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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  22  */
  23
  24 #include "config.h"
  25
  26 #include <stdio.h>
  27 #undef _GNU_SOURCE
  28 #include <string.h>
  29 #include <stdlib.h>
  30 #include <dirent.h>
  31 #include <errno.h>
  32 #include <unistd.h>
  33 #include <signal.h>
  34 #include <fcntl.h>
  35 #include <termios.h>
  36 #include <sys/param.h>
  37 #include <sys/file.h>
  38 #include <sys/mount.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/socket.h>
  42 #include <sys/prctl.h>
  43 #include <sys/types.h>
  44 #include <sys/wait.h>
  45 #include <sys/un.h>
  46 #include <sys/poll.h>
  47 #include <sys/syscall.h>
  48
  49 #if HAVE_SYS_CAPABILITY_H
  50 #include <sys/capability.h>
  51 #endif
  52
  53 #ifdef HAVE_SYS_SIGNALFD_H
  54 #  include <sys/signalfd.h>
  55 #else
  56 /* assume kernel headers are too old */
  57 #include <stdint.h>
  58 struct signalfd_siginfo
  59 {
  60         uint32_t ssi_signo;
  61         int32_t ssi_errno;
  62         int32_t ssi_code;
  63         uint32_t ssi_pid;
  64         uint32_t ssi_uid;
  65         int32_t ssi_fd;
  66         uint32_t ssi_tid;
  67         uint32_t ssi_band;
  68         uint32_t ssi_overrun;
  69         uint32_t ssi_trapno;
  70         int32_t ssi_status;
  71         int32_t ssi_int;
  72         uint64_t ssi_ptr;
  73         uint64_t ssi_utime;
  74         uint64_t ssi_stime;
  75         uint64_t ssi_addr;
  76         uint8_t __pad[48];
  77 };
  78
  79 #  ifndef __NR_signalfd4
  80 /* assume kernel headers are too old */
  81 #    if __i386__
  82 #      define __NR_signalfd4 327
  83 #    elif __x86_64__
  84 #      define __NR_signalfd4 289
  85 #    elif __powerpc__
  86 #      define __NR_signalfd4 313
  87 #    elif __s390x__
  88 #      define __NR_signalfd4 322
  89 #    endif
  90 #endif
  91
  92 #  ifndef __NR_signalfd
  93 /* assume kernel headers are too old */
  94 #    if __i386__
  95 #      define __NR_signalfd 321
  96 #    elif __x86_64__
  97 #      define __NR_signalfd 282
  98 #    elif __powerpc__
  99 #      define __NR_signalfd 305
 100 #    elif __s390x__
 101 #      define __NR_signalfd 316
 102 #    endif
 103 #endif
 104
 105 int signalfd(int fd, const sigset_t *mask, int flags)
 106 {
 107         int retval;
 108
 109         retval = syscall (__NR_signalfd4, fd, mask, _NSIG / 8, flags);
 110         if (errno == ENOSYS && flags == 0)
 111                 retval = syscall (__NR_signalfd, fd, mask, _NSIG / 8);
 112         return retval;
 113 }
 114 #endif
 115
 116 #if !HAVE_DECL_PR_CAPBSET_DROP
 117 #define PR_CAPBSET_DROP 24
 118 #endif
 119
 120 #include "start.h"
 121 #include "conf.h"
 122 #include "log.h"
 123 #include "cgroup.h"
 124 #include "error.h"
 125 #include "af_unix.h"
 126 #include "mainloop.h"
 127 #include "utils.h"
 128 #include "lxcutmp.h"
 129 #include "monitor.h"
 130 #include "commands.h"
 131 #include "console.h"
 132 #include "sync.h"
 133 #include "namespace.h"
 134 #include "apparmor.h"
 135 #include "lxcseccomp.h"
 136 #include "caps.h"
 137
 138 lxc_log_define(lxc_start, lxc);
 139
 140 static int match_fd(int fd)
 141 {
 142         return (fd == 0 || fd == 1 || fd == 2);
 143 }
 144
 145 int lxc_check_inherited(struct lxc_conf *conf, int fd_to_ignore)
 146 {
 147         struct dirent dirent, *direntp;
 148         int fd, fddir;
 149         DIR *dir;
 150
 151 restart:
 152         dir = opendir("/proc/self/fd");
 153         if (!dir) {
 154                 WARN("failed to open directory: %m");
 155                 return -1;
 156         }
 157
 158         fddir = dirfd(dir);
 159
 160         while (!readdir_r(dir, &dirent, &direntp)) {
 161                 if (!direntp)
 162                         break;
 163
 164                 if (!strcmp(direntp->d_name, "."))
