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_LIBCAP
  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                 NOTICE("Received SIGCHLD from pid %d instead of container init %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_LIBCAP
 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_handler(const char *name, struct lxc_conf *conf,
 400                                      const char *lxcpath)
 401 {
 402         int i;
 403         struct lxc_handler *handler;
 404
 405         handler = malloc(sizeof(*handler));
 406         if (!handler) {
 407                 ERROR("failed to allocate memory");
 408                 return NULL;
 409         }
 410
 411         memset(handler, 0, sizeof(*handler));
 412
 413         handler->ttysock[0] = handler->ttysock[1] = -1;
 414         handler->conf = conf;
 415         handler->lxcpath = lxcpath;
 416         handler->pinfd = -1;
 417
 418         for (i = 0; i < LXC_NS_MAX; i++)
 419                 handler->nsfd[i] = -1;
 420
 421         handler->name = strdup(name);
 422         if (!handler->name) {
 423                 ERROR("failed to allocate memory");
 424                 goto do_partial_cleanup;
 425         }
 426
 427         if (lxc_cmd_init(name, handler, lxcpath)) {
 428                 ERROR("failed to set up command socket");
 429                 goto do_full_cleanup;
 430         }
 431
 432         TRACE("unix domain socket %d for command server is ready",
 433               handler->conf->maincmd_fd);
 434
 435         return handler;
 436
 437 do_full_cleanup:
 438         free(handler->name);
 439
 440 do_partial_cleanup:
 441         free(handler);
 442
 443         return NULL;
 444 }
 445
 446 int lxc_init(const char *name, struct lxc_handler *handler)
 447 {
 448         struct lxc_conf *conf = handler->conf;
 449
 450         lsm_init();
 451         TRACE("initialized LSM");
 452
 453         if (lxc_read_seccomp_config(conf) != 0) {
 454                 ERROR("Failed loading seccomp policy.");
 455                 goto out_close_maincmd_fd;
 456         }
 457
 458         /* Begin by setting the state to STARTING. */
 459         if (lxc_set_state(name, handler, STARTING)) {
 460                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING));
 461                 goto out_close_maincmd_fd;
 462         }
 463
 464         /* Start of environment variable setup for hooks. */
 465         if (name && setenv("LXC_NAME", name, 1))
 466                 SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name);
 467
 468         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1))
 469                 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile);
 470
 471         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1))
 472                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount);
 473
 474         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1))
 475                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path);
 476
 477         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1))
 478                 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path);
 479
 480         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1))
 481                 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path);
 482
 483         if (setenv("LXC_CGNS_AWARE", "1", 1))
 484                 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1.");
 485         /* End of environment variable setup for hooks. */
 486
 487         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 488                 ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name);
 489                 goto out_aborting;
 490         }
 491
 492         /* The signal fd has to be created before forking otherwise if the child
 493          * process exits before we setup the signal fd, the event will be lost
 494          * and the command will be stuck.
 495          */
 496         handler->sigfd = setup_signal_fd(&handler->oldmask);
 497         if (handler->sigfd < 0) {
 498                 ERROR("Failed to setup SIGCHLD fd handler.");
 499                 goto out_delete_tty;
 500         }
 501
 502         /* Do this after setting up signals since it might unblock SIGWINCH. */
 503         if (lxc_console_create(conf)) {
 504                 ERROR("Failed to create console for container \"%s\".", name);
 505                 goto out_restore_sigmask;
 506         }
 507
 508         if (lxc_ttys_shift_ids(conf) < 0) {
 509                 ERROR("Failed to shift tty into container.");
 510                 goto out_restore_sigmask;
 511         }
 512
 513         INFO("Container \"%s\" is initialized.", name);
 514         return 0;
 515
 516 out_restore_sigmask:
 517         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 518 out_delete_tty:
 519         lxc_delete_tty(&conf->tty_info);
 520 out_aborting:
 521         lxc_set_state(name, handler, ABORTING);
 522 out_close_maincmd_fd:
 523         close(conf->maincmd_fd);
 524         conf->maincmd_fd = -1;
 525         return -1;
 526 }
 527
 528 void lxc_fini(const char *name, struct lxc_handler *handler)
 529 {
 530         int i, rc;
 531         pid_t self = getpid();
 532         char *namespaces[LXC_NS_MAX+1];
 533         size_t namespace_count = 0;
 534
 535         /* The STOPPING state is there for future cleanup code which can take
 536          * awhile.
