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 #define _GNU_SOURCE
  25 #include "config.h"
  26
  27 #include <alloca.h>
  28 #include <dirent.h>
  29 #include <errno.h>
  30 #include <fcntl.h>
  31 #include <grp.h>
  32 #include <poll.h>
  33 #include <signal.h>
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <string.h>
  37 #include <unistd.h>
  38 #include <sys/file.h>
  39 #include <sys/mount.h>
  40 #include <sys/param.h>
  41 #include <sys/prctl.h>
  42 #include <sys/socket.h>
  43 #include <sys/stat.h>
  44 #include <sys/syscall.h>
  45 #include <sys/types.h>
  46 #include <sys/un.h>
  47 #include <sys/wait.h>
  48
  49 #if HAVE_SYS_CAPABILITY_H
  50 #include <sys/capability.h>
  51 #endif
  52
  53 #ifndef HAVE_DECL_PR_CAPBSET_DROP
  54 #define PR_CAPBSET_DROP 24
  55 #endif
  56
  57 #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS
  58 #define PR_SET_NO_NEW_PRIVS 38
  59 #endif
  60
  61 #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS
  62 #define PR_GET_NO_NEW_PRIVS 39
  63 #endif
  64
  65 #include "af_unix.h"
  66 #include "bdev.h"
  67 #include "caps.h"
  68 #include "cgroup.h"
  69 #include "commands.h"
  70 #include "conf.h"
  71 #include "console.h"
  72 #include "error.h"
  73 #include "log.h"
  74 #include "lxclock.h"
  75 #include "lxcseccomp.h"
  76 #include "lxcutmp.h"
  77 #include "mainloop.h"
  78 #include "monitor.h"
  79 #include "namespace.h"
  80 #include "start.h"
  81 #include "sync.h"
  82 #include "utils.h"
  83 #include "lsm/lsm.h"
  84
  85 lxc_log_define(lxc_start, lxc);
  86
  87 const struct ns_info ns_info[LXC_NS_MAX] = {
  88         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  89         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  90         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  91         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  92         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  93         [LXC_NS_NET] = {"net", CLONE_NEWNET},
  94         [LXC_NS_CGROUP] = {"cgroup", CLONE_NEWCGROUP}
  95 };
  96
  97 extern void mod_all_rdeps(struct lxc_container *c, bool inc);
  98 static bool do_destroy_container(struct lxc_conf *conf);
  99 static int lxc_rmdir_onedev_wrapper(void *data);
 100 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
 101                                             const char *name);
 102
 103 static void print_top_failing_dir(const char *path)
 104 {
 105         size_t len = strlen(path);
 106         char *copy = alloca(len + 1), *p, *e, saved;
 107         strcpy(copy, path);
 108
 109         p = copy;
 110         e = copy + len;
 111         while (p < e) {
 112                 while (p < e && *p == '/')
 113                         p++;
 114                 while (p < e && *p != '/')
 115                         p++;
 116                 saved = *p;
 117                 *p = '\0';
 118                 if (access(copy, X_OK)) {
 119                         SYSERROR("could not access %s.  Please grant it 'x' " \
 120                               "access, or add an ACL for the container root.",
 121                               copy);
 122                         return;
 123                 }
 124                 *p = saved;
 125         }
 126 }
 127
 128 static void close_ns(int ns_fd[LXC_NS_MAX]) {
 129         int i;
 130
 131         for (i = 0; i < LXC_NS_MAX; i++) {
 132                 if (ns_fd[i] > -1) {
 133                         close(ns_fd[i]);
 134                         ns_fd[i] = -1;
 135                 }
 136         }
 137 }
 138
 139 /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
 140  * in clone_flags.
 141  * Return true on success, false on failure.
 142  */
 143 static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid)
 144 {
 145         int i, ret;
 146
 147         for (i = 0; i < LXC_NS_MAX; i++)
 148                 ns_fd[i] = -1;
 149
 150         ret = lxc_preserve_ns(pid, "");
 151         if (ret < 0) {
 152                 SYSERROR("Kernel does not support attaching to namespaces.");
 153                 return false;
 154         } else {
 155                 close(ret);
 156         }
 157
 158         for (i = 0; i < LXC_NS_MAX; i++) {
 159                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 160                         continue;
 161                 ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name);
 162                 if (ns_fd[i] < 0)
 163                         goto error;
 164         }
 165
 166         return true;
 167
 168 error:
 169         if (errno == ENOENT)
 170                 SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name);
 171         else
 172                 SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno));
 173         close_ns(ns_fd);
 174         return false;
 175 }
 176
 177 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 178         int i;
 179
 180         for (i = 0; i < LXC_NS_MAX; i++) {
 181                 if (ns_fd[i] < 0)
 182                         continue;
 183
 184                 if (setns(ns_fd[i], 0) != 0)
 185                         goto error;
 186         }
 187         return 0;
 188
 189 error:
 190         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 191         return -1;
 192 }
 193
 194 static int match_fd(int fd)
 195 {
 196         return (fd == 0 || fd == 1 || fd == 2);
 197 }
 198
 199 /*
 200  * Check for any fds we need to close
 201  * * if fd_to_ignore != -1, then if we find that fd open we will ignore it.
 202  * * By default we warn about open fds we find.
 203  * * If closeall is true, we will close open fds.
 204  * * If lxc-start was passed "-C", then conf->close_all_fds will be true,
 205  *     in which case we also close all open fds.
 206  * * A daemonized container will always pass closeall=true.
 207  */
 208 int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore)
 209 {
 210         struct dirent *direntp;
 211         int fd, fddir;
 212         DIR *dir;
 213
 214         if (conf && conf->close_all_fds)
 215                 closeall = true;
 216
 217 restart:
 218         dir = opendir("/proc/self/fd");
 219         if (!dir) {
 220                 WARN("failed to open directory: %m");
 221                 return -1;
 222         }
 223
 224         fddir = dirfd(dir);
 225
 226         while ((direntp = readdir(dir))) {
 227                 if (!direntp)
 228                         break;
 229
 230                 if (!strcmp(direntp->d_name, "."))