 165                         continue;
 166
 167                 if (!strcmp(direntp->d_name, ".."))
 168                         continue;
 169
 170                 fd = atoi(direntp->d_name);
 171
 172                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 173                         continue;
 174
 175                 if (match_fd(fd))
 176                         continue;
 177
 178                 if (conf->close_all_fds) {
 179                         close(fd);
 180                         closedir(dir);
 181                         INFO("closed inherited fd %d", fd);
 182                         goto restart;
 183                 }
 184                 WARN("inherited fd %d", fd);
 185         }
 186
 187         closedir(dir); /* cannot fail */
 188         return 0;
 189 }
 190
 191 static int setup_signal_fd(sigset_t *oldmask)
 192 {
 193         sigset_t mask;
 194         int fd;
 195
 196         /* Block everything except serious error signals */
 197         if (sigfillset(&mask) ||
 198             sigdelset(&mask, SIGILL) ||
 199             sigdelset(&mask, SIGSEGV) ||
 200             sigdelset(&mask, SIGBUS) ||
 201             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 202                 SYSERROR("failed to set signal mask");
 203                 return -1;
 204         }
 205
 206         fd = signalfd(-1, &mask, 0);
 207         if (fd < 0) {
 208                 SYSERROR("failed to create the signal fd");
 209                 return -1;
 210         }
 211
 212         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 213                 SYSERROR("failed to set sigfd to close-on-exec");
 214                 close(fd);
 215                 return -1;
 216         }
 217
 218         DEBUG("sigchild handler set");
 219
 220         return fd;
 221 }
 222
 223 static int signal_handler(int fd, void *data,
 224                            struct lxc_epoll_descr *descr)
 225 {
 226         struct signalfd_siginfo siginfo;
 227         int ret;
 228         pid_t *pid = data;
 229
 230         ret = read(fd, &siginfo, sizeof(siginfo));
 231         if (ret < 0) {
 232                 ERROR("failed to read signal info");
 233                 return -1;
 234         }
 235
 236         if (ret != sizeof(siginfo)) {
 237                 ERROR("unexpected siginfo size");
 238                 return -1;
 239         }
 240
 241         if (siginfo.ssi_signo != SIGCHLD) {
 242                 kill(*pid, siginfo.ssi_signo);
 243                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 244                 return 0;
 245         }
 246
 247         if (siginfo.ssi_code == CLD_STOPPED ||
 248             siginfo.ssi_code == CLD_CONTINUED) {
 249                 INFO("container init process was stopped/continued");
 250                 return 0;
 251         }
 252
 253         /* more robustness, protect ourself from a SIGCHLD sent
 254          * by a process different from the container init
 255          */
 256         if (siginfo.ssi_pid != *pid) {
 257                 WARN("invalid pid for SIGCHLD");
 258                 return 0;
 259         }
 260
 261         DEBUG("container init process exited");
 262         return 1;
 263 }
 264
 265 int lxc_pid_callback(int fd, struct lxc_request *request,
 266                      struct lxc_handler *handler)
 267 {
 268         struct lxc_answer answer;
 269         int ret;
 270
 271         memset(&answer, 0, sizeof(answer));
 272         answer.pid = handler->pid;
 273         answer.ret = 0;
 274
 275         ret = send(fd, &answer, sizeof(answer), 0);
 276         if (ret < 0) {
 277                 WARN("failed to send answer to the peer");
 278                 return -1;
 279         }
 280
 281         if (ret != sizeof(answer)) {
 282                 ERROR("partial answer sent");
 283                 return -1;
 284         }
 285
 286         return 0;
 287 }
 288
 289 int lxc_cgroup_callback(int fd, struct lxc_request *request,
 290                      struct lxc_handler *handler)
 291 {
 292         struct lxc_answer answer;
 293         int ret;
 294
 295         memset(&answer, 0, sizeof(answer));
 296         answer.pathlen = strlen(handler->cgroup) + 1;
 297         answer.path = handler->cgroup;
 298         answer.ret = 0;
 299
 300         ret = send(fd, &answer, sizeof(answer), 0);
 301         if (ret < 0) {
