out_free_name;
 357
 358         if (lxc_read_seccomp_config(conf) != 0) {
 359                 ERROR("failed loading seccomp policy");
 360                 goto out_close_maincmd_fd;
 361         }
 362
 363         /* Begin by setting the state to STARTING */
 364         if (lxc_set_state(name, handler, STARTING)) {
 365                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 366                 goto out_close_maincmd_fd;
 367         }
 368
 369         /* Start of environment variable setup for hooks */
 370         if (setenv("LXC_NAME", name, 1)) {
 371                 SYSERROR("failed to set environment variable for container name");
 372         }
 373         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 374                 SYSERROR("failed to set environment variable for config path");
 375         }
 376         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 377                 SYSERROR("failed to set environment variable for rootfs mount");
 378         }
 379         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 380                 SYSERROR("failed to set environment variable for rootfs mount");
 381         }
 382         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 383                 SYSERROR("failed to set environment variable for console path");
 384         }
 385         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 386                 SYSERROR("failed to set environment variable for console log");
 387         }
 388         /* End of environment variable setup for hooks */
 389
 390         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 391                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 392                 goto out_aborting;
 393         }
 394
 395         if (lxc_create_tty(name, conf)) {
 396                 ERROR("failed to create the ttys");
 397                 goto out_aborting;
 398         }
 399
 400         /* the signal fd has to be created before forking otherwise
 401          * if the child process exits before we setup the signal fd,
 402          * the event will be lost and the command will be stuck */
 403         handler->sigfd = setup_signal_fd(&handler->oldmask);
 404         if (handler->sigfd < 0) {
 405                 ERROR("failed to set sigchild fd handler");
 406                 goto out_delete_tty;
 407         }
 408
 409         /* do this after setting up signals since it might unblock SIGWINCH */
 410         if (lxc_console_create(conf)) {
 411                 ERROR("failed to create console");
 412                 goto out_restore_sigmask;
 413         }
 414
 415         if (ttys_shift_ids(conf) < 0) {
 416                 ERROR("Failed to shift tty into container");
 417                 goto out_restore_sigmask;
 418         }
 419
 420         INFO("'%s' is initialized", name);
 421         return handler;
 422
 423 out_restore_sigmask:
 424         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 425 out_delete_tty:
 426         lxc_delete_tty(&conf->tty_info);
 427 out_aborting:
 428         lxc_set_state(name, handler, ABORTING);
 429 out_close_maincmd_fd:
 430         close(conf->maincmd_fd);
 431         conf->maincmd_fd = -1;
 432 out_free_name:
 433         free(handler->name);
 434         handler->name = NULL;
 435 out_free:
 436         free(handler);
 437         return NULL;
 438 }
 439
 440 static void lxc_fini(const char *name, struct lxc_handler *handler)
 441 {
 442         /* The STOPPING state is there for future cleanup code
 443          * which can take awhile
 444          */
 445         lxc_set_state(name, handler, STOPPING);
 446         lxc_set_state(name, handler, STOPPED);
 447
 448         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 449                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 450
 451         /* reset mask set by setup_signal_fd */
 452         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 453                 WARN("failed to restore sigprocmask");
 454
 455         lxc_console_delete(&handler->conf->console);
 456         lxc_delete_tty(&handler->conf->tty_info);
 457         close(handler->conf->maincmd_fd);
 458         handler->conf->maincmd_fd = -1;
 459         free(handler->name);
 460         if (handler->cgroup) {
 461                 lxc_cgroup_process_info_free_and_remove(handler->cgroup);
 462                 handler->cgroup = NULL;
 463         }
 464         free(handler);
 465 }
 466
 467 static void lxc_abort(const char *name, struct lxc_handler *handler)
 468 {
 469         int ret, status;
 470
 471         lxc_set_state(name, handler, ABORTING);
 472         if (handler->pid > 0)
 473                 kill(handler->pid, SIGKILL);
 474         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 475 }
 476
 477 #include <sys/reboot.h>
 478 #include <linux/reboot.h>
 479
 480 /*
 481  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 482  * in a child pid namespace if container reboot support exists.
 483  * Otherwise, it will either succeed or return -EPERM.