 231                         continue;
 232
 233                 if (!strcmp(direntp->d_name, ".."))
 234                         continue;
 235
 236                 fd = atoi(direntp->d_name);
 237
 238                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 239                         continue;
 240
 241                 if (current_config && fd == current_config->logfd)
 242                         continue;
 243
 244                 if (match_fd(fd))
 245                         continue;
 246
 247                 if (closeall) {
 248                         close(fd);
 249                         closedir(dir);
 250                         INFO("closed inherited fd %d", fd);
 251                         goto restart;
 252                 }
 253                 WARN("inherited fd %d", fd);
 254         }
 255
 256         /*
 257          * only enable syslog at this point to avoid the above logging function
 258          * to open a new fd and make the check_inherited function enter an
 259          * infinite loop.
 260          */
 261         lxc_log_enable_syslog();
 262
 263         closedir(dir); /* cannot fail */
 264         return 0;
 265 }
 266
 267 static int setup_signal_fd(sigset_t *oldmask)
 268 {
 269         sigset_t mask;
 270         int fd;
 271
 272         /* Block everything except serious error signals */
 273         if (sigfillset(&mask) ||
 274             sigdelset(&mask, SIGILL) ||
 275             sigdelset(&mask, SIGSEGV) ||
 276             sigdelset(&mask, SIGBUS) ||
 277             sigdelset(&mask, SIGWINCH) ||
 278             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 279                 SYSERROR("failed to set signal mask");
 280                 return -1;
 281         }
 282
 283         fd = signalfd(-1, &mask, 0);
 284         if (fd < 0) {
 285                 SYSERROR("failed to create the signal fd");
 286                 return -1;
 287         }
 288
 289         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 290                 SYSERROR("failed to set sigfd to close-on-exec");
 291                 close(fd);
 292                 return -1;
 293         }
 294
 295         DEBUG("sigchild handler set");
 296
 297         return fd;
 298 }
 299
 300 static int signal_handler(int fd, uint32_t events, void *data,
 301                           struct lxc_epoll_descr *descr)
 302 {
 303         struct signalfd_siginfo siginfo;
 304         siginfo_t info;
 305         int ret;
 306         pid_t *pid = data;
 307         bool init_died = false;
 308
 309         ret = read(fd, &siginfo, sizeof(siginfo));
 310         if (ret < 0) {
 311                 ERROR("failed to read signal info");
 312                 return -1;
 313         }
 314
 315         if (ret != sizeof(siginfo)) {
 316                 ERROR("unexpected siginfo size");
 317                 return -1;
 318         }
 319
 320         // check whether init is running
 321         info.si_pid = 0;
 322         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 323         if (ret == 0 && info.si_pid == *pid) {
 324                 init_died = true;
 325         }
 326
 327         if (siginfo.ssi_signo != SIGCHLD) {
 328                 kill(*pid, siginfo.ssi_signo);
 329                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 330                 return init_died ? 1 : 0;
 331         }
 332
 333         if (siginfo.ssi_code == CLD_STOPPED ||
 334             siginfo.ssi_code == CLD_CONTINUED) {
 335                 INFO("container init process was stopped/continued");
 336                 return init_died ? 1 : 0;
 337         }
 338
 339         /* more robustness, protect ourself from a SIGCHLD sent
 340          * by a process different from the container init
 341          */
 342         if (siginfo.ssi_pid != *pid) {
 343                 WARN("invalid pid for SIGCHLD");
 344                 return init_died ? 1 : 0;
 345         }
 346
 347         DEBUG("container init process exited");
 348         return 1;
 349 }
 350
 351 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 352 {
 353         handler->state = state;
 354         lxc_monitor_send_state(name, state, handler->lxcpath);
 355         return 0;
 356 }
 357
 358 int lxc_poll(const char *name, struct lxc_handler *handler)
 359 {
 360         int sigfd = handler->sigfd;
 361         int pid = handler->pid;
 362         struct lxc_epoll_descr descr;
 363
 364         if (lxc_mainloop_open(&descr)) {
 365                 ERROR("failed to create mainloop");
 366                 goto out_sigfd;
 367         }
 368
 369         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 370                 ERROR("failed to add handler for the signal");
 371                 goto out_mainloop_open;
 372         }
 373
 374         if (lxc_console_mainloop_add(&descr, handler->conf)) {
 375                 ERROR("failed to add console handler to mainloop");
 376                 goto out_mainloop_open;
 377         }
 378
 379         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 380                 ERROR("failed to add command handler to mainloop");
 381                 goto out_mainloop_open;
 382         }
 383
 384         if (handler->conf->need_utmp_watch) {
 385                 #if HAVE_SYS_CAPABILITY_H
 386                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 387                         ERROR("failed to add utmp handler to mainloop");
 388                         goto out_mainloop_open;
 389                 }
 390                 #else
 391                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
 392                 #endif
 393         }
 394
 395         return lxc_mainloop(&descr, -1);
 396
 397 out_mainloop_open:
 398         lxc_mainloop_close(&descr);
 399 out_sigfd:
 400         close(sigfd);
 401         return -1;
 402 }
 403
 404 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 405 {
 406         int i;
 407         struct lxc_handler *handler;
 408
 409         handler = malloc(sizeof(*handler));
 410         if (!handler)
 411                 return NULL;
 412
 413         memset(handler, 0, sizeof(*handler));
 414
 415         handler->ttysock[0] = handler->ttysock[1] = -1;
 416         handler->conf = conf;
 417         handler->lxcpath = lxcpath;
 418         handler->pinfd = -1;
 419
 420         for (i = 0; i < LXC_NS_MAX; i++)
 421                 handler->nsfd[i] = -1;
 422
 423         lsm_init();
 424
 425         handler->name = strdup(name);
 426         if (!handler->name) {
 427                 ERROR("failed to allocate memory");
 428                 goto out_free;
 429         }
 430
 431         if (lxc_cmd_init(name, handler, lxcpath))
 432                 goto out_free_name;
 433
 434         if (lxc_read_seccomp_config(conf) != 0) {
 435                 ERROR("failed loading seccomp policy");
 436                 goto out_close_maincmd_fd;
 437         }
 438
 439         /* Begin by setting the state to STARTING */