 302                 WARN("failed to send answer to the peer");
 303                 return -1;
 304         }
 305
 306         if (ret != sizeof(answer)) {
 307                 ERROR("partial answer sent");
 308                 return -1;
 309         }
 310
 311         ret = send(fd, answer.path, answer.pathlen, 0);
 312         if (ret < 0) {
 313                 WARN("failed to send answer to the peer");
 314                 return -1;
 315         }
 316
 317         if (ret != answer.pathlen) {
 318                 ERROR("partial answer sent");
 319                 return -1;
 320         }
 321
 322         return 0;
 323 }
 324
 325 int lxc_clone_flags_callback(int fd, struct lxc_request *request,
 326                              struct lxc_handler *handler)
 327 {
 328         struct lxc_answer answer;
 329         int ret;
 330
 331         memset(&answer, 0, sizeof(answer));
 332         answer.pid = 0;
 333         answer.ret = handler->clone_flags;
 334
 335         ret = send(fd, &answer, sizeof(answer), 0);
 336         if (ret < 0) {
 337                 WARN("failed to send answer to the peer");
 338                 return -1;
 339         }
 340
 341         if (ret != sizeof(answer)) {
 342                 ERROR("partial answer sent");
 343                 return -1;
 344         }
 345
 346         return 0;
 347 }
 348
 349 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 350 {
 351         handler->state = state;
 352         lxc_monitor_send_state(name, state, handler->lxcpath);
 353         return 0;
 354 }
 355
 356 int lxc_poll(const char *name, struct lxc_handler *handler)
 357 {
 358         int sigfd = handler->sigfd;
 359         int pid = handler->pid;
 360         struct lxc_epoll_descr descr;
 361
 362         if (lxc_mainloop_open(&descr)) {
 363                 ERROR("failed to create mainloop");
 364                 goto out_sigfd;
 365         }
 366
 367         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 368                 ERROR("failed to add handler for the signal");
 369                 goto out_mainloop_open;
 370         }
 371
 372         if (lxc_console_mainloop_add(&descr, handler)) {
 373                 ERROR("failed to add console handler to mainloop");
 374                 goto out_mainloop_open;
 375         }
 376
 377         if (lxc_command_mainloop_add(name, &descr, handler)) {
 378                 ERROR("failed to add command handler to mainloop");
 379                 goto out_mainloop_open;
 380         }
 381
 382         if (handler->conf->need_utmp_watch) {
 383                 #if HAVE_SYS_CAPABILITY_H
 384                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 385                         ERROR("failed to add utmp handler to mainloop");
 386                         goto out_mainloop_open;
 387                 }
 388                 #else
 389                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support\n");
 390                 #endif
 391         }
 392
 393         return lxc_mainloop(&descr);
 394
 395 out_mainloop_open:
 396         lxc_mainloop_close(&descr);
 397 out_sigfd:
 398         close(sigfd);
 399         return -1;
 400 }
 401
 402 extern int lxc_caps_check(void);
 403
 404 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 405 {
 406         struct lxc_handler *handler;
 407
 408         if (!lxc_caps_check()) {
 409                 ERROR("Not running with sufficient privilege");
 410                 return NULL;
 411         }
 412
 413         handler = malloc(sizeof(*handler));
 414         if (!handler)
 415                 return NULL;
 416
 417         memset(handler, 0, sizeof(*handler));
 418
 419         handler->conf = conf;
 420         handler->lxcpath = lxcpath;
 421
 422         apparmor_handler_init(handler);
 423         handler->name = strdup(name);
 424         if (!handler->name) {
 425                 ERROR("failed to allocate memory");
 426                 goto out_free;
 427         }
 428
 429         if (lxc_command_init(name, handler, lxcpath))
 430                 goto out_free_name;
 431
 432         if (lxc_read_seccomp_config(conf) != 0) {
 433                 ERROR("failed loading seccomp policy");
 434                 goto out_close_maincmd_fd;
 435         }
 436
 437         /* Begin the set the state to STARTING*/
 438         if (lxc_set_state(name, handler, STARTING)) {
