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 const struct ns_info ns_info[LXC_NS_MAX] = {
  79         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  80         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  81         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  82         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  83         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  84         [LXC_NS_NET] = {"net", CLONE_NEWNET}
  85 };
  86
  87 static void close_ns(int ns_fd[LXC_NS_MAX]) {
  88         int i;
  89
  90         process_lock();
  91         for (i = 0; i < LXC_NS_MAX; i++) {
  92                 if (ns_fd[i] > -1) {
  93                         close(ns_fd[i]);
  94                         ns_fd[i] = -1;
  95                 }
  96         }
  97         process_unlock();
  98 }
  99
 100 static int preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags) {
 101         int i, saved_errno;
 102         char path[MAXPATHLEN];
 103
 104         if (access("/proc/self/ns", X_OK)) {
 105                 ERROR("Does this kernel version support 'attach'?");
 106                 return -1;
 107         }
 108
 109         for (i = 0; i < LXC_NS_MAX; i++)
 110                 ns_fd[i] = -1;
 111
 112         for (i = 0; i < LXC_NS_MAX; i++) {
 113                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 114                         continue;
 115                 snprintf(path, MAXPATHLEN, "/proc/self/ns/%s", ns_info[i].proc_name);
 116                 process_lock();
 117                 ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC);
 118                 process_unlock();
 119                 if (ns_fd[i] < 0)
 120                         goto error;
 121         }
 122
 123         return 0;
 124
 125 error:
 126         saved_errno = errno;
 127         close_ns(ns_fd);
 128         errno = saved_errno;
 129         SYSERROR("failed to open '%s'", path);
 130         return -1;
 131 }
 132
 133 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 134         int i;
 135
 136         for (i = 0; i < LXC_NS_MAX; i++) {
 137                 if (ns_fd[i] < 0)
 138                         continue;
 139
 140                 if (setns(ns_fd[i], 0) != 0)
 141                         goto error;
 142         }
 143         return 0;
 144
 145 error:
 146         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 147         return -1;
 148 }
 149
 150 static int match_fd(int fd)
 151 {
 152         return (fd == 0 || fd == 1 || fd == 2);
 153 }
 154
 155 int lxc_check_inherited(struct lxc_conf *conf, int fd_to_ignore)
 156 {
 157         struct dirent dirent, *direntp;
 158         int fd, fddir;
 159         DIR *dir;
 160
 161 restart:
 162         process_lock();
 163         dir = opendir("/proc/self/fd");
 164         process_unlock();
 165         if (!dir) {
 166                 WARN("failed to open directory: %m");
 167                 return -1;
 168         }
 169
 170         fddir = dirfd(dir);
 171
 172         while (!readdir_r(dir, &dirent, &direntp)) {
 173                 if (!direntp)
 174                         break;
 175
 176                 if (!strcmp(direntp->d_name, "."))
 177                         continue;
 178
 179                 if (!strcmp(direntp->d_name, ".."))
 180                         continue;
 181
 182                 fd = atoi(direntp->d_name);
 183
 184                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 185                         continue;
 186
 187                 if (match_fd(fd))
 188                         continue;
 189
 190                 if (conf->close_all_fds) {
 191                         process_lock();
 192                         close(fd);
 193                         closedir(dir);
 194                         process_unlock();
 195                         INFO("closed inherited fd %d", fd);
 196                         goto restart;
 197                 }
 198                 WARN("inherited fd %d", fd);
 199         }
 200
 201         process_lock();
 202         closedir(dir); /* cannot fail */
 203         process_unlock();
 204         return 0;
 205 }
 206
 207 static int setup_signal_fd(sigset_t *oldmask)
 208 {
 209         sigset_t mask;
 210         int fd;
 211
 212         /* Block everything except serious error signals */
 213         if (sigfillset(&mask) ||
 214             sigdelset(&mask, SIGILL) ||
 215             sigdelset(&mask, SIGSEGV) ||
 216             sigdelset(&mask, SIGBUS) ||
 217             sigdelset(&mask, SIGWINCH) ||
 218             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 219                 SYSERROR("failed to set signal mask");
 220                 return -1;
 221         }
 222
 223         fd = signalfd(-1, &mask, 0);
 224         if (fd < 0) {
 225                 SYSERROR("failed to create the signal fd");
 226                 return -1;
 227         }