 440         if (lxc_set_state(name, handler, STARTING)) {
 441                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 442                 goto out_close_maincmd_fd;
 443         }
 444
 445         /* Start of environment variable setup for hooks */
 446         if (name && setenv("LXC_NAME", name, 1)) {
 447                 SYSERROR("failed to set environment variable for container name");
 448         }
 449         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 450                 SYSERROR("failed to set environment variable for config path");
 451         }
 452         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 453                 SYSERROR("failed to set environment variable for rootfs mount");
 454         }
 455         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 456                 SYSERROR("failed to set environment variable for rootfs mount");
 457         }
 458         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 459                 SYSERROR("failed to set environment variable for console path");
 460         }
 461         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 462                 SYSERROR("failed to set environment variable for console log");
 463         }
 464         if (setenv("LXC_CGNS_AWARE", "1", 1)) {
 465                 SYSERROR("failed to set LXC_CGNS_AWARE environment variable");
 466         }
 467         /* End of environment variable setup for hooks */
 468
 469         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 470                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 471                 goto out_aborting;
 472         }
 473
 474         /* the signal fd has to be created before forking otherwise
 475          * if the child process exits before we setup the signal fd,
 476          * the event will be lost and the command will be stuck */
 477         handler->sigfd = setup_signal_fd(&handler->oldmask);
 478         if (handler->sigfd < 0) {
 479                 ERROR("failed to set sigchild fd handler");
 480                 goto out_delete_tty;
 481         }
 482
 483         /* do this after setting up signals since it might unblock SIGWINCH */
 484         if (lxc_console_create(conf)) {
 485                 ERROR("failed to create console");
 486                 goto out_restore_sigmask;
 487         }
 488
 489         if (ttys_shift_ids(conf) < 0) {
 490                 ERROR("Failed to shift tty into container");
 491                 goto out_restore_sigmask;
 492         }
 493
 494         INFO("'%s' is initialized", name);
 495         return handler;
 496
 497 out_restore_sigmask:
 498         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 499 out_delete_tty:
 500         lxc_delete_tty(&conf->tty_info);
 501 out_aborting:
 502         lxc_set_state(name, handler, ABORTING);
 503 out_close_maincmd_fd:
 504         close(conf->maincmd_fd);
 505         conf->maincmd_fd = -1;
 506 out_free_name:
 507         free(handler->name);
 508         handler->name = NULL;
 509 out_free:
 510         free(handler);
 511         return NULL;
 512 }
 513
 514 void lxc_fini(const char *name, struct lxc_handler *handler)
 515 {
 516         int i, rc;
 517         pid_t self = getpid();
 518         char *namespaces[LXC_NS_MAX+1];
 519         size_t namespace_count = 0;
 520
 521         /* The STOPPING state is there for future cleanup code
 522          * which can take awhile
 523          */
 524         lxc_set_state(name, handler, STOPPING);
 525
 526         for (i = 0; i < LXC_NS_MAX; i++) {
 527                 if (handler->nsfd[i] != -1) {
 528                         rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d",
 529                                       ns_info[i].proc_name, self, handler->nsfd[i]);
 530                         if (rc == -1) {
 531                                 SYSERROR("failed to allocate memory");
 532                                 break;
 533                         }
 534                         ++namespace_count;
 535                 }
 536         }
 537         namespaces[namespace_count] = NULL;
 538
 539         if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) {
 540                 SYSERROR("failed to set environment variable for stop target");
 541         }
 542         if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) {
 543                 SYSERROR("failed to set environment variable for stop target");
 544         }
 545
 546         if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces))
 547                 ERROR("failed to run stop hooks for container '%s'.", name);
 548
 549         while (namespace_count--)
 550                 free(namespaces[namespace_count]);
 551         for (i = 0; i < LXC_NS_MAX; i++) {
 552                 if (handler->nsfd[i] != -1) {
 553                         close(handler->nsfd[i]);
 554                         handler->nsfd[i] = -1;
 555                 }
 556         }
 557
 558         if (handler->netnsfd >= 0) {
 559                 close(handler->netnsfd);
 560                 handler->netnsfd = -1;
 561         }
 562
 563         lxc_set_state(name, handler, STOPPED);
 564
 565         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) {
 566                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 567                 if (handler->conf->reboot) {
 568                         WARN("Container will be stopped instead of rebooted.");
 569                         handler->conf->reboot = 0;
 570                         setenv("LXC_TARGET", "stop", 1);
 571                 }
 572         }
 573
 574         /* reset mask set by setup_signal_fd */
 575         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 576                 WARN("failed to restore sigprocmask");
 577
 578         lxc_console_delete(&handler->conf->console);
 579         lxc_delete_tty(&handler->conf->tty_info);
 580         close(handler->conf->maincmd_fd);
 581         handler->conf->maincmd_fd = -1;
 582         free(handler->name);
 583         if (handler->ttysock[0] != -1) {
 584                 close(handler->ttysock[0]);
 585                 close(handler->ttysock[1]);
 586         }
 587
 588         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1)
 589                 lxc_destroy_container_on_signal(handler, name);
 590
 591         cgroup_destroy(handler);
 592         free(handler);
 593 }
 594
 595 void lxc_abort(const char *name, struct lxc_handler *handler)
 596 {
 597         int ret, status;
 598
 599         lxc_set_state(name, handler, ABORTING);
 600         if (handler->pid > 0)
 601                 kill(handler->pid, SIGKILL);
 602         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 603 }
 604
 605 #include <sys/reboot.h>
 606 #include <linux/reboot.h>
 607
 608 /*
 609  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 610  * in a child pid namespace if container reboot support exists.
 611  * Otherwise, it will either succeed or return -EPERM.