 439                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 440                 goto out_free_name;
 441         }
 442
 443         /* Start of environment variable setup for hooks */
 444         if (setenv("LXC_NAME", name, 1)) {
 445                 SYSERROR("failed to set environment variable for container name");
 446         }
 447         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 448                 SYSERROR("failed to set environment variable for config path");
 449         }
 450         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 451                 SYSERROR("failed to set environment variable for rootfs mount");
 452         }
 453         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 454                 SYSERROR("failed to set environment variable for rootfs mount");
 455         }
 456         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 457                 SYSERROR("failed to set environment variable for console path");
 458         }
 459         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 460                 SYSERROR("failed to set environment variable for console log");
 461         }
 462         /* End of environment variable setup for hooks */
 463
 464         if (run_lxc_hooks(name, "pre-start", conf)) {
 465                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 466                 goto out_aborting;
 467         }
 468
 469         if (lxc_create_tty(name, conf)) {
 470                 ERROR("failed to create the ttys");
 471                 goto out_aborting;
 472         }
 473
 474         if (lxc_create_console(conf)) {
 475                 ERROR("failed to create console");
 476                 goto out_delete_tty;
 477         }
 478
 479         /* the signal fd has to be created before forking otherwise
 480          * if the child process exits before we setup the signal fd,
 481          * the event will be lost and the command will be stuck */
 482         handler->sigfd = setup_signal_fd(&handler->oldmask);
 483         if (handler->sigfd < 0) {
 484                 ERROR("failed to set sigchild fd handler");
 485                 goto out_delete_console;
 486         }
 487
 488         INFO("'%s' is initialized", name);
 489         return handler;
 490
 491 out_delete_console:
 492         lxc_delete_console(&conf->console);
 493 out_delete_tty:
 494         lxc_delete_tty(&conf->tty_info);
 495 out_aborting:
 496         lxc_set_state(name, handler, ABORTING);
 497 out_close_maincmd_fd:
 498         close(conf->maincmd_fd);
 499         conf->maincmd_fd = -1;
 500 out_free_name:
 501         free(handler->name);
 502         handler->name = NULL;
 503 out_free:
 504         free(handler);
 505         return NULL;
 506 }
 507
 508 static void lxc_fini(const char *name, struct lxc_handler *handler)
 509 {
 510         /* The STOPPING state is there for future cleanup code
 511          * which can take awhile
 512          */
 513         lxc_set_state(name, handler, STOPPING);
 514         lxc_set_state(name, handler, STOPPED);
 515
 516         if (run_lxc_hooks(name, "post-stop", handler->conf))
 517                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 518
 519         /* reset mask set by setup_signal_fd */
 520         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 521                 WARN("failed to restore sigprocmask");
 522
 523         lxc_delete_console(&handler->conf->console);
 524         lxc_delete_tty(&handler->conf->tty_info);
 525         close(handler->conf->maincmd_fd);
 526         handler->conf->maincmd_fd = -1;
 527         free(handler->name);
 528         if (handler->cgroup) {
 529                 lxc_cgroup_destroy(handler->cgroup);
 530                 free(handler->cgroup);
 531                 handler->cgroup = NULL;
 532         }
 533         free(handler);
 534 }
 535
 536 void lxc_abort(const char *name, struct lxc_handler *handler)
 537 {
 538         lxc_set_state(name, handler, ABORTING);
 539         if (handler->pid > 0)
 540                 kill(handler->pid, SIGKILL);
 541 }
 542
 543 #include <sys/reboot.h>
 544 #include <linux/reboot.h>
 545
 546 /*
 547  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 548  * in a child pid namespace if container reboot support exists.
 549  * Otherwise, it will either succeed or return -EPERM.