 228
 229         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 230                 SYSERROR("failed to set sigfd to close-on-exec");
 231                 close(fd);
 232                 return -1;
 233         }
 234
 235         DEBUG("sigchild handler set");
 236
 237         return fd;
 238 }
 239
 240 static int signal_handler(int fd, uint32_t events, void *data,
 241                            struct lxc_epoll_descr *descr)
 242 {
 243         struct signalfd_siginfo siginfo;
 244         siginfo_t info;
 245         int ret;
 246         pid_t *pid = data;
 247         bool init_died = false;
 248
 249         ret = read(fd, &siginfo, sizeof(siginfo));
 250         if (ret < 0) {
 251                 ERROR("failed to read signal info");
 252                 return -1;
 253         }
 254
 255         if (ret != sizeof(siginfo)) {
 256                 ERROR("unexpected siginfo size");
 257                 return -1;
 258         }
 259
 260         // check whether init is running
 261         info.si_pid = 0;
 262         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 263         if (ret == 0 && info.si_pid == *pid) {
 264                 init_died = true;
 265         }
 266
 267         if (siginfo.ssi_signo != SIGCHLD) {
 268                 kill(*pid, siginfo.ssi_signo);
 269                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 270                 return init_died ? 1 : 0;
 271         }
 272
 273         if (siginfo.ssi_code == CLD_STOPPED ||
 274             siginfo.ssi_code == CLD_CONTINUED) {
 275                 INFO("container init process was stopped/continued");
 276                 return init_died ? 1 : 0;
 277         }
 278
 279         /* more robustness, protect ourself from a SIGCHLD sent
 280          * by a process different from the container init
 281          */
 282         if (siginfo.ssi_pid != *pid) {
 283                 WARN("invalid pid for SIGCHLD");
 284                 return init_died ? 1 : 0;
 285         }
 286
 287         DEBUG("container init process exited");
 288         return 1;
 289 }
 290
 291 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 292 {
 293         handler->state = state;
 294         lxc_monitor_send_state(name, state, handler->lxcpath);
 295         return 0;
 296 }
 297
 298 int lxc_poll(const char *name, struct lxc_handler *handler)
 299 {
 300         int sigfd = handler->sigfd;
 301         int pid = handler->pid;
 302         struct lxc_epoll_descr descr;
 303
 304         if (lxc_mainloop_open(&descr)) {
 305                 ERROR("failed to create mainloop");
 306                 goto out_sigfd;
 307         }
 308
 309         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 310                 ERROR("failed to add handler for the signal");
 311                 goto out_mainloop_open;
 312         }
 313
 314         if (lxc_console_mainloop_add(&descr, handler)) {
 315                 ERROR("failed to add console handler to mainloop");
 316                 goto out_mainloop_open;
 317         }
 318
 319         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 320                 ERROR("failed to add command handler to mainloop");
 321                 goto out_mainloop_open;
 322         }
 323
 324         if (handler->conf->need_utmp_watch) {
 325                 #if HAVE_SYS_CAPABILITY_H
 326                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 327                         ERROR("failed to add utmp handler to mainloop");
 328                         goto out_mainloop_open;
 329                 }
 330                 #else
 331                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support\n");
 332                 #endif
 333         }
 334
 335         return lxc_mainloop(&descr, -1);
 336
 337 out_mainloop_open:
 338         lxc_mainloop_close(&descr);
 339 out_sigfd:
 340         process_lock();
 341         close(sigfd);
 342         process_unlock();
 343         return -1;
 344 }
 345
 346 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 347 {
 348         struct lxc_handler *handler;
 349
 350         handler = malloc(sizeof(*handler));
 351         if (!handler)
 352                 return NULL;
 353
 354         memset(handler, 0, sizeof(*handler));
 355
 356         handler->conf = conf;
 357         handler->lxcpath = lxcpath;
 358         handler->pinfd = -1;
 359
 360         lsm_init();
 361
 362         handler->name = strdup(name);
 363         if (!handler->name) {
 364                 ERROR("failed to allocate memory");
 365                 goto out_free;
 366         }
 367
 368         if (lxc_cmd_init(name, handler, lxcpath))
 369                 goto out_free_name;
 370
 371         if (lxc_read_seccomp_config(conf) != 0) {
 372                 ERROR("failed loading seccomp policy");
 373                 goto out_close_maincmd_fd;
 374         }
 375
 376         /* Begin by setting the state to STARTING */