 612  */
 613 static int container_reboot_supported(void *arg)
 614 {
 615         int *cmd = arg;
 616         int ret;
 617
 618         ret = reboot(*cmd);
 619         if (ret == -1 && errno == EINVAL)
 620                 return 1;
 621         return 0;
 622 }
 623
 624 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 625 {
 626         FILE *f;
 627         int ret, cmd, v, flags;
 628         long stack_size = 4096;
 629         void *stack = alloca(stack_size);
 630         int status;
 631         pid_t pid;
 632
 633         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 634         if (!f) {
 635                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 636                 return 1;
 637         }
 638
 639         ret = fscanf(f, "%d", &v);
 640         fclose(f);
 641         if (ret != 1) {
 642                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 643                 return 1;
 644         }
 645         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 646
 647         flags = CLONE_NEWPID | SIGCHLD;
 648         if (!lxc_list_empty(&conf->id_map))
 649                 flags |= CLONE_NEWUSER;
 650
 651 #ifdef __ia64__
 652         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 653 #else
 654         stack += stack_size;
 655         pid = clone(container_reboot_supported, stack, flags, &cmd);
 656 #endif
 657         if (pid < 0) {
 658                 if (flags & CLONE_NEWUSER)
 659                         ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno));
 660                 else
 661                         ERROR("failed to clone (%#x): %s", flags, strerror(errno));
 662                 return -1;
 663         }
 664         if (wait(&status) < 0) {
 665                 SYSERROR("unexpected wait error: %m");
 666                 return -1;
 667         }
 668
 669         if (WEXITSTATUS(status) != 1)
 670                 return 1;
 671
 672         return 0;
 673 }
 674
 675 /*
 676  * netpipe is used in the unprivileged case to transfer the ifindexes
 677  * from parent to child
 678  */
 679 static int netpipe = -1;
 680
 681 static inline int count_veths(struct lxc_list *network)
 682 {
 683         struct lxc_list *iterator;
 684         struct lxc_netdev *netdev;
 685         int count = 0;
 686
 687         lxc_list_for_each(iterator, network) {
 688                 netdev = iterator->elem;
 689                 if (netdev->type != LXC_NET_VETH)
 690                         continue;
 691                 count++;
 692         }
 693         return count;
 694 }
 695
 696 static int read_unpriv_netifindex(struct lxc_list *network)
 697 {
 698         struct lxc_list *iterator;
 699         struct lxc_netdev *netdev;
 700
 701         if (netpipe == -1)
 702                 return 0;
 703         lxc_list_for_each(iterator, network) {
 704                 netdev = iterator->elem;
 705                 if (netdev->type != LXC_NET_VETH)
 706                         continue;
 707                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 708                         ERROR("Out of memory");
 709                         close(netpipe);
 710                         return -1;
 711                 }
 712                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 713                         close(netpipe);
 714                         return -1;
 715                 }
 716         }
 717         close(netpipe);
 718         return 0;
 719 }
 720
 721 static int do_start(void *data)
 722 {
 723         struct lxc_list *iterator;
 724         struct lxc_handler *handler = data;
 725         int devnull_fd = -1, ret;
 726         char path[PATH_MAX];
 727
 728         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 729                 SYSERROR("failed to set sigprocmask");
 730                 return -1;
 731         }
 732
 733         /* This prctl must be before the synchro, so if the parent
 734          * dies before we set the parent death signal, we will detect
 735          * its death with the synchro right after, otherwise we have
 736          * a window where the parent can exit before we set the pdeath
 737          * signal leading to a unsupervized container.
 738          */
 739         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 740                 SYSERROR("failed to set pdeath signal");
 741                 return -1;
 742         }
 743
 744         lxc_sync_fini_parent(handler);
 745
 746         /* don't leak the pinfd to the container */
 747         if (handler->pinfd >= 0) {
 748                 close(handler->pinfd);
 749         }
 750
 751         if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP))
 752                 return -1;
 753
 754         /* Unshare CLONE_NEWNET after CLONE_NEWUSER  - see
 755           https://github.com/lxc/lxd/issues/1978 */
 756         if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) ==
 757                         (CLONE_NEWNET | CLONE_NEWUSER)) {
 758                 ret = unshare(CLONE_NEWNET);
 759                 if (ret < 0) {
 760                         SYSERROR("Error unsharing network namespace");
 761                         goto out_warn_father;
 762                 } else {
 763                         INFO("Unshared NET namespace.");
 764                 }
 765         }
 766
 767         /* Tell the parent task it can begin to configure the
 768          * container and wait for it to finish
 769          */
 770         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 771                 return -1;
 772
 773         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 774                 goto out_warn_father;
 775
 776         /*
 777          * if we are in a new user namespace, become root there to have
 778          * privilege over our namespace. When using lxc-execute we default to root,
 779          * but this can be overriden using the lxc.init_uid and lxc.init_gid
 780          * configuration options.
 781          */
 782         if (!lxc_list_empty(&handler->conf->id_map)) {
 783                 gid_t new_gid = 0;
 784                 if (handler->conf->is_execute && handler->conf->init_gid)
 785                         new_gid = handler->conf->init_gid;
 786
 787                 uid_t new_uid = 0;
 788                 if (handler->conf->is_execute && handler->conf->init_uid)
 789                         new_uid = handler->conf->init_uid;
 790
 791                 NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid);
 792                 if (setgid(new_gid)) {
 793                         SYSERROR("setgid");
 794                         goto out_warn_father;
 795                 }
 796                 if (setuid(new_uid)) {
 797                         SYSERROR("setuid");
 798                         goto out_warn_father;
 799                 }
 800                 if (setgroups(0, NULL)) {
 801                         SYSERROR("setgroups");
 802                         goto out_warn_father;
 803                 }
 804         }
 805
 806         if (access(handler->lxcpath, X_OK)) {
 807                 print_top_failing_dir(handler->lxcpath);
 808                 goto out_warn_father;
 809         }
 810
 811         #if HAVE_SYS_CAPABILITY_H
 812         if (handler->conf->need_utmp_watch) {
 813                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 814                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 815                         goto out_warn_father;
 816                 }
 817                 DEBUG("Dropped cap_sys_boot");
 818         }
 819         #endif
 820
 821         ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount);
 822         if (ret < 0 || ret >= sizeof(path)) {
 823                 SYSERROR("sprintf'd too many chars");
 824                 goto out_warn_father;
 825         }
 826
 827         /* In order to checkpoint restore, we need to have everything in the
 828          * same mount namespace. However, some containers may not have a
 829          * reasonable /dev (in particular, they may not have /dev/null), so we
 830          * can't set init's std fds to /dev/null by opening it from inside the
 831          * container.