 484  */
 485 static int container_reboot_supported(void *arg)
 486 {
 487         int *cmd = arg;
 488         int ret;
 489
 490         ret = reboot(*cmd);
 491         if (ret == -1 && errno == EINVAL)
 492                 return 1;
 493         return 0;
 494 }
 495
 496 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 497 {
 498         FILE *f;
 499         int ret, cmd, v, flags;
 500         long stack_size = 4096;
 501         void *stack = alloca(stack_size);
 502         int status;
 503         pid_t pid;
 504
 505         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 506         if (!f) {
 507                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 508                 return 1;
 509         }
 510
 511         ret = fscanf(f, "%d", &v);
 512         fclose(f);
 513         if (ret != 1) {
 514                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 515                 return 1;
 516         }
 517         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 518
 519         flags = CLONE_NEWPID | SIGCHLD;
 520         if (!lxc_list_empty(&conf->id_map))
 521                 flags |= CLONE_NEWUSER;
 522
 523 #ifdef __ia64__
 524         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 525 #else
 526         stack += stack_size;
 527         pid = clone(container_reboot_supported, stack, flags, &cmd);
 528 #endif
 529         if (pid < 0) {
 530                 SYSERROR("failed to clone\n");
 531                 return -1;
 532         }
 533         if (wait(&status) < 0) {
 534                 SYSERROR("unexpected wait error: %m\n");
 535                 return -1;
 536         }
 537
 538         if (WEXITSTATUS(status) != 1)
 539                 return 1;
 540
 541         return 0;
 542 }
 543
 544 static int do_start(void *data)
 545 {
 546         struct lxc_handler *handler = data;
 547         const char *lsm_label = NULL;
 548
 549         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 550                 SYSERROR("failed to set sigprocmask");
 551                 return -1;
 552         }
 553
 554         /* This prctl must be before the synchro, so if the parent
 555          * dies before we set the parent death signal, we will detect
 556          * its death with the synchro right after, otherwise we have
 557          * a window where the parent can exit before we set the pdeath
 558          * signal leading to a unsupervized container.
 559          */
 560         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 561                 SYSERROR("failed to set pdeath signal");
 562                 return -1;
 563         }
 564
 565         lxc_sync_fini_parent(handler);
 566
 567         /* don't leak the pinfd to the container */
 568         if (handler->pinfd >= 0) {
 569                 close(handler->pinfd);
 570         }
 571
 572         /* Tell the parent task it can begin to configure the
 573          * container and wait for it to finish
 574          */
 575         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 576                 return -1;
 577
 578         /*
 579          * if we are in a new user namespace, become root there to have
 580          * privilege over our namespace
 581          */
 582         if (!lxc_list_empty(&handler->conf->id_map)) {
 583                 NOTICE("switching to gid/uid 0 in new user namespace");
 584                 if (setgid(0)) {
 585                         SYSERROR("setgid");
 586                         goto out_warn_father;
 587                 }
 588                 if (setuid(0)) {
 589                         SYSERROR("setuid");
 590                         goto out_warn_father;
 591                 }
 592         }
 593
 594         #if HAVE_SYS_CAPABILITY_H
 595         if (handler->conf->need_utmp_watch) {
 596                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 597                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 598                         goto out_warn_father;
 599                 }
 600                 DEBUG("Dropped cap_sys_boot\n");
 601         }
 602         #endif
 603
 604         /* Setup the container, ip, names, utsname, ... */
 605         if (lxc_setup(handler->name, handler->conf, handler->lxcpath, handler->cgroup, handler->data) ){
 606                 ERROR("failed to setup the container");
 607                 goto out_warn_father;
 608         }
 609
 610         /* ask father to setup cgroups and wait for him to finish */
 611         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 612                 return -1;
 613
 614         /* Set the label to change to when we exec(2) the container's init */
 615         if (!strcmp(lsm_name(), "AppArmor"))
 616                 lsm_label = handler->conf->lsm_aa_profile;
 617         else if (!strcmp(lsm_name(), "SELinux"))
 618                 lsm_label = handler->conf->lsm_se_context;
 619         if (lsm_process_label_set(lsm_label, 1, 1) < 0)
 620                 goto out_warn_father;
 621         lsm_proc_unmount(handler->conf);
 622
 623         if (lxc_seccomp_load(handler->conf) != 0)
 624                 goto out_warn_father;
 625
 626         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 627                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 628                 goto out_warn_father;
 629         }
 630
 631         /* The clearenv() and putenv() calls have been moved here
 632          * to allow us to use enviroment variables passed to the various
 633          * hooks, such as the start hook above.  Not all of the
 634          * variables like CONFIG_PATH or ROOTFS are valid in this
 635          * context but others are. */
 636         if (clearenv()) {
 637                 SYSERROR("failed to clear environment");
 638                 /* don't error out though */
 639         }