 550  */
 551 static int container_reboot_supported(void *arg)
 552 {
 553         int *cmd = arg;
 554         int ret;
 555
 556         ret = reboot(*cmd);
 557         if (ret == -1 && errno == EINVAL)
 558                 return 1;
 559         return 0;
 560 }
 561
 562 static int must_drop_cap_sys_boot(void)
 563 {
 564         FILE *f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 565         int ret, cmd, v;
 566         long stack_size = 4096;
 567         void *stack = alloca(stack_size);
 568         int status;
 569         pid_t pid;
 570
 571         if (!f) {
 572                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 573                 return 1;
 574         }
 575
 576         ret = fscanf(f, "%d", &v);
 577         fclose(f);
 578         if (ret != 1) {
 579                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 580                 return 1;
 581         }
 582         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 583
 584 #ifdef __ia64__
 585         pid = __clone2(container_reboot_supported, stack, stack_size, CLONE_NEWPID | SIGCHLD, &cmd);
 586 #else
 587         stack += stack_size;
 588         pid = clone(container_reboot_supported, stack, CLONE_NEWPID | SIGCHLD, &cmd);
 589 #endif
 590         if (pid < 0) {
 591                 SYSERROR("failed to clone\n");
 592                 return -1;
 593         }
 594         if (wait(&status) < 0) {
 595                 SYSERROR("unexpected wait error: %m\n");
 596                 return -1;
 597         }
 598
 599         if (WEXITSTATUS(status) != 1)
 600                 return 1;
 601
 602         return 0;
 603 }
 604
 605 static int do_start(void *data)
 606 {
 607         struct lxc_handler *handler = data;
 608
 609         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 610                 SYSERROR("failed to set sigprocmask");
 611                 return -1;
 612         }
 613
 614         /* This prctl must be before the synchro, so if the parent
 615          * dies before we set the parent death signal, we will detect
 616          * its death with the synchro right after, otherwise we have
 617          * a window where the parent can exit before we set the pdeath
 618          * signal leading to a unsupervized container.
 619          */
 620         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 621                 SYSERROR("failed to set pdeath signal");
 622                 return -1;
 623         }
 624
 625         lxc_sync_fini_parent(handler);
 626
 627         /* don't leak the pinfd to the container */
 628         if (handler->pinfd >= 0)
 629                 close(handler->pinfd);
 630
 631         /* Tell the parent task it can begin to configure the
 632          * container and wait for it to finish
 633          */
 634         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 635                 return -1;
 636
 637         /*
 638          * if we are in a new user namespace, become root there to have
 639          * privilege over our namespace
 640          */
 641         if (!lxc_list_empty(&handler->conf->id_map)) {
 642                 NOTICE("switching to gid/uid 0 in new user namespace");
 643                 if (setgid(0)) {
 644                         SYSERROR("setgid");
 645                         goto out_warn_father;
 646                 }
 647                 if (setuid(0)) {
 648                         SYSERROR("setuid");
 649                         goto out_warn_father;
 650                 }
 651         }
 652
 653         #if HAVE_SYS_CAPABILITY_H
 654         if (handler->conf->need_utmp_watch) {
 655                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 656                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 657                         goto out_warn_father;
 658                 }
 659                 DEBUG("Dropped cap_sys_boot\n");
 660         }
 661         #endif
 662
 663         /* Setup the container, ip, names, utsname, ... */
 664         if (lxc_setup(handler->name, handler->conf)) {
 665                 ERROR("failed to setup the container");
 666                 goto out_warn_father;
 667         }
 668
 669         /* ask father to setup cgroups and wait for him to finish */
 670         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 671                 return -1;
 672
 673         if (apparmor_load(handler) < 0)
 674                 goto out_warn_father;
 675
 676         if (lxc_seccomp_load(handler->conf) != 0)
 677                 goto out_warn_father;
 678
 679         if (run_lxc_hooks(handler->name, "start", handler->conf)) {
 680                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 681                 goto out_warn_father;
 682         }
 683
 684         /* The clearenv() and putenv() calls have been moved here
 685          * to allow us to use enviroment variables passed to the various
 686          * hooks, such as the start hook above.  Not all of the
 687          * variables like CONFIG_PATH or ROOTFS are valid in this
 688          * context but others are. */
 689         if (clearenv()) {
 690                 SYSERROR("failed to clear environment");
 691                 /* don't error out though */
 692         }
 693
 694         if (putenv("container=lxc")) {
 695                 SYSERROR("failed to set environment variable");
 696                 goto out_warn_father;
 697         }
 698
 699         close(handler->sigfd);
 700
 701         /* after this call, we are in error because this
 702          * ops should not return as it execs */
 703         handler->ops->start(handler, handler->data);
 704
 705 out_warn_father:
 706         /* we want the parent to know something went wrong, so any
 707          * value other than what it expects is ok. */
 708         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 709         return -1;