 832          *
 833          * If that's the case, fall back to using the host's /dev/null. This
 834          * means that migration won't work, but at least we won't spew output
 835          * where it isn't wanted.
 836          */
 837         if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) {
 838                 devnull_fd = open_devnull();
 839
 840                 if (devnull_fd < 0)
 841                         goto out_warn_father;
 842                 WARN("using host's /dev/null for container init's std fds, migraiton won't work");
 843         }
 844
 845         /* Setup the container, ip, names, utsname, ... */
 846         if (lxc_setup(handler)) {
 847                 ERROR("failed to setup the container");
 848                 goto out_warn_father;
 849         }
 850
 851         /* ask father to setup cgroups and wait for him to finish */
 852         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 853                 goto out_error;
 854
 855         /* Unshare cgroup namespace after we have setup our cgroups. If we do it
 856          * earlier we end up with a wrong view of /proc/self/cgroup. For
 857          * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
 858          * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
 859          * /proc/self/cgroup would show us:
 860          *
 861          *      8:cpuset:/lxc/c
 862          *
 863          * whereas it should actually show
 864          *
 865          *      8:cpuset:/
 866          */
 867         if (cgns_supported()) {
 868                 if (unshare(CLONE_NEWCGROUP) < 0) {
 869                         INFO("Failed to unshare CLONE_NEWCGROUP.");
 870                         goto out_warn_father;
 871                 }
 872                 INFO("Unshared CLONE_NEWCGROUP.");
 873         }
 874
 875         /* Set the label to change to when we exec(2) the container's init */
 876         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 877                 goto out_warn_father;
 878
 879         /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
 880          * before we aren't allowed anymore. */
 881         if (handler->conf->no_new_privs) {
 882                 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
 883                         SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 884                         goto out_warn_father;
 885                 }
 886                 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 887         }
 888
 889         /* Some init's such as busybox will set sane tty settings on stdin,
 890          * stdout, stderr which it thinks is the console. We already set them
 891          * the way we wanted on the real terminal, and we want init to do its
 892          * setup on its console ie. the pty allocated in lxc_console_create()
 893          * so make sure that that pty is stdin,stdout,stderr.
 894          */
 895         if (lxc_console_set_stdfds(handler->conf->console.slave) < 0)
 896                 goto out_warn_father;
 897
 898         /* If we mounted a temporary proc, then unmount it now */
 899         tmp_proc_unmount(handler->conf);
 900
 901         if (lxc_seccomp_load(handler->conf) != 0)
 902                 goto out_warn_father;
 903
 904         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 905                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 906                 goto out_warn_father;
 907         }
 908
 909         /* The clearenv() and putenv() calls have been moved here
 910          * to allow us to use environment variables passed to the various
 911          * hooks, such as the start hook above.  Not all of the
 912          * variables like CONFIG_PATH or ROOTFS are valid in this
 913          * context but others are. */
 914         if (clearenv()) {
 915                 SYSERROR("failed to clear environment");
 916                 /* don't error out though */
 917         }
 918
 919         lxc_list_for_each(iterator, &handler->conf->environment) {
 920                 if (putenv((char *)iterator->elem)) {
 921                         SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem);
 922                         goto out_warn_father;
 923                 }
 924         }
 925
 926         if (putenv("container=lxc")) {
 927                 SYSERROR("failed to set environment variable 'container=lxc'");
 928                 goto out_warn_father;
 929         }
 930
 931         if (handler->conf->pty_names) {
 932                 if (putenv(handler->conf->pty_names)) {
 933                         SYSERROR("failed to set environment variable for container ptys");
 934                         goto out_warn_father;
 935                 }
 936         }
 937
 938         close(handler->sigfd);
 939
 940         if (devnull_fd < 0) {
 941                 devnull_fd = open_devnull();
 942
 943                 if (devnull_fd < 0)
 944                         goto out_warn_father;
 945         }
 946
 947         if (handler->backgrounded && set_stdfds(devnull_fd))
 948                 goto out_warn_father;
 949
 950         if (devnull_fd >= 0) {
 951                 close(devnull_fd);
 952                 devnull_fd = -1;
 953         }
 954
 955         setsid();
 956
 957         /* after this call, we are in error because this
 958          * ops should not return as it execs */
 959         handler->ops->start(handler, handler->data);
 960
 961 out_warn_father:
 962         /* we want the parent to know something went wrong, so we return a special
 963          * error code. */
 964         lxc_sync_wake_parent(handler, LXC_SYNC_ERROR);
 965
 966 out_error:
 967         if (devnull_fd >= 0)
 968                 close(devnull_fd);
 969
 970         return -1;
 971 }
 972
 973 static int save_phys_nics(struct lxc_conf *conf)
 974 {
 975         struct lxc_list *iterator;
 976         int am_root = (getuid() == 0);
 977
 978         if (!am_root)
 979                 return 0;
 980
 981         lxc_list_for_each(iterator, &conf->network) {
 982                 struct lxc_netdev *netdev = iterator->elem;
 983
 984                 if (netdev->type != LXC_NET_PHYS)
 985                         continue;
 986                 conf->saved_nics = realloc(conf->saved_nics,
 987                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 988                 if (!conf->saved_nics) {