 537          */
 538         lxc_set_state(name, handler, STOPPING);
 539
 540         for (i = 0; i < LXC_NS_MAX; i++) {
 541                 if (handler->nsfd[i] != -1) {
 542                         rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d",
 543                                       ns_info[i].proc_name, self, handler->nsfd[i]);
 544                         if (rc == -1) {
 545                                 SYSERROR("Failed to allocate memory.");
 546                                 break;
 547                         }
 548                         ++namespace_count;
 549                 }
 550         }
 551         namespaces[namespace_count] = NULL;
 552
 553         if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1))
 554                 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot.");
 555
 556         if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1))
 557                 SYSERROR("Failed to set environment variable: LXC_TARGET=stop.");
 558
 559         if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces))
 560                 ERROR("Failed to run lxc.hook.stop for container \"%s\".", name);
 561
 562         while (namespace_count--)
 563                 free(namespaces[namespace_count]);
 564         for (i = 0; i < LXC_NS_MAX; i++) {
 565                 if (handler->nsfd[i] != -1) {
 566                         close(handler->nsfd[i]);
 567                         handler->nsfd[i] = -1;
 568                 }
 569         }
 570
 571         if (handler->netnsfd >= 0) {
 572                 close(handler->netnsfd);
 573                 handler->netnsfd = -1;
 574         }
 575
 576         lxc_set_state(name, handler, STOPPED);
 577
 578         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) {
 579                 ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name);
 580                 if (handler->conf->reboot) {
 581                         WARN("Container will be stopped instead of rebooted.");
 582                         handler->conf->reboot = 0;
 583                         if (setenv("LXC_TARGET", "stop", 1))
 584                                 WARN("Failed to set environment variable: LXC_TARGET=stop.");
 585                 }
 586         }
 587
 588         /* Reset mask set by setup_signal_fd. */
 589         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 590                 WARN("Failed to restore signal mask.");
 591
 592         lxc_console_delete(&handler->conf->console);
 593         lxc_delete_tty(&handler->conf->tty_info);
 594         close(handler->conf->maincmd_fd);
 595         handler->conf->maincmd_fd = -1;
 596         free(handler->name);
 597         if (handler->ttysock[0] != -1) {
 598                 close(handler->ttysock[0]);
 599                 close(handler->ttysock[1]);
 600         }
 601
 602         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1)
 603                 lxc_destroy_container_on_signal(handler, name);
 604
 605         cgroup_destroy(handler);
 606         free(handler);
 607 }
 608
 609 void lxc_abort(const char *name, struct lxc_handler *handler)
 610 {
 611         int ret, status;
 612
 613         lxc_set_state(name, handler, ABORTING);
 614         if (handler->pid > 0)
 615                 kill(handler->pid, SIGKILL);
 616         while ((ret = waitpid(-1, &status, 0)) > 0) {
 617                 ;
 618         }
 619 }
 620
 621 #include <sys/reboot.h>
 622 #include <linux/reboot.h>
 623
 624 /* reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL in a child pid namespace
 625  * if container reboot support exists.  Otherwise, it will either succeed or
 626  * return -EPERM.