 377         if (lxc_set_state(name, handler, STARTING)) {
 378                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 379                 goto out_close_maincmd_fd;
 380         }
 381
 382         /* Start of environment variable setup for hooks */
 383         if (setenv("LXC_NAME", name, 1)) {
 384                 SYSERROR("failed to set environment variable for container name");
 385         }
 386         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 387                 SYSERROR("failed to set environment variable for config path");
 388         }
 389         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 390                 SYSERROR("failed to set environment variable for rootfs mount");
 391         }
 392         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 393                 SYSERROR("failed to set environment variable for rootfs mount");
 394         }
 395         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 396                 SYSERROR("failed to set environment variable for console path");
 397         }
 398         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 399                 SYSERROR("failed to set environment variable for console log");
 400         }
 401         /* End of environment variable setup for hooks */
 402
 403         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 404                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 405                 goto out_aborting;
 406         }
 407
 408         if (lxc_create_tty(name, conf)) {
 409                 ERROR("failed to create the ttys");
 410                 goto out_aborting;
 411         }
 412
 413         /* the signal fd has to be created before forking otherwise
 414          * if the child process exits before we setup the signal fd,
 415          * the event will be lost and the command will be stuck */
 416         handler->sigfd = setup_signal_fd(&handler->oldmask);
 417         if (handler->sigfd < 0) {
 418                 ERROR("failed to set sigchild fd handler");
 419                 goto out_delete_tty;
 420         }
 421
 422         /* do this after setting up signals since it might unblock SIGWINCH */
 423         if (lxc_console_create(conf)) {
 424                 ERROR("failed to create console");
 425                 goto out_restore_sigmask;
 426         }
 427
 428         if (ttys_shift_ids(conf) < 0) {
 429                 ERROR("Failed to shift tty into container");
 430                 goto out_restore_sigmask;
 431         }
 432
 433         INFO("'%s' is initialized", name);
 434         return handler;
 435
 436 out_restore_sigmask:
 437         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 438 out_delete_tty:
 439         lxc_delete_tty(&conf->tty_info);
 440 out_aborting:
 441         lxc_set_state(name, handler, ABORTING);
 442 out_close_maincmd_fd:
 443         process_lock();
 444         close(conf->maincmd_fd);
 445         process_unlock();
 446         conf->maincmd_fd = -1;
 447 out_free_name:
 448         free(handler->name);
 449         handler->name = NULL;
 450 out_free:
 451         free(handler);
 452         return NULL;
 453 }
 454
 455 static void lxc_fini(const char *name, struct lxc_handler *handler)
 456 {
 457         /* The STOPPING state is there for future cleanup code
 458          * which can take awhile
 459          */
 460         lxc_set_state(name, handler, STOPPING);
 461         lxc_set_state(name, handler, STOPPED);
 462
 463         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 464                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 465
 466         /* reset mask set by setup_signal_fd */
 467         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 468                 WARN("failed to restore sigprocmask");
 469
 470         lxc_console_delete(&handler->conf->console);
 471         lxc_delete_tty(&handler->conf->tty_info);
 472         process_lock();
 473         close(handler->conf->maincmd_fd);
 474         process_unlock();
 475         handler->conf->maincmd_fd = -1;
 476         free(handler->name);
 477         if (handler->cgroup) {
 478                 lxc_cgroup_process_info_free_and_remove(handler->cgroup);
 479                 handler->cgroup = NULL;
 480         }
 481         free(handler);
 482 }
 483
 484 void lxc_abort(const char *name, struct lxc_handler *handler)
 485 {
 486         int ret, status;
 487
 488         lxc_set_state(name, handler, ABORTING);
 489         if (handler->pid > 0)
 490                 kill(handler->pid, SIGKILL);
 491         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 492 }
 493
 494 #include <sys/reboot.h>
 495 #include <linux/reboot.h>
 496
 497 /*
 498  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 499  * in a child pid namespace if container reboot support exists.
 500  * Otherwise, it will either succeed or return -EPERM.