 640
 641         if (putenv("container=lxc")) {
 642                 SYSERROR("failed to set environment variable");
 643                 goto out_warn_father;
 644         }
 645
 646         close(handler->sigfd);
 647
 648         /* after this call, we are in error because this
 649          * ops should not return as it execs */
 650         handler->ops->start(handler, handler->data);
 651
 652 out_warn_father:
 653         /* we want the parent to know something went wrong, so any
 654          * value other than what it expects is ok. */
 655         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 656         return -1;
 657 }
 658
 659 static int save_phys_nics(struct lxc_conf *conf)
 660 {
 661         struct lxc_list *iterator;
 662
 663         lxc_list_for_each(iterator, &conf->network) {
 664                 struct lxc_netdev *netdev = iterator->elem;
 665
 666                 if (netdev->type != LXC_NET_PHYS)
 667                         continue;
 668                 conf->saved_nics = realloc(conf->saved_nics,
 669                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 670                 if (!conf->saved_nics) {
 671                         SYSERROR("failed to allocate memory");
 672                         return -1;
 673                 }
 674                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 675                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 676                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 677                         SYSERROR("failed to allocate memory");
 678                         return -1;
 679                 }
 680                 INFO("stored saved_nic #%d idx %d name %s\n", conf->num_savednics,
 681                         conf->saved_nics[conf->num_savednics].ifindex,
 682                         conf->saved_nics[conf->num_savednics].orig_name);
 683                 conf->num_savednics++;
 684         }
 685
 686         return 0;
 687 }
 688
 689 static int lxc_spawn(struct lxc_handler *handler)
 690 {
 691         int failed_before_rename = 0;
 692         const char *name = handler->name;
 693         struct cgroup_meta_data *cgroup_meta = NULL;
 694         const char *cgroup_pattern = NULL;
 695         int saved_ns_fd[LXC_NS_MAX];
 696         int preserve_mask = 0, i;
 697
 698         for (i = 0; i < LXC_NS_MAX; i++)
 699                 if (handler->conf->inherit_ns_fd[i] != -1)
 700                         preserve_mask |= ns_info[i].clone_flag;
 701
 702         if (lxc_sync_init(handler))
 703                 return -1;
 704
 705         handler->clone_flags = CLONE_NEWPID|CLONE_NEWNS;
 706         if (!lxc_list_empty(&handler->conf->id_map)) {
 707                 INFO("Cloning a new user namespace");
 708                 handler->clone_flags |= CLONE_NEWUSER;
 709         }
 710
 711         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 712                 if (!lxc_requests_empty_network(handler))
 713                         handler->clone_flags |= CLONE_NEWNET;
 714
 715                 if (!lxc_list_empty(&handler->conf->network)) {
 716
 717                         /* Find gateway addresses from the link device, which is
 718                          * no longer accessible inside the container. Do this
 719                          * before creating network interfaces, since goto
 720                          * out_delete_net does not work before lxc_clone. */
 721                         if (lxc_find_gateway_addresses(handler)) {
 722                                 ERROR("failed to find gateway addresses");
 723                                 lxc_sync_fini(handler);
 724                                 return -1;
 725                         }
 726
 727                         /* that should be done before the clone because we will
 728                          * fill the netdev index and use them in the child
 729                          */
 730                         if (lxc_create_network(handler)) {
 731                                 ERROR("failed to create the network");
 732                                 lxc_sync_fini(handler);
 733                                 return -1;
 734                         }
 735                 }
 736
 737                 if (save_phys_nics(handler->conf)) {
 738                         ERROR("failed to save physical nic info");
 739                         goto out_abort;
 740                 }
 741         } else {
 742                 INFO("Inheriting a net namespace");
 743         }
 744
 745         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 746                 handler->clone_flags |= CLONE_NEWIPC;
 747         } else {
 748                 INFO("Inheriting an IPC namespace");
 749         }
 750
 751         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 752                 handler->clone_flags |= CLONE_NEWUTS;
 753         } else {
 754                 INFO("Inheriting a UTS namespace");
 755         }
 756
 757
 758         cgroup_meta = lxc_cgroup_load_meta();
 759         if (!cgroup_meta) {
 760                 ERROR("failed to detect cgroup metadata");
 761                 goto out_delete_net;
 762         }
 763
 764         /* if we are running as root, use system cgroup pattern, otherwise
 765          * just create a cgroup under the current one. But also fall back to
 766          * that if for some reason reading the configuration fails and no
 767          * default value is available
 768          */
 769         if (getuid() == 0)
 770                 cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern");
 771         if (!cgroup_pattern)
 772                 cgroup_pattern = "%n";
 773
 774         /* Create cgroup before doing clone(), so the child will know from
 775          * handler which cgroup it is going to be put in later.