 710 }
 711
 712 int save_phys_nics(struct lxc_conf *conf)
 713 {
 714         struct lxc_list *iterator;
 715
 716         lxc_list_for_each(iterator, &conf->network) {
 717                 struct lxc_netdev *netdev = iterator->elem;
 718
 719                 if (netdev->type != LXC_NET_PHYS)
 720                         continue;
 721                 conf->saved_nics = realloc(conf->saved_nics,
 722                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 723                 if (!conf->saved_nics) {
 724                         SYSERROR("failed to allocate memory");
 725                         return -1;
 726                 }
 727                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 728                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 729                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 730                         SYSERROR("failed to allocate memory");
 731                         return -1;
 732                 }
 733                 INFO("stored saved_nic #%d idx %d name %s\n", conf->num_savednics,
 734                         conf->saved_nics[conf->num_savednics].ifindex,
 735                         conf->saved_nics[conf->num_savednics].orig_name);
 736                 conf->num_savednics++;
 737         }
 738
 739         return 0;
 740 }
 741
 742
 743 int lxc_spawn(struct lxc_handler *handler)
 744 {
 745         int failed_before_rename = 0;
 746         const char *name = handler->name;
 747
 748         if (lxc_sync_init(handler))
 749                 return -1;
 750
 751         handler->clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 752         if (!lxc_list_empty(&handler->conf->id_map)) {
 753                 INFO("Cloning a new user namespace");
 754                 handler->clone_flags |= CLONE_NEWUSER;
 755         }
 756         if (!lxc_list_empty(&handler->conf->network)) {
 757
 758                 handler->clone_flags |= CLONE_NEWNET;
 759
 760                 /* Find gateway addresses from the link device, which is
 761                  * no longer accessible inside the container. Do this
 762                  * before creating network interfaces, since goto
 763                  * out_delete_net does not work before lxc_clone. */
 764                 if (lxc_find_gateway_addresses(handler)) {
 765                         ERROR("failed to find gateway addresses");
 766                         lxc_sync_fini(handler);
 767                         return -1;
 768                 }
 769
 770                 /* that should be done before the clone because we will
 771                  * fill the netdev index and use them in the child
 772                  */
 773                 if (lxc_create_network(handler)) {
 774                         ERROR("failed to create the network");
 775                         lxc_sync_fini(handler);
 776                         return -1;
 777                 }
 778         }
 779
 780         if (save_phys_nics(handler->conf)) {
 781                 ERROR("failed to save physical nic info");
 782                 goto out_abort;
 783         }
 784
 785         /*
 786          * if the rootfs is not a blockdev, prevent the container from
 787          * marking it readonly.
 788          */
 789
 790         handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 791         if (handler->pinfd == -1) {
 792                 ERROR("failed to pin the container's rootfs");
 793                 goto out_delete_net;
 794         }
 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         lxc_sync_fini_child(handler);
 804
 805         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 806                 failed_before_rename = 1;
 807
 808         /* TODO - pass lxc.cgroup.dir (or user's pam cgroup) in for first argument */
 809         if ((handler->cgroup = lxc_cgroup_path_create(NULL, name)) == NULL)
 810                 goto out_delete_net;
 811
 812         if (lxc_cgroup_enter(handler->cgroup, handler->pid) < 0)
 813                 goto out_delete_net;
 814
 815         if (failed_before_rename)
 816                 goto out_delete_net;
 817
 818         /* Create the network configuration */
 819         if (handler->clone_flags & CLONE_NEWNET) {
 820                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 821                         ERROR("failed to create the configured network");
 822                         goto out_delete_net;
 823                 }
 824         }
 825
 826         /* map the container uids - the container became an invalid
 827          * userid the moment it was cloned with CLONE_NEWUSER - this
 828          * call doesn't change anything immediately, but allows the
 829          * container to setuid(0) (0 being mapped to something else on
 830          * the host) later to become a valid uid again */
 831         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
 832                 ERROR("failed to set up id mapping");
 833                 goto out_delete_net;
 834         }
 835
 836         /* Tell the child to continue its initialization.  we'll get
 837          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
 838          */
 839         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 840                 goto out_delete_net;
 841
 842         if (setup_cgroup(handler->cgroup, &handler->conf->cgroup)) {
 843                 ERROR("failed to setup the cgroups for '%s'", name);
 844                 goto out_delete_net;
 845         }
 846
 847
 848         /* Tell the child to complete its initialization and wait for
 849          * it to exec or return an error.  (the child will never
 850          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
 851          * sync pipe, causing lxc_sync_barrier_child to return
 852          * success, or return a different value, causing us to error
 853          * out).