 501  */
 502 static int container_reboot_supported(void *arg)
 503 {
 504         int *cmd = arg;
 505         int ret;
 506
 507         ret = reboot(*cmd);
 508         if (ret == -1 && errno == EINVAL)
 509                 return 1;
 510         return 0;
 511 }
 512
 513 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 514 {
 515         FILE *f;
 516         int ret, cmd, v, flags;
 517         long stack_size = 4096;
 518         void *stack = alloca(stack_size);
 519         int status;
 520         pid_t pid;
 521
 522         process_lock();
 523         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 524         process_unlock();
 525         if (!f) {
 526                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 527                 return 1;
 528         }
 529
 530         ret = fscanf(f, "%d", &v);
 531         process_lock();
 532         fclose(f);
 533         process_unlock();
 534         if (ret != 1) {
 535                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 536                 return 1;
 537         }
 538         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 539
 540         flags = CLONE_NEWPID | SIGCHLD;
 541         if (!lxc_list_empty(&conf->id_map))
 542                 flags |= CLONE_NEWUSER;
 543
 544 #ifdef __ia64__
 545         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 546 #else
 547         stack += stack_size;
 548         pid = clone(container_reboot_supported, stack, flags, &cmd);
 549 #endif
 550         if (pid < 0) {
 551                 SYSERROR("failed to clone\n");
 552                 return -1;
 553         }
 554         if (wait(&status) < 0) {
 555                 SYSERROR("unexpected wait error: %m\n");
 556                 return -1;
 557         }
 558
 559         if (WEXITSTATUS(status) != 1)
 560                 return 1;
 561
 562         return 0;
 563 }
 564
 565 static int do_start(void *data)
 566 {
 567         struct lxc_handler *handler = data;
 568         const char *lsm_label = NULL;
 569
 570         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 571                 SYSERROR("failed to set sigprocmask");
 572                 return -1;
 573         }
 574
 575         /* This prctl must be before the synchro, so if the parent
 576          * dies before we set the parent death signal, we will detect
 577          * its death with the synchro right after, otherwise we have
 578          * a window where the parent can exit before we set the pdeath
 579          * signal leading to a unsupervized container.
 580          */
 581         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 582                 SYSERROR("failed to set pdeath signal");
 583                 return -1;
 584         }
 585
 586         lxc_sync_fini_parent(handler);
 587
 588         /* don't leak the pinfd to the container */
 589         if (handler->pinfd >= 0) {
 590                 process_lock();
 591                 close(handler->pinfd);
 592                 process_unlock();
 593         }
 594
 595         /* Tell the parent task it can begin to configure the
 596          * container and wait for it to finish
 597          */
 598         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 599                 return -1;
 600
 601         /*
 602          * if we are in a new user namespace, become root there to have
 603          * privilege over our namespace
 604          */
 605         if (!lxc_list_empty(&handler->conf->id_map)) {
 606                 NOTICE("switching to gid/uid 0 in new user namespace");
 607                 if (setgid(0)) {
 608                         SYSERROR("setgid");
 609                         goto out_warn_father;
 610                 }
 611                 if (setuid(0)) {
 612                         SYSERROR("setuid");
 613                         goto out_warn_father;
 614                 }
 615         }
 616
 617         #if HAVE_SYS_CAPABILITY_H
 618         if (handler->conf->need_utmp_watch) {
 619                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 620                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 621                         goto out_warn_father;
 622                 }
 623                 DEBUG("Dropped cap_sys_boot\n");
 624         }
 625         #endif
 626
 627         /* Setup the container, ip, names, utsname, ... */
 628         if (lxc_setup(handler->name, handler->conf, handler->lxcpath, handler->cgroup, handler->data) ){
 629                 ERROR("failed to setup the container");
 630                 goto out_warn_father;
 631         }
 632
 633         /* ask father to setup cgroups and wait for him to finish */
 634         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 635                 return -1;
 636
 637         /* Set the label to change to when we exec(2) the container's init */
 638         if (!strcmp(lsm_name(), "AppArmor"))
 639                 lsm_label = handler->conf->lsm_aa_profile;
 640         else if (!strcmp(lsm_name(), "SELinux"))
 641                 lsm_label = handler->conf->lsm_se_context;
 642         if (lsm_process_label_set(lsm_label, 1, 1) < 0)
 643                 goto out_warn_father;
 644         lsm_proc_unmount(handler->conf);
 645
 646         if (lxc_seccomp_load(handler->conf) != 0)
 647                 goto out_warn_father;
 648
 649         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 650                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 651                 goto out_warn_father;
 652         }
 653
 654         /* The clearenv() and putenv() calls have been moved here
 655          * to allow us to use enviroment variables passed to the various
 656          * hooks, such as the start hook above.  Not all of the
 657          * variables like CONFIG_PATH or ROOTFS are valid in this
 658          * context but others are. */
 659         if (clearenv()) {
 660                 SYSERROR("failed to clear environment");