 627  */
 628 static int container_reboot_supported(void *arg)
 629 {
 630         int *cmd = arg;
 631         int ret;
 632
 633         ret = reboot(*cmd);
 634         if (ret == -1 && errno == EINVAL)
 635                 return 1;
 636         return 0;
 637 }
 638
 639 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 640 {
 641         FILE *f;
 642         int ret, cmd, v, flags;
 643         long stack_size = 4096;
 644         void *stack = alloca(stack_size);
 645         int status;
 646         pid_t pid;
 647
 648         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 649         if (!f) {
 650                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 651                 return 1;
 652         }
 653
 654         ret = fscanf(f, "%d", &v);
 655         fclose(f);
 656         if (ret != 1) {
 657                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del.");
 658                 return 1;
 659         }
 660         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 661
 662         flags = CLONE_NEWPID | SIGCHLD;
 663         if (!lxc_list_empty(&conf->id_map))
 664                 flags |= CLONE_NEWUSER;
 665
 666 #ifdef __ia64__
 667         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 668 #else
 669         stack += stack_size;
 670         pid = clone(container_reboot_supported, stack, flags, &cmd);
 671 #endif
 672         if (pid < 0) {
 673                 if (flags & CLONE_NEWUSER)
 674                         ERROR("Failed to clone (%#x): %s (includes CLONE_NEWUSER).", flags, strerror(errno));
 675                 else
 676                         ERROR("Failed to clone (%#x): %s.", flags, strerror(errno));
 677                 return -1;
 678         }
 679         if (wait(&status) < 0) {
 680                 SYSERROR("Unexpected wait error: %m.");
 681                 return -1;
 682         }
 683
 684         if (WEXITSTATUS(status) != 1)
 685                 return 1;
 686
 687         return 0;
 688 }
 689
 690 /* netpipe is used in the unprivileged case to transfer the ifindexes from
 691  * parent to child
 692  */
 693 static int netpipe = -1;
 694
 695 static inline int count_veths(struct lxc_list *network)
 696 {
 697         struct lxc_list *iterator;
 698         struct lxc_netdev *netdev;
 699         int count = 0;
 700
 701         lxc_list_for_each(iterator, network) {
 702                 netdev = iterator->elem;
 703                 if (netdev->type != LXC_NET_VETH)
 704                         continue;
 705                 count++;
 706         }
 707         return count;
 708 }
 709
 710 static int read_unpriv_netifindex(struct lxc_list *network)
 711 {
 712         struct lxc_list *iterator;
 713         struct lxc_netdev *netdev;
 714
 715         if (netpipe == -1)
 716                 return 0;
 717         lxc_list_for_each(iterator, network) {
 718                 netdev = iterator->elem;
 719                 if (netdev->type != LXC_NET_VETH)
 720                         continue;
 721                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 722                         ERROR("Out of memory.");
 723                         close(netpipe);
 724                         return -1;
 725                 }
 726                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 727                         close(netpipe);
 728                         return -1;
 729                 }
 730         }
 731         close(netpipe);
 732         return 0;
 733 }
 734
 735 static int do_start(void *data)
 736 {
 737         struct lxc_list *iterator;
 738         struct lxc_handler *handler = data;
 739         int devnull_fd = -1, ret;
 740         char path[PATH_MAX];
 741
 742         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 743                 SYSERROR("Failed to set signal mask.");
 744                 return -1;
 745         }
 746
 747         /* This prctl must be before the synchro, so if the parent dies before
 748          * we set the parent death signal, we will detect its death with the
 749          * synchro right after, otherwise we have a window where the parent can
 750          * exit before we set the pdeath signal leading to a unsupervized
 751          * container.
 752          */
 753         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 754                 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL.");
 755                 return -1;
 756         }
 757
 758         lxc_sync_fini_parent(handler);
 759
 760         /* Don't leak the pinfd to the container. */
 761         if (handler->pinfd >= 0) {
 762                 close(handler->pinfd);
 763         }
 764
 765         if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP))
 766                 return -1;
 767
 768         /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
 769          * https://github.com/lxc/lxd/issues/1978.
 770          */
 771         if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) ==
 772             (CLONE_NEWNET | CLONE_NEWUSER)) {
 773                 ret = unshare(CLONE_NEWNET);
 774                 if (ret < 0) {
 775                         SYSERROR("Failed to unshare CLONE_NEWNET.");
 776                         goto out_warn_father;
 777                 }
 778                 INFO("Unshared CLONE_NEWNET.");
 779         }
 780
 781         /* Tell the parent task it can begin to configure the container and wait
 782          * for it to finish.
 783          */
 784         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 785                 return -1;
 786
 787         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 788                 goto out_warn_father;
 789
 790         /* If we are in a new user namespace, become root there to have
 791          * privilege over our namespace.
 792          */
 793         if (!lxc_list_empty(&handler->conf->id_map)) {
 794                 if (lxc_switch_uid_gid(0, 0) < 0)
 795                         goto out_warn_father;
 796
 797                 /* Drop groups only after we switched to a valid gid in the new
 798                  * user namespace.