 854          */
 855         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
 856                 return -1;
 857
 858         if (detect_shared_rootfs())
 859                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
 860
 861         /* If child is in a fresh user namespace, chown his ptys for
 862          * it */
 863         if (uid_shift_ttys(handler->pid, handler->conf))
 864                 DEBUG("Failed to chown ptys.\n");
 865
 866         if (handler->ops->post_start(handler, handler->data))
 867                 goto out_abort;
 868
 869         if (lxc_set_state(name, handler, RUNNING)) {
 870                 ERROR("failed to set state to %s",
 871                               lxc_state2str(RUNNING));
 872                 goto out_abort;
 873         }
 874
 875         lxc_sync_fini(handler);
 876
 877         if (handler->pinfd >= 0)
 878                 close(handler->pinfd);
 879
 880         return 0;
 881
 882 out_delete_net:
 883         if (handler->clone_flags & CLONE_NEWNET)
 884                 lxc_delete_network(handler);
 885 out_abort:
 886         lxc_abort(name, handler);
 887         lxc_sync_fini(handler);
 888         return -1;
 889 }
 890
 891 int __lxc_start(const char *name, struct lxc_conf *conf,
 892                 struct lxc_operations* ops, void *data, const char *lxcpath)
 893 {
 894         struct lxc_handler *handler;
 895         int err = -1;
 896         int status;
 897
 898         handler = lxc_init(name, conf, lxcpath);
 899         if (!handler) {
 900                 ERROR("failed to initialize the container");
 901                 return -1;
 902         }
 903         handler->ops = ops;
 904         handler->data = data;
 905
 906         if (must_drop_cap_sys_boot()) {
 907                 #if HAVE_SYS_CAPABILITY_H
 908                 DEBUG("Dropping cap_sys_boot\n");
 909                 #else
 910                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported\n");
 911                 #endif
 912         } else {
 913                 DEBUG("Not dropping cap_sys_boot or watching utmp\n");
 914                 handler->conf->need_utmp_watch = 0;
 915         }
 916
 917         err = lxc_spawn(handler);
 918         if (err) {
 919                 ERROR("failed to spawn '%s'", name);
 920                 goto out_fini_nonet;
 921         }
 922
 923         err = lxc_poll(name, handler);
 924         if (err) {
 925                 ERROR("mainloop exited with an error");
 926                 goto out_abort;
 927         }
 928
 929         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 930                 continue;
 931
 932         /*
 933          * If the child process exited but was not signaled,
 934          * it didn't call reboot.  This should mean it was an
 935          * lxc-execute which simply exited.  In any case, treat
 936          * it as a 'halt'
 937          */
 938         if (WIFSIGNALED(status)) {
 939                 switch(WTERMSIG(status)) {
 940                 case SIGINT: /* halt */
 941                         DEBUG("Container halting");
 942                         break;
 943                 case SIGHUP: /* reboot */
 944                         DEBUG("Container rebooting");
 945                         handler->conf->reboot = 1;
 946                         break;
 947                 default:
 948                         DEBUG("unknown exit status for init: %d\n", WTERMSIG(status));
 949                         break;
 950                 }
 951         }
 952
 953         lxc_rename_phys_nics_on_shutdown(handler->conf);
 954
 955         err =  lxc_error_set_and_log(handler->pid, status);
 956 out_fini:
 957         lxc_delete_network(handler);
 958
 959 out_fini_nonet:
 960         lxc_fini(name, handler);
 961         return err;
 962
 963 out_abort:
 964         lxc_abort(name, handler);
 965         goto out_fini;
 966 }
 967
 968 struct start_args {
 969         char *const *argv;
 970 };
 971
 972 static int start(struct lxc_handler *handler, void* data)
 973 {
 974         struct start_args *arg = data;
 975
 976         NOTICE("exec'ing '%s'", arg->argv[0]);
 977
 978         execvp(arg->argv[0], arg->argv);
 979         SYSERROR("failed to exec %s", arg->argv[0]);
 980         return 0;
 981 }
 982
 983 static int post_start(struct lxc_handler *handler, void* data)
 984 {
 985         struct start_args *arg = data;
 986
 987         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
 988         return 0;
 989 }
 990
 991 static struct lxc_operations start_ops = {
 992         .start = start,
 993         .post_start = post_start
 994 };
 995
 996 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
 997               const char *lxcpath)
 998 {
 999         struct start_args start_arg = {
1000                 .argv = argv,
1001         };
1002
1003         if (lxc_check_inherited(conf, -1))
1004                 return -1;
1005
1006         conf->need_utmp_watch = 1;
1007         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath);
1008 }