 661                 /* don't error out though */
 662         }
 663
 664         if (putenv("container=lxc")) {
 665                 SYSERROR("failed to set environment variable");
 666                 goto out_warn_father;
 667         }
 668
 669         process_lock();
 670         close(handler->sigfd);
 671         process_unlock();
 672
 673         /* after this call, we are in error because this
 674          * ops should not return as it execs */
 675         handler->ops->start(handler, handler->data);
 676
 677 out_warn_father:
 678         /* we want the parent to know something went wrong, so any
 679          * value other than what it expects is ok. */
 680         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 681         return -1;
 682 }
 683
 684 int save_phys_nics(struct lxc_conf *conf)
 685 {
 686         struct lxc_list *iterator;
 687
 688         lxc_list_for_each(iterator, &conf->network) {
 689                 struct lxc_netdev *netdev = iterator->elem;
 690
 691                 if (netdev->type != LXC_NET_PHYS)
 692                         continue;
 693                 conf->saved_nics = realloc(conf->saved_nics,
 694                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 695                 if (!conf->saved_nics) {
 696                         SYSERROR("failed to allocate memory");
 697                         return -1;
 698                 }
 699                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 700                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 701                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 702                         SYSERROR("failed to allocate memory");
 703                         return -1;
 704                 }
 705                 INFO("stored saved_nic #%d idx %d name %s\n", conf->num_savednics,
 706                         conf->saved_nics[conf->num_savednics].ifindex,
 707                         conf->saved_nics[conf->num_savednics].orig_name);
 708                 conf->num_savednics++;
 709         }
 710
 711         return 0;
 712 }
 713
 714 int lxc_spawn(struct lxc_handler *handler)
 715 {
 716         int failed_before_rename = 0;
 717         const char *name = handler->name;
 718         struct cgroup_meta_data *cgroup_meta = NULL;
 719         const char *cgroup_pattern = NULL;
 720         int saved_ns_fd[LXC_NS_MAX];
 721         int preserve_mask = 0, i;
 722
 723         for (i = 0; i < LXC_NS_MAX; i++)
 724                 if (handler->conf->inherit_ns_fd[i] != -1)
 725                         preserve_mask |= ns_info[i].clone_flag;
 726
 727         if (lxc_sync_init(handler))
 728                 return -1;
 729
 730         handler->clone_flags = CLONE_NEWPID|CLONE_NEWNS;
 731         if (!lxc_list_empty(&handler->conf->id_map)) {
 732                 INFO("Cloning a new user namespace");
 733                 handler->clone_flags |= CLONE_NEWUSER;
 734         }
 735
 736         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 737                 if (!lxc_requests_empty_network(handler))
 738                         handler->clone_flags |= CLONE_NEWNET;
 739
 740                 if (!lxc_list_empty(&handler->conf->network)) {
 741
 742                         /* Find gateway addresses from the link device, which is
 743                          * no longer accessible inside the container. Do this
 744                          * before creating network interfaces, since goto
 745                          * out_delete_net does not work before lxc_clone. */
 746                         if (lxc_find_gateway_addresses(handler)) {
 747                                 ERROR("failed to find gateway addresses");
 748                                 lxc_sync_fini(handler);
 749                                 return -1;
 750                         }
 751
 752                         /* that should be done before the clone because we will
 753                          * fill the netdev index and use them in the child
 754                          */
 755                         if (lxc_create_network(handler)) {
 756                                 ERROR("failed to create the network");
 757                                 lxc_sync_fini(handler);
 758                                 return -1;
 759                         }
 760                 }
 761
 762                 if (save_phys_nics(handler->conf)) {
 763                         ERROR("failed to save physical nic info");
 764                         goto out_abort;
 765                 }
 766         } else {
 767                 INFO("Inheriting a net namespace");
 768         }
 769
 770         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 771                 handler->clone_flags |= CLONE_NEWIPC;
 772         } else {
 773                 INFO("Inheriting an IPC namespace");
 774         }
 775
 776         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 777                 handler->clone_flags |= CLONE_NEWUTS;
 778         } else {
 779                 INFO("Inheriting a UTS namespace");
 780         }
 781
 782
 783         cgroup_meta = lxc_cgroup_load_meta();
 784         if (!cgroup_meta) {
 785                 ERROR("failed to detect cgroup metadata");
 786                 goto out_delete_net;
 787         }
 788
 789         /* if we are running as root, use system cgroup pattern, otherwise
 790          * just create a cgroup under the current one. But also fall back to
 791          * that if for some reason reading the configuration fails and no
 792          * default value is available
 793          */
 794         if (getuid() == 0)
 795                 cgroup_pattern = default_cgroup_pattern();
 796         if (!cgroup_pattern)
 797                 cgroup_pattern = "%n";
 798
 799         /* Create cgroup before doing clone(), so the child will know from
 800          * handler which cgroup it is going to be put in later.