 799                  */
 800                 if (lxc_setgroups(0, NULL) < 0)
 801                         goto out_warn_father;
 802         }
 803
 804         if (access(handler->lxcpath, X_OK)) {
 805                 print_top_failing_dir(handler->lxcpath);
 806                 goto out_warn_father;
 807         }
 808
 809         #if HAVE_LIBCAP
 810         if (handler->conf->need_utmp_watch) {
 811                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 812                         SYSERROR("Failed to remove the CAP_SYS_BOOT capability.");
 813                         goto out_warn_father;
 814                 }
 815                 DEBUG("Dropped the CAP_SYS_BOOT capability.");
 816         }
 817         #endif
 818
 819         ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount);
 820         if (ret < 0 || ret >= sizeof(path))
 821                 goto out_warn_father;
 822
 823         /* In order to checkpoint restore, we need to have everything in the
 824          * same mount namespace. However, some containers may not have a
 825          * reasonable /dev (in particular, they may not have /dev/null), so we
 826          * can't set init's std fds to /dev/null by opening it from inside the
 827          * container.
 828          *
 829          * If that's the case, fall back to using the host's /dev/null. This
 830          * means that migration won't work, but at least we won't spew output
 831          * where it isn't wanted.
 832          */
 833         if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) {
 834                 devnull_fd = open_devnull();
 835
 836                 if (devnull_fd < 0)
 837                         goto out_warn_father;
 838                 WARN("Using /dev/null from the host for container init's "
 839                      "standard file descriptors. Migration will not work.");
 840         }
 841
 842         /* Setup the container, ip, names, utsname, ... */
 843         if (lxc_setup(handler)) {
 844                 ERROR("Failed to setup container \"%s\".", handler->name);
 845                 goto out_warn_father;
 846         }
 847
 848         /* Ask father to setup cgroups and wait for him to finish. */
 849         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 850                 goto out_error;
 851
 852         /* Unshare cgroup namespace after we have setup our cgroups. If we do it
 853          * earlier we end up with a wrong view of /proc/self/cgroup. For
 854          * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
 855          * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
 856          * /proc/self/cgroup would show us:
 857          *
 858          *      8:cpuset:/lxc/c
 859          *
 860          * whereas it should actually show
 861          *
 862          *      8:cpuset:/
 863          */
 864         if (cgns_supported()) {
 865                 if (unshare(CLONE_NEWCGROUP) < 0) {
 866                         INFO("Failed to unshare CLONE_NEWCGROUP.");
 867                         goto out_warn_father;
 868                 }
 869                 INFO("Unshared CLONE_NEWCGROUP.");
 870         }
 871
 872         /* Set the label to change to when we exec(2) the container's init. */
 873         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 874                 goto out_warn_father;
 875
 876         /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
 877          * before we aren't allowed anymore.
 878          */
 879         if (handler->conf->no_new_privs) {
 880                 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
 881                         SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 882                         goto out_warn_father;
 883                 }
 884                 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 885         }
 886
 887         /* Some init's such as busybox will set sane tty settings on stdin,
 888          * stdout, stderr which it thinks is the console. We already set them
 889          * the way we wanted on the real terminal, and we want init to do its
 890          * setup on its console ie. the pty allocated in lxc_console_create() so
 891          * make sure that that pty is stdin,stdout,stderr.
 892          */
 893         if (lxc_console_set_stdfds(handler->conf->console.slave) < 0)
 894                 goto out_warn_father;
 895
 896         /* If we mounted a temporary proc, then unmount it now. */
 897         tmp_proc_unmount(handler->conf);
 898
 899         if (lxc_seccomp_load(handler->conf) != 0)
 900                 goto out_warn_father;
 901
 902         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 903                 ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name);
 904                 goto out_warn_father;
 905         }
 906
 907         /* The container has been setup. We can now switch to an unprivileged
 908          * uid/gid.
 909          */
 910         if (handler->conf->is_execute) {
 911                 bool have_cap_setgid;
 912                 uid_t new_uid = handler->conf->init_uid;
 913                 gid_t new_gid = handler->conf->init_gid;
 914
 915                 /* If we are in a new user namespace we already dropped all
 916                  * groups when we switched to root in the new user namespace
 917                  * further above. Only drop groups if we can, so ensure that we
 918                  * have necessary privilege.