 989                         SYSERROR("failed to allocate memory");
 990                         return -1;
 991                 }
 992                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 993                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 994                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 995                         SYSERROR("failed to allocate memory");
 996                         return -1;
 997                 }
 998                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 999                         conf->saved_nics[conf->num_savednics].ifindex,
1000                         conf->saved_nics[conf->num_savednics].orig_name);
1001                 conf->num_savednics++;
1002         }
1003
1004         return 0;
1005 }
1006
1007 static int recv_fd(int sock, int *fd)
1008 {
1009         if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) {
1010                 SYSERROR("Error receiving tty fd from child");
1011                 return -1;
1012         }
1013         if (*fd == -1)
1014                 return -1;
1015         return 0;
1016 }
1017
1018 static int recv_ttys_from_child(struct lxc_handler *handler)
1019 {
1020         struct lxc_conf *conf = handler->conf;
1021         int i, sock = handler->ttysock[1];
1022         struct lxc_tty_info *tty_info = &conf->tty_info;
1023
1024         if (!conf->tty)
1025                 return 0;
1026
1027         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty);
1028         if (!tty_info->pty_info) {
1029                 SYSERROR("failed to allocate pty_info");
1030                 return -1;
1031         }
1032
1033         for (i = 0; i < conf->tty; i++) {
1034                 struct lxc_pty_info *pty_info = &tty_info->pty_info[i];
1035                 pty_info->busy = 0;
1036                 if (recv_fd(sock, &pty_info->slave) < 0 ||
1037                                 recv_fd(sock, &pty_info->master) < 0) {
1038                         ERROR("Error receiving tty info from child");
1039                         return -1;
1040                 }
1041         }
1042         tty_info->nbtty = conf->tty;
1043
1044         return 0;
1045 }
1046
1047 void resolve_clone_flags(struct lxc_handler *handler)
1048 {
1049         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
1050         INFO("Adding CLONE_NEWPID to clone flags.");
1051         INFO("Adding CLONE_NEWNS to clone flags.");
1052
1053         if (!lxc_list_empty(&handler->conf->id_map)) {
1054                 INFO("Adding CLONE_NEWUSER to clone flags.");
1055                 handler->clone_flags |= CLONE_NEWUSER;
1056         }
1057
1058         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
1059                 if (!lxc_requests_empty_network(handler)) {
1060                         INFO("Adding CLONE_NEWNET to clone flags.");
1061                         handler->clone_flags |= CLONE_NEWNET;
1062                 }
1063         } else {
1064                 INFO("Inheriting a NET namespace.");
1065         }
1066
1067         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
1068                 INFO("Adding CLONE_NEWIPC to clone flags.");
1069                 handler->clone_flags |= CLONE_NEWIPC;
1070         } else {
1071                 INFO("Inheriting an IPC namespace.");
1072         }
1073
1074         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
1075                 INFO("Adding CLONE_NEWUTS to clone flags.");
1076                 handler->clone_flags |= CLONE_NEWUTS;
1077         } else {
1078                 INFO("Inheriting a UTS namespace.");
1079         }
1080 }
1081
1082 static int lxc_spawn(struct lxc_handler *handler)
1083 {
1084         int failed_before_rename = 0;
1085         const char *name = handler->name;
1086         bool cgroups_connected = false;
1087         int saved_ns_fd[LXC_NS_MAX];
1088         int preserve_mask = 0, i, flags;
1089         int netpipepair[2], nveths;
1090
1091         netpipe = -1;
1092
1093         for (i = 0; i < LXC_NS_MAX; i++)
1094                 if (handler->conf->inherit_ns_fd[i] != -1)
1095                         preserve_mask |= ns_info[i].clone_flag;
1096
1097         if (lxc_sync_init(handler))
1098                 return -1;
1099
1100         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
1101                 lxc_sync_fini(handler);
1102                 return -1;
1103         }
1104
1105         resolve_clone_flags(handler);
1106
1107         if (handler->clone_flags & CLONE_NEWNET) {
1108                 if (!lxc_list_empty(&handler->conf->network)) {
1109
1110                         /* Find gateway addresses from the link device, which is
1111                          * no longer accessible inside the container. Do this
1112                          * before creating network interfaces, since goto
1113                          * out_delete_net does not work before lxc_clone. */
1114                         if (lxc_find_gateway_addresses(handler)) {
1115                                 ERROR("failed to find gateway addresses");
1116                                 lxc_sync_fini(handler);
1117                                 return -1;
1118                         }
1119
1120                         /* that should be done before the clone because we will
1121                          * fill the netdev index and use them in the child
1122                          */
1123                         if (lxc_create_network(handler)) {
1124                                 ERROR("failed to create the network");
1125                                 lxc_sync_fini(handler);
1126                                 return -1;
1127                         }
1128                 }
1129
1130                 if (save_phys_nics(handler->conf)) {
1131                         ERROR("failed to save physical nic info");
1132                         goto out_abort;
1133                 }
1134         }
1135
1136         if (!cgroup_init(handler)) {
1137                 ERROR("failed initializing cgroup support");
1138                 goto out_delete_net;
1139         }
1140
1141         cgroups_connected = true;
1142
1143         if (!cgroup_create(handler)) {
1144                 ERROR("failed creating cgroups");
1145                 goto out_delete_net;
1146         }
1147
1148         /*
1149          * if the rootfs is not a blockdev, prevent the container from
1150          * marking it readonly.