 801          */
 802         if ((handler->cgroup = lxc_cgroup_create(name, cgroup_pattern, cgroup_meta, NULL)) == NULL) {
 803                 ERROR("failed to create cgroups for '%s'", name);
 804                 goto out_delete_net;
 805         }
 806
 807         /*
 808          * if the rootfs is not a blockdev, prevent the container from
 809          * marking it readonly.
 810          */
 811
 812         handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 813         if (handler->pinfd == -1)
 814                 INFO("failed to pin the container's rootfs");
 815
 816         preserve_ns(saved_ns_fd, preserve_mask);
 817         attach_ns(handler->conf->inherit_ns_fd);
 818
 819         /* Create a process in a new set of namespaces */
 820         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 821         if (handler->pid < 0) {
 822                 SYSERROR("failed to fork into a new namespace");
 823                 goto out_delete_net;
 824         }
 825
 826         attach_ns(saved_ns_fd);
 827
 828         lxc_sync_fini_child(handler);
 829
 830         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 831                 failed_before_rename = 1;
 832
 833         /* In case there is still legacy ns cgroup support in the kernel.
 834          * Should be removed at some later point in time.
 835          */
 836         if (lxc_cgroup_create_legacy(handler->cgroup, name, handler->pid) < 0) {
 837                 ERROR("failed to create legacy ns cgroups for '%s'", name);
 838                 goto out_delete_net;
 839         }
 840
 841         if (lxc_setup_cgroup_without_devices(handler, &handler->conf->cgroup)) {
 842                 ERROR("failed to setup the cgroups for '%s'", name);
 843                 goto out_delete_net;
 844         }
 845
 846         if (lxc_cgroup_enter(handler->cgroup, handler->pid, false) < 0)
 847                 goto out_delete_net;
 848
 849         if (failed_before_rename)
 850                 goto out_delete_net;
 851
 852         /* Create the network configuration */
 853         if (handler->clone_flags & CLONE_NEWNET) {
 854                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 855                         ERROR("failed to create the configured network");
 856                         goto out_delete_net;
 857                 }
 858         }
 859
 860         /* map the container uids - the container became an invalid
 861          * userid the moment it was cloned with CLONE_NEWUSER - this
 862          * call doesn't change anything immediately, but allows the
 863          * container to setuid(0) (0 being mapped to something else on
 864          * the host) later to become a valid uid again */
 865         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
 866                 ERROR("failed to set up id mapping");
 867                 goto out_delete_net;
 868         }
 869
 870         /* Tell the child to continue its initialization.  we'll get
 871          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
 872          */
 873         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 874                 goto out_delete_net;
 875
 876         if (lxc_setup_cgroup_devices(handler, &handler->conf->cgroup)) {
 877                 ERROR("failed to setup the devices cgroup for '%s'", name);
 878                 goto out_delete_net;
 879         }
 880
 881         /* Tell the child to complete its initialization and wait for
 882          * it to exec or return an error.  (the child will never
 883          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
 884          * sync pipe, causing lxc_sync_barrier_child to return
 885          * success, or return a different value, causing us to error
 886          * out).