 919                  */
 920                 #if HAVE_LIBCAP
 921                 have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE);
 922                 #else
 923                 have_cap_setgid = false;
 924                 #endif
 925                 if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) {
 926                         if (lxc_setgroups(0, NULL) < 0)
 927                                 goto out_warn_father;
 928                 }
 929
 930                 if (lxc_switch_uid_gid(new_uid, new_gid) < 0)
 931                         goto out_warn_father;
 932         }
 933
 934         /* The clearenv() and putenv() calls have been moved here to allow us to
 935          * use environment variables passed to the various hooks, such as the
 936          * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS
 937          * are valid in this context but others are.
 938          */
 939         if (clearenv()) {
 940                 SYSERROR("Failed to clear environment.");
 941                 /* Don't error out though. */
 942         }
 943
 944         lxc_list_for_each(iterator, &handler->conf->environment) {
 945                 if (putenv((char *)iterator->elem)) {
 946                         SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem);
 947                         goto out_warn_father;
 948                 }
 949         }
 950
 951         if (putenv("container=lxc")) {
 952                 SYSERROR("Failed to set environment variable: container=lxc.");
 953                 goto out_warn_father;
 954         }
 955
 956         if (handler->conf->pty_names) {
 957                 if (putenv(handler->conf->pty_names)) {
 958                         SYSERROR("Failed to set environment variable for container ptys.");
 959                         goto out_warn_father;
 960                 }
 961         }
 962
 963         close(handler->sigfd);
 964
 965         if (devnull_fd < 0) {
 966                 devnull_fd = open_devnull();
 967
 968                 if (devnull_fd < 0)
 969                         goto out_warn_father;
 970         }
 971
 972         if (handler->backgrounded && set_stdfds(devnull_fd))
 973                 goto out_warn_father;
 974
 975         if (devnull_fd >= 0) {
 976                 close(devnull_fd);
 977                 devnull_fd = -1;
 978         }
 979
 980         setsid();
 981
 982         /* After this call, we are in error because this ops should not return
 983          * as it execs.
 984          */
 985         handler->ops->start(handler, handler->data);
 986
 987 out_warn_father:
 988         /* We want the parent to know something went wrong, so we return a
 989          * special error code.
 990          */
 991         lxc_sync_wake_parent(handler, LXC_SYNC_ERROR);
 992
 993 out_error:
 994         if (devnull_fd >= 0)
 995                 close(devnull_fd);
 996
 997         return -1;
 998 }
 999
1000 static int save_phys_nics(struct lxc_conf *conf)
1001 {
1002         struct lxc_list *iterator;
1003         int am_root = (getuid() == 0);
1004
1005         if (!am_root)
1006                 return 0;
1007
1008         lxc_list_for_each(iterator, &conf->network) {
1009                 struct lxc_netdev *netdev = iterator->elem;
1010
1011                 if (netdev->type != LXC_NET_PHYS)
1012                         continue;
1013                 conf->saved_nics = realloc(conf->saved_nics,
1014                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
1015                 if (!conf->saved_nics)
1016                         return -1;
1017                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
1018                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
1019                 if (!conf->saved_nics[conf->num_savednics].orig_name)
1020                         return -1;
1021                 INFO("Stored saved_nic #%d idx %d name %s.", conf->num_savednics,
1022                         conf->saved_nics[conf->num_savednics].ifindex,
1023                         conf->saved_nics[conf->num_savednics].orig_name);
1024                 conf->num_savednics++;
1025         }
1026
1027         return 0;
1028 }
1029
1030 static int lxc_recv_ttys_from_child(struct lxc_handler *handler)
1031 {
1032         int i;
1033         int *ttyfds;
1034         struct lxc_pty_info *pty_info;
1035         int ret = -1;
1036         int sock = handler->ttysock[1];
1037         struct lxc_conf *conf = handler->conf;
1038         struct lxc_tty_info *tty_info = &conf->tty_info;