1151          *
1152          * if the container is unprivileged then skip rootfs pinning
1153          */
1154         if (lxc_list_empty(&handler->conf->id_map)) {
1155                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
1156                 if (handler->pinfd == -1)
1157                         INFO("failed to pin the container's rootfs");
1158         }
1159
1160         if (!preserve_ns(saved_ns_fd, preserve_mask, getpid()))
1161                 goto out_delete_net;
1162
1163         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
1164                 goto out_delete_net;
1165
1166         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
1167                 if (pipe(netpipepair) < 0) {
1168                         SYSERROR("Error creating pipe");
1169                         goto out_delete_net;
1170                 }
1171                 /* store netpipe in the global var for do_start's use */
1172                 netpipe = netpipepair[0];
1173         }
1174
1175         /* Create a process in a new set of namespaces */
1176         flags = handler->clone_flags;
1177         if (handler->clone_flags & CLONE_NEWUSER)
1178                 flags &= ~CLONE_NEWNET;
1179         handler->pid = lxc_clone(do_start, handler, flags);
1180         if (handler->pid < 0) {
1181                 SYSERROR("Failed to fork into a set of new namespaces.");
1182                 goto out_delete_net;
1183         } else {
1184                 INFO("Cloned a set of new namespaces.");
1185         }
1186
1187         if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid))
1188                 INFO("Failed to preserve namespace for lxc.hook.stop.");
1189
1190         if (attach_ns(saved_ns_fd))
1191                 WARN("failed to restore saved namespaces");
1192
1193         lxc_sync_fini_child(handler);
1194
1195         /* map the container uids - the container became an invalid
1196          * userid the moment it was cloned with CLONE_NEWUSER - this
1197          * call doesn't change anything immediately, but allows the
1198          * container to setuid(0) (0 being mapped to something else on
1199          * the host) later to become a valid uid again */
1200         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
1201                 ERROR("failed to set up id mapping");
1202                 goto out_delete_net;
1203         }
1204
1205         if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) {
1206                 failed_before_rename = 1;
1207                 goto out_delete_net;
1208         }
1209
1210         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) {
1211                 failed_before_rename = 1;
1212                 goto out_delete_net;
1213         }
1214
1215         if (!cgroup_create_legacy(handler)) {
1216                 ERROR("failed to setup the legacy cgroups for %s", name);
1217                 goto out_delete_net;
1218         }
1219         if (!cgroup_setup_limits(handler, false)) {
1220                 ERROR("failed to setup the cgroup limits for '%s'", name);
1221                 goto out_delete_net;
1222         }
1223
1224         if (!cgroup_enter(handler))
1225                 goto out_delete_net;
1226
1227         if (!cgroup_chown(handler))
1228                 goto out_delete_net;
1229
1230         if (failed_before_rename)
1231                 goto out_delete_net;
1232
1233         /* Create the network configuration */
1234         if (handler->clone_flags & CLONE_NEWNET) {
1235                 if (lxc_assign_network(handler->lxcpath, handler->name,
1236                                         &handler->conf->network, handler->pid)) {
1237                         ERROR("failed to create the configured network");
1238                         goto out_delete_net;
1239                 }
1240         }
1241
1242         if (netpipe != -1) {
1243                 struct lxc_list *iterator;
1244                 struct lxc_netdev *netdev;
1245
1246                 close(netpipe);
1247                 lxc_list_for_each(iterator, &handler->conf->network) {
1248                         netdev = iterator->elem;
1249                         if (netdev->type != LXC_NET_VETH)
1250                                 continue;
1251                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1252                                 ERROR("Error writing veth name to container");
1253                                 goto out_delete_net;
1254                         }
1255                 }
1256                 close(netpipepair[1]);
1257         }
1258
1259         /* Tell the child to continue its initialization.  we'll get
1260          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1261          */
1262         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1263                 goto out_delete_net;
1264
1265         if (!cgroup_setup_limits(handler, true)) {
1266                 ERROR("failed to setup the devices cgroup for '%s'", name);
1267                 goto out_delete_net;
1268         }
1269
1270         cgroup_disconnect();
1271         cgroups_connected = false;
1272
1273         /* read tty fds allocated by child */
1274         if (recv_ttys_from_child(handler) < 0) {
1275                 ERROR("failed to receive tty info from child");
1276                 goto out_delete_net;
1277         }
1278
1279         /* Tell the child to complete its initialization and wait for
1280          * it to exec or return an error.  (the child will never
1281          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
1282          * sync pipe, causing lxc_sync_barrier_child to return
1283          * success, or return a different value, causing us to error
1284          * out).
1285          */
1286         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1287                 return -1;
1288
1289         if (detect_shared_rootfs())
1290                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
1291
1292         if (handler->ops->post_start(handler, handler->data))
1293                 goto out_abort;
1294
1295         if (lxc_set_state(name, handler, RUNNING)) {
1296                 ERROR("failed to set state to %s",
1297                               lxc_state2str(RUNNING));
1298                 goto out_abort;
1299         }
1300
1301         lxc_sync_fini(handler);
1302         handler->netnsfd = lxc_preserve_ns(handler->pid, "net");
1303
1304         return 0;
1305
1306 out_delete_net:
1307         if (cgroups_connected)
1308                 cgroup_disconnect();
1309         if (handler->clone_flags & CLONE_NEWNET)
1310                 lxc_delete_network(handler);
1311 out_abort:
1312         lxc_abort(name, handler);
1313         lxc_sync_fini(handler);
1314         if (handler->pinfd >= 0) {
1315                 close(handler->pinfd);
1316                 handler->pinfd = -1;
1317         }
1318
1319         return -1;
1320 }
1321
1322 int __lxc_start(const char *name, struct lxc_conf *conf,
1323                 struct lxc_operations* ops, void *data, const char *lxcpath,
1324                 bool backgrounded)
1325 {
1326         struct lxc_handler *handler;
1327         int err = -1;
1328         int status;
1329         bool removed_all_netdevs = true;
1330
1331         handler = lxc_init(name, conf, lxcpath);
1332         if (!handler) {
1333                 ERROR("failed to initialize the container");
1334                 return -1;
1335         }
1336         handler->ops = ops;
1337         handler->data = data;
1338         handler->backgrounded = backgrounded;
1339         handler->netnsfd = -1;
1340
1341         if (must_drop_cap_sys_boot(handler->conf)) {
1342                 #if HAVE_SYS_CAPABILITY_H
1343                 DEBUG("Dropping cap_sys_boot");