 776          */
 777         if ((handler->cgroup = lxc_cgroup_create(name, cgroup_pattern, cgroup_meta, NULL)) == NULL) {
 778                 ERROR("failed to create cgroups for '%s'", name);
 779                 goto out_delete_net;
 780         }
 781
 782         /*
 783          * if the rootfs is not a blockdev, prevent the container from
 784          * marking it readonly.
 785          */
 786
 787         handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 788         if (handler->pinfd == -1)
 789                 INFO("failed to pin the container's rootfs");
 790
 791         if (preserve_ns(saved_ns_fd, preserve_mask) < 0)
 792                 goto out_delete_net;
 793         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
 794                 goto out_delete_net;
 795
 796         /* Create a process in a new set of namespaces */
 797         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 798         if (handler->pid < 0) {
 799                 SYSERROR("failed to fork into a new namespace");
 800                 goto out_delete_net;
 801         }
 802
 803         if (attach_ns(saved_ns_fd))
 804                 WARN("failed to restore saved namespaces");
 805
 806         lxc_sync_fini_child(handler);
 807
 808         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 809                 failed_before_rename = 1;
 810
 811         /* In case there is still legacy ns cgroup support in the kernel.
 812          * Should be removed at some later point in time.
 813          */
 814         if (lxc_cgroup_create_legacy(handler->cgroup, name, handler->pid) < 0) {
 815                 ERROR("failed to create legacy ns cgroups for '%s'", name);
 816                 goto out_delete_net;
 817         }
 818
 819         if (lxc_setup_cgroup_without_devices(handler, &handler->conf->cgroup)) {
 820                 ERROR("failed to setup the cgroups for '%s'", name);
 821                 goto out_delete_net;
 822         }
 823
 824         if (lxc_cgroup_enter(handler->cgroup, handler->pid, false) < 0)
 825                 goto out_delete_net;
 826
 827         if (failed_before_rename)
 828                 goto out_delete_net;
 829
 830         /* Create the network configuration */
 831         if (handler->clone_flags & CLONE_NEWNET) {
 832                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 833                         ERROR("failed to create the configured network");
 834                         goto out_delete_net;
 835                 }
 836         }
 837
 838         /* map the container uids - the container became an invalid
 839          * userid the moment it was cloned with CLONE_NEWUSER - this
 840          * call doesn't change anything immediately, but allows the
 841          * container to setuid(0) (0 being mapped to something else on
 842          * the host) later to become a valid uid again */
 843         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
 844                 ERROR("failed to set up id mapping");
 845                 goto out_delete_net;
 846         }
 847
 848         /* Tell the child to continue its initialization.  we'll get
 849          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
 850          */
 851         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 852                 goto out_delete_net;
 853
 854         if (lxc_setup_cgroup_devices(handler, &handler->conf->cgroup)) {
 855                 ERROR("failed to setup the devices cgroup for '%s'", name);
 856                 goto out_delete_net;
 857         }
 858
 859         /* Tell the child to complete its initialization and wait for
 860          * it to exec or return an error.  (the child will never
 861          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
 862          * sync pipe, causing lxc_sync_barrier_child to return
 863          * success, or return a different value, causing us to error
 864          * out).