1039         size_t num_ttyfds = (2 * conf->tty);
1040
1041         if (!conf->tty)
1042                 return 0;
1043
1044         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty);
1045         if (!tty_info->pty_info)
1046                 return -1;
1047
1048         ttyfds = malloc(num_ttyfds * sizeof(int));
1049         if (!ttyfds)
1050                 return -1;
1051
1052         ret = lxc_abstract_unix_recv_fds(sock, ttyfds, num_ttyfds, NULL, 0);
1053         for (i = 0; (ret >= 0 && *ttyfds != -1) && (i < num_ttyfds); i++) {
1054                 pty_info = &tty_info->pty_info[i / 2];
1055                 pty_info->busy = 0;
1056                 pty_info->slave = ttyfds[i++];
1057                 pty_info->master = ttyfds[i];
1058                 TRACE("received pty with master fd %d and slave fd %d from "
1059                       "parent", pty_info->master, pty_info->slave);
1060         }
1061
1062         tty_info->nbtty = conf->tty;
1063
1064         free(ttyfds);
1065
1066         if (ret < 0)
1067                 ERROR("failed to receive %d ttys from child: %s", conf->tty,
1068                       strerror(errno));
1069         else
1070                 TRACE("received %d ttys from child", conf->tty);
1071
1072         return ret;
1073 }
1074
1075 void resolve_clone_flags(struct lxc_handler *handler)
1076 {
1077         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
1078
1079         if (!lxc_list_empty(&handler->conf->id_map))
1080                 handler->clone_flags |= CLONE_NEWUSER;
1081
1082         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
1083                 if (!lxc_requests_empty_network(handler))
1084                         handler->clone_flags |= CLONE_NEWNET;
1085         } else {
1086                 INFO("Inheriting a NET namespace.");
1087         }
1088
1089         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1)
1090                 handler->clone_flags |= CLONE_NEWIPC;
1091         else
1092                 INFO("Inheriting an IPC namespace.");
1093
1094         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1)
1095                 handler->clone_flags |= CLONE_NEWUTS;
1096         else
1097                 INFO("Inheriting a UTS namespace.");
1098 }
1099
1100 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1101  * exec()s the requested container binary.
1102  * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1103  * right here should be double checked if they'd pose a security risk. (For
1104  * example, any {u}mount() operations performed here will be reflected on the
1105  * host!)
1106  */
1107 static int lxc_spawn(struct lxc_handler *handler)
1108 {
1109         int failed_before_rename = 0;
1110         const char *name = handler->name;
1111         bool cgroups_connected = false;
1112         int saved_ns_fd[LXC_NS_MAX];
1113         int preserve_mask = 0, i, flags;
1114         int netpipepair[2], nveths;
1115         bool wants_to_map_ids;
1116         struct lxc_list *id_map;
1117
1118         netpipe = -1;
1119         id_map = &handler->conf->id_map;
1120         wants_to_map_ids = !lxc_list_empty(id_map);
1121
1122         for (i = 0; i < LXC_NS_MAX; i++)
1123                 if (handler->conf->inherit_ns_fd[i] != -1)
1124                         preserve_mask |= ns_info[i].clone_flag;
1125
1126         if (lxc_sync_init(handler))
1127                 return -1;
1128
1129         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
1130                 lxc_sync_fini(handler);
1131                 return -1;
1132         }
1133
1134         resolve_clone_flags(handler);
1135
1136         if (handler->clone_flags & CLONE_NEWNET) {
1137                 if (!lxc_list_empty(&handler->conf->network)) {
1138
1139                         /* Find gateway addresses from the link device, which is
1140                          * no longer accessible inside the container. Do this
1141                          * before creating network interfaces, since goto
1142                          * out_delete_net does not work before lxc_clone.
1143                          */
1144                         if (lxc_find_gateway_addresses(handler)) {
1145                                 ERROR("Failed to find gateway addresses.");
1146                                 lxc_sync_fini(handler);
1147                                 return -1;
1148                         }
1149
1150                         /* That should be done before the clone because we will
1151                          * fill the netdev index and use them in the child.