 887          */
 888         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
 889                 return -1;
 890
 891         if (detect_shared_rootfs())
 892                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
 893
 894         if (handler->ops->post_start(handler, handler->data))
 895                 goto out_abort;
 896
 897         if (lxc_set_state(name, handler, RUNNING)) {
 898                 ERROR("failed to set state to %s",
 899                               lxc_state2str(RUNNING));
 900                 goto out_abort;
 901         }
 902
 903         lxc_cgroup_put_meta(cgroup_meta);
 904         lxc_sync_fini(handler);
 905
 906         return 0;
 907
 908 out_delete_net:
 909         if (handler->clone_flags & CLONE_NEWNET)
 910                 lxc_delete_network(handler);
 911 out_abort:
 912         lxc_cgroup_put_meta(cgroup_meta);
 913         lxc_abort(name, handler);
 914         lxc_sync_fini(handler);
 915         if (handler->pinfd >= 0) {
 916                 process_lock();
 917                 close(handler->pinfd);
 918                 process_unlock();
 919                 handler->pinfd = -1;
 920         }
 921
 922         return -1;
 923 }
 924
 925 int __lxc_start(const char *name, struct lxc_conf *conf,
 926                 struct lxc_operations* ops, void *data, const char *lxcpath)
 927 {
 928         struct lxc_handler *handler;
 929         int err = -1;
 930         int status;
 931
 932         handler = lxc_init(name, conf, lxcpath);
 933         if (!handler) {
 934                 ERROR("failed to initialize the container");
 935                 return -1;
 936         }
 937         handler->ops = ops;
 938         handler->data = data;
 939
 940         if (must_drop_cap_sys_boot(handler->conf)) {
 941                 #if HAVE_SYS_CAPABILITY_H
 942                 DEBUG("Dropping cap_sys_boot\n");
 943                 #else
 944                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported\n");
 945                 #endif
 946         } else {
 947                 DEBUG("Not dropping cap_sys_boot or watching utmp\n");
 948                 handler->conf->need_utmp_watch = 0;
 949         }
 950
 951         err = lxc_spawn(handler);
 952         if (err) {
 953                 ERROR("failed to spawn '%s'", name);
 954                 goto out_fini_nonet;
 955         }
 956
 957         err = lxc_poll(name, handler);
 958         if (err) {
 959                 ERROR("mainloop exited with an error");
 960                 goto out_abort;
 961         }
 962
 963         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 964                 continue;
 965
 966         /*
 967          * If the child process exited but was not signaled,
 968          * it didn't call reboot.  This should mean it was an
 969          * lxc-execute which simply exited.  In any case, treat
 970          * it as a 'halt'
 971          */
 972         if (WIFSIGNALED(status)) {
 973                 switch(WTERMSIG(status)) {
 974                 case SIGINT: /* halt */
 975                         DEBUG("Container halting");
 976                         break;
 977                 case SIGHUP: /* reboot */
 978                         DEBUG("Container rebooting");
 979                         handler->conf->reboot = 1;
 980                         break;
 981                 default:
 982                         DEBUG("unknown exit status for init: %d\n", WTERMSIG(status));
 983                         break;
 984                 }
 985         }
 986
 987         lxc_rename_phys_nics_on_shutdown(handler->conf);
 988
 989         if (handler->pinfd >= 0) {
 990                 process_lock();
 991                 close(handler->pinfd);
 992                 process_unlock();
 993                 handler->pinfd = -1;
 994         }
 995
 996         err =  lxc_error_set_and_log(handler->pid, status);
 997 out_fini:
 998         lxc_delete_network(handler);
 999
1000 out_fini_nonet:
1001         lxc_fini(name, handler);
1002         return err;
1003
1004 out_abort:
1005         lxc_abort(name, handler);
1006         goto out_fini;
1007 }
1008
1009 struct start_args {
1010         char *const *argv;
1011 };
1012
1013 static int start(struct lxc_handler *handler, void* data)
1014 {
1015         struct start_args *arg = data;
1016
1017         NOTICE("exec'ing '%s'", arg->argv[0]);
1018
1019         execvp(arg->argv[0], arg->argv);
1020         SYSERROR("failed to exec %s", arg->argv[0]);
1021         return 0;
1022 }
1023
1024 static int post_start(struct lxc_handler *handler, void* data)
1025 {
1026         struct start_args *arg = data;
1027
1028         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1029         return 0;
1030 }
1031
1032 static struct lxc_operations start_ops = {
1033         .start = start,
1034         .post_start = post_start
1035 };
1036
1037 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1038               const char *lxcpath)
1039 {
1040         struct start_args start_arg = {
1041                 .argv = argv,
1042         };
1043
1044         if (lxc_check_inherited(conf, -1))
1045                 return -1;
1046
1047         conf->need_utmp_watch = 1;
1048         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath);
1049 }