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