1152                          */
1153                         if (lxc_create_network(handler)) {
1154                                 ERROR("Failed to create the network.");
1155                                 lxc_sync_fini(handler);
1156                                 return -1;
1157                         }
1158                 }
1159
1160                 if (save_phys_nics(handler->conf)) {
1161                         ERROR("Failed to save physical nic info.");
1162                         goto out_abort;
1163                 }
1164         }
1165
1166         if (!cgroup_init(handler)) {
1167                 ERROR("Failed initializing cgroup support.");
1168                 goto out_delete_net;
1169         }
1170
1171         cgroups_connected = true;
1172
1173         if (!cgroup_create(handler)) {
1174                 ERROR("Failed creating cgroups.");
1175                 goto out_delete_net;
1176         }
1177
1178         /* If the rootfs is not a blockdev, prevent the container from marking
1179          * it readonly.
1180          * If the container is unprivileged then skip rootfs pinning.
1181          */
1182         if (!wants_to_map_ids) {
1183                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
1184                 if (handler->pinfd == -1)
1185                         INFO("Failed to pin the rootfs for container \"%s\".", handler->name);
1186         }
1187
1188         if (!preserve_ns(saved_ns_fd, preserve_mask, getpid()))
1189                 goto out_delete_net;
1190
1191         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
1192                 goto out_delete_net;
1193
1194         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
1195                 if (pipe(netpipepair) < 0) {
1196                         SYSERROR("Failed to create pipe.");
1197                         goto out_delete_net;
1198                 }
1199                 /* Store netpipe in the global var for do_start's use. */
1200                 netpipe = netpipepair[0];
1201         }
1202
1203         /* Create a process in a new set of namespaces. */
1204         flags = handler->clone_flags;
1205         if (handler->clone_flags & CLONE_NEWUSER) {
1206                 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1207                  * clone a new user namespace first and only later unshare our
1208                  * network namespace to ensure that network devices ownership is
1209                  * set up correctly.
1210                  */
1211                 flags &= ~CLONE_NEWNET;
1212         }
1213         handler->pid = lxc_clone(do_start, handler, flags);
1214         if (handler->pid < 0) {
1215                 SYSERROR("Failed to clone a new set of namespaces.");
1216                 goto out_delete_net;
1217         }
1218         for (i = 0; i < LXC_NS_MAX; i++)
1219                 if (flags & ns_info[i].clone_flag)
1220                         INFO("Cloned %s.", ns_info[i].flag_name);
1221
1222         if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid))
1223                 INFO("Failed to preserve namespace for lxc.hook.stop.");
1224
1225         if (attach_ns(saved_ns_fd))
1226                 WARN("Failed to restore saved namespaces.");
1227
1228         lxc_sync_fini_child(handler);
1229
1230         /* Map the container uids. The container became an invalid userid the
1231          * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1232          * anything immediately, but allows the container to setuid(0) (0 being
1233          * mapped to something else on the host.) later to become a valid uid
1234          * again.
1235          */
1236         if (wants_to_map_ids && lxc_map_ids(id_map, handler->pid)) {
1237                 ERROR("Failed to set up id mapping.");
1238                 goto out_delete_net;
1239         }
1240
1241         if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) {
1242                 failed_before_rename = 1;
1243                 goto out_delete_net;
1244         }
1245
1246         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) {
1247                 failed_before_rename = 1;
1248                 goto out_delete_net;
1249         }
1250
1251         if (!cgroup_create_legacy(handler)) {
1252                 ERROR("Failed to setup legacy cgroups for container \"%s\".", name);
1253                 goto out_delete_net;
1254         }
1255         if (!cgroup_setup_limits(handler, false)) {
1256                 ERROR("Failed to setup cgroup limits for container \"%s\".", name);
1257                 goto out_delete_net;
1258         }
1259
1260         if (!cgroup_enter(handler))
1261                 goto out_delete_net;
1262
1263         if (!cgroup_chown(handler))
1264                 goto out_delete_net;
1265
1266         if (failed_before_rename)
1267                 goto out_delete_net;
1268
1269         /* Create the network configuration. */
1270         if (handler->clone_flags & CLONE_NEWNET) {
1271                 if (lxc_assign_network(handler->lxcpath, handler->name,
1272                                        &handler->conf->network, handler->pid)) {