 865          */
 866         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
 867                 return -1;
 868
 869         if (detect_shared_rootfs())
 870                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
 871
 872         if (handler->ops->post_start(handler, handler->data))
 873                 goto out_abort;
 874
 875         if (lxc_set_state(name, handler, RUNNING)) {
 876                 ERROR("failed to set state to %s",
 877                               lxc_state2str(RUNNING));
 878                 goto out_abort;
 879         }
 880
 881         lxc_cgroup_put_meta(cgroup_meta);
 882         lxc_sync_fini(handler);
 883
 884         return 0;
 885
 886 out_delete_net:
 887         if (handler->clone_flags & CLONE_NEWNET)
 888                 lxc_delete_network(handler);
 889 out_abort:
 890         lxc_cgroup_put_meta(cgroup_meta);
 891         lxc_abort(name, handler);
 892         lxc_sync_fini(handler);
 893         if (handler->pinfd >= 0) {
 894                 close(handler->pinfd);
 895                 handler->pinfd = -1;
 896         }
 897
 898         return -1;
 899 }
 900
 901 int __lxc_start(const char *name, struct lxc_conf *conf,
 902                 struct lxc_operations* ops, void *data, const char *lxcpath)
 903 {
 904         struct lxc_handler *handler;
 905         int err = -1;
 906         int status;
 907
 908         handler = lxc_init(name, conf, lxcpath);
 909         if (!handler) {
 910                 ERROR("failed to initialize the container");
 911                 return -1;
 912         }
 913         handler->ops = ops;
 914         handler->data = data;
 915
 916         if (must_drop_cap_sys_boot(handler->conf)) {
 917                 #if HAVE_SYS_CAPABILITY_H
 918                 DEBUG("Dropping cap_sys_boot\n");
 919                 #else
 920                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported\n");
 921                 #endif
 922         } else {
 923                 DEBUG("Not dropping cap_sys_boot or watching utmp\n");
 924                 handler->conf->need_utmp_watch = 0;
 925         }
 926
 927         err = lxc_spawn(handler);
 928         if (err) {
 929                 ERROR("failed to spawn '%s'", name);
 930                 goto out_fini_nonet;
 931         }
 932
 933         err = lxc_poll(name, handler);
 934         if (err) {
 935                 ERROR("mainloop exited with an error");
 936                 goto out_abort;
 937         }
 938
 939         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 940                 continue;
 941
 942         /*
 943          * If the child process exited but was not signaled,
 944          * it didn't call reboot.  This should mean it was an
 945          * lxc-execute which simply exited.  In any case, treat
 946          * it as a 'halt'
 947          */
 948         if (WIFSIGNALED(status)) {
 949                 switch(WTERMSIG(status)) {
 950                 case SIGINT: /* halt */
 951                         DEBUG("Container halting");
 952                         break;
 953                 case SIGHUP: /* reboot */
 954                         DEBUG("Container rebooting");
 955                         handler->conf->reboot = 1;
 956                         break;
 957                 default:
 958                         DEBUG("unknown exit status for init: %d\n", WTERMSIG(status));
 959                         break;
 960                 }
 961         }
 962
 963         lxc_rename_phys_nics_on_shutdown(handler->conf);
 964
 965         if (handler->pinfd >= 0) {
 966                 close(handler->pinfd);
 967                 handler->pinfd = -1;
 968         }
 969
 970         err =  lxc_error_set_and_log(handler->pid, status);
 971 out_fini:
 972         lxc_delete_network(handler);
 973
 974 out_fini_nonet:
 975         lxc_fini(name, handler);
 976         return err;
 977
 978 out_abort:
 979         lxc_abort(name, handler);
 980         goto out_fini;
 981 }
 982
 983 struct start_args {
 984         char *const *argv;
 985 };
 986
 987 static int start(struct lxc_handler *handler, void* data)
 988 {
 989         struct start_args *arg = data;
 990
 991         NOTICE("exec'ing '%s'", arg->argv[0]);
 992
 993         execvp(arg->argv[0], arg->argv);
 994         SYSERROR("failed to exec %s", arg->argv[0]);
 995         return 0;
 996 }
 997
 998 static int post_start(struct lxc_handler *handler, void* data)
 999 {
1000         struct start_args *arg = data;
1001
1002         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1003         return 0;
1004 }
1005
1006 static struct lxc_operations start_ops = {
1007         .start = start,
1008         .post_start = post_start
1009 };
1010
1011 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1012               const char *lxcpath)
1013 {
1014         struct start_args start_arg = {
1015                 .argv = argv,
1016         };
1017
1018         if (lxc_check_inherited(conf, -1))
1019                 return -1;
1020
1021         conf->need_utmp_watch = 1;
1022         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath);
1023 }