1344                 #else
1345                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1346                 #endif
1347         } else {
1348                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1349                 handler->conf->need_utmp_watch = 0;
1350         }
1351
1352         if (!attach_block_device(handler->conf)) {
1353                 ERROR("Failure attaching block device");
1354                 goto out_fini_nonet;
1355         }
1356
1357         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1358                 /* if the backing store is a device, mount it here and now */
1359                 if (rootfs_is_blockdev(conf)) {
1360                         if (unshare(CLONE_NEWNS) < 0) {
1361                                 ERROR("Error unsharing MOUNT namespace.");
1362                                 goto out_fini_nonet;
1363                         } else {
1364                                 INFO("Unshared MOUNT namespace.");
1365                         }
1366                         remount_all_slave();
1367                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1368                                 ERROR("Error setting up rootfs mount as root before spawn");
1369                                 goto out_fini_nonet;
1370                         }
1371                         INFO("Set up container rootfs as host root");
1372                 }
1373         }
1374
1375         err = lxc_spawn(handler);
1376         if (err) {
1377                 ERROR("failed to spawn '%s'", name);
1378                 goto out_detach_blockdev;
1379         }
1380
1381         handler->conf->reboot = 0;
1382
1383         err = lxc_poll(name, handler);
1384         if (err) {
1385                 ERROR("mainloop exited with an error");
1386                 if (handler->netnsfd >= 0) {
1387                         close(handler->netnsfd);
1388                         handler->netnsfd = -1;
1389                 }
1390                 goto out_abort;
1391         }
1392
1393         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1394                 continue;
1395
1396         /*
1397          * If the child process exited but was not signaled,
1398          * it didn't call reboot.  This should mean it was an
1399          * lxc-execute which simply exited.  In any case, treat
1400          * it as a 'halt'
1401          */
1402         if (WIFSIGNALED(status)) {
1403                 switch(WTERMSIG(status)) {
1404                 case SIGINT: /* halt */
1405                         DEBUG("Container halting");
1406                         break;
1407                 case SIGHUP: /* reboot */
1408                         DEBUG("Container rebooting");
1409                         handler->conf->reboot = 1;
1410                         break;
1411                 case SIGSYS: /* seccomp */
1412                         DEBUG("Container violated its seccomp policy");
1413                         break;
1414                 default:
1415                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1416                         break;
1417                 }
1418         }
1419
1420         DEBUG("Pushing physical nics back to host namespace");
1421         lxc_restore_phys_nics_to_netns(handler->netnsfd, handler->conf);
1422
1423         DEBUG("Tearing down virtual network devices used by container");
1424         removed_all_netdevs = lxc_delete_network(handler);
1425
1426         if (handler->pinfd >= 0) {
1427                 close(handler->pinfd);
1428                 handler->pinfd = -1;
1429         }
1430
1431         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1432         err =  lxc_error_set_and_log(handler->pid, status);
1433 out_fini:
1434         if (!removed_all_netdevs) {
1435                 DEBUG("Failed tearing down all network devices used by container. Trying again!");
1436                 removed_all_netdevs = lxc_delete_network(handler);
1437                 if (!removed_all_netdevs)
1438                         DEBUG("Failed tearing down network devices used by container. Not trying again!");
1439         }
1440
1441 out_detach_blockdev:
1442         detach_block_device(handler->conf);
1443
1444 out_fini_nonet:
1445         lxc_fini(name, handler);
1446         return err;
1447
1448 out_abort:
1449         lxc_abort(name, handler);
1450         goto out_fini;
1451 }
1452
1453 struct start_args {
1454         char *const *argv;
1455 };
1456
1457 static int start(struct lxc_handler *handler, void* data)
1458 {
1459         struct start_args *arg = data;
1460
1461         NOTICE("exec'ing '%s'", arg->argv[0]);
1462
1463         execvp(arg->argv[0], arg->argv);
1464         SYSERROR("failed to exec %s", arg->argv[0]);
1465         return 0;
1466 }
1467
1468 static int post_start(struct lxc_handler *handler, void* data)
1469 {
1470         struct start_args *arg = data;
1471
1472         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1473         return 0;
1474 }
1475
1476 static struct lxc_operations start_ops = {
1477         .start = start,
1478         .post_start = post_start
1479 };
1480
1481 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1482               const char *lxcpath, bool backgrounded)
1483 {
1484         struct start_args start_arg = {
1485                 .argv = argv,
1486         };
1487
1488         conf->need_utmp_watch = 1;
1489         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded);
1490 }
1491
1492 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1493                                             const char *name)
1494 {
1495         char destroy[MAXPATHLEN];
1496         bool bret = true;
1497         int ret = 0;
1498         struct lxc_container *c;
1499         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1500                 bret = do_destroy_container(handler->conf);
1501                 if (!bret) {
1502                         ERROR("Error destroying rootfs for %s", name);
1503                         return;
1504                 }
1505         }
1506         INFO("Destroyed rootfs for %s", name);
1507
1508         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1509         if (ret < 0 || ret >= MAXPATHLEN) {
1510                 ERROR("Error printing path for %s", name);
1511                 ERROR("Error destroying directory for %s", name);
1512                 return;
1513         }
1514
1515         c = lxc_container_new(name, handler->lxcpath);
1516         if (c) {
1517                 if (container_disk_lock(c)) {
1518                         INFO("Could not update lxc_snapshots file");
1519                         lxc_container_put(c);
1520                 } else {
1521                         mod_all_rdeps(c, false);
1522                         container_disk_unlock(c);
1523                         lxc_container_put(c);
1524                 }
1525         }
1526
1527         if (am_unpriv())
1528                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, destroy);
1529         else
1530                 ret = lxc_rmdir_onedev(destroy, NULL);
1531
1532         if (ret < 0) {
1533                 ERROR("Error destroying directory for %s", name);
1534                 return;
1535         }
1536         INFO("Destroyed directory for %s", name);
1537 }
1538
1539 static int lxc_rmdir_onedev_wrapper(void *data)
1540 {
1541         char *arg = (char *) data;
1542         return lxc_rmdir_onedev(arg, NULL);
1543 }
1544
1545 static bool do_destroy_container(struct lxc_conf *conf) {
1546         if (am_unpriv()) {
1547                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf) < 0)
1548                         return false;
1549                 return true;
1550         }
1551         return bdev_destroy(conf);
1552 }