1273                         ERROR("Failed to create the configured network.");
1274                         goto out_delete_net;
1275                 }
1276         }
1277
1278         if (netpipe != -1) {
1279                 struct lxc_list *iterator;
1280                 struct lxc_netdev *netdev;
1281
1282                 close(netpipe);
1283                 lxc_list_for_each(iterator, &handler->conf->network) {
1284                         netdev = iterator->elem;
1285                         if (netdev->type != LXC_NET_VETH)
1286                                 continue;
1287                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1288                                 ERROR("Error writing veth name to container.");
1289                                 goto out_delete_net;
1290                         }
1291                 }
1292                 close(netpipepair[1]);
1293         }
1294
1295         /* Tell the child to continue its initialization. We'll get
1296          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1297          */
1298         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1299                 goto out_delete_net;
1300
1301         if (!lxc_list_empty(&handler->conf->limits) && setup_resource_limits(&handler->conf->limits, handler->pid)) {
1302                 ERROR("failed to setup resource limits for '%s'", name);
1303                 goto out_delete_net;
1304         }
1305
1306         if (!cgroup_setup_limits(handler, true)) {
1307                 ERROR("Failed to setup the devices cgroup for container \"%s\".", name);
1308                 goto out_delete_net;
1309         }
1310
1311         cgroup_disconnect();
1312         cgroups_connected = false;
1313
1314         /* Read tty fds allocated by child. */
1315         if (lxc_recv_ttys_from_child(handler) < 0) {
1316                 ERROR("Failed to receive tty info from child process.");
1317                 goto out_delete_net;
1318         }
1319
1320         /* Tell the child to complete its initialization and wait for it to exec
1321          * or return an error. (The child will never return
1322          * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing
1323          * lxc_sync_barrier_child to return success, or return a different
1324          * value, causing us to error out).
1325          */
1326         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1327                 return -1;
1328
1329         if (handler->ops->post_start(handler, handler->data))
1330                 goto out_abort;
1331
1332         if (lxc_set_state(name, handler, RUNNING)) {
1333                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name,
1334                       lxc_state2str(RUNNING));
1335                 goto out_abort;
1336         }
1337
1338         lxc_sync_fini(handler);
1339         handler->netnsfd = lxc_preserve_ns(handler->pid, "net");
1340
1341         return 0;
1342
1343 out_delete_net:
1344         if (cgroups_connected)
1345                 cgroup_disconnect();
1346         if (handler->clone_flags & CLONE_NEWNET)
1347                 lxc_delete_network(handler);
1348 out_abort:
1349         lxc_abort(name, handler);
1350         lxc_sync_fini(handler);
1351         if (handler->pinfd >= 0) {
1352                 close(handler->pinfd);
1353                 handler->pinfd = -1;
1354         }
1355
1356         return -1;
1357 }
1358
1359 int __lxc_start(const char *name, struct lxc_handler *handler,
1360                 struct lxc_operations* ops, void *data, const char *lxcpath,
1361                 bool backgrounded)
1362 {
1363         int status;
1364         int err = -1;
1365         bool removed_all_netdevs = true;
1366         struct lxc_conf *conf = handler->conf;
1367
1368         if (lxc_init(name, handler) < 0) {
1369                 ERROR("Failed to initialize container \"%s\".", name);
1370                 return -1;
1371         }
1372         handler->ops = ops;
1373         handler->data = data;
1374         handler->backgrounded = backgrounded;
1375         handler->netnsfd = -1;
1376
1377         if (must_drop_cap_sys_boot(handler->conf)) {
1378                 #if HAVE_LIBCAP
1379                 DEBUG("Dropping CAP_SYS_BOOT capability.");
1380                 #else
1381                 DEBUG("Not dropping CAP_SYS_BOOT capability as capabilities aren't supported.");
1382                 #endif
1383         } else {
1384                 DEBUG("Not dropping CAP_SYS_BOOT or watching utmp.");
1385                 handler->conf->need_utmp_watch = 0;
1386         }
1387
1388         if (!attach_block_device(handler->conf)) {
1389                 ERROR("Failed to attach block device.");
1390                 goto out_fini_nonet;
1391         }
1392
1393         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1394                 /* If the backing store is a device, mount it here and now. */
1395                 if (rootfs_is_blockdev(conf)) {
1396                         if (unshare(CLONE_NEWNS) < 0) {
1397                                 ERROR("Failed to unshare CLONE_NEWNS.");
1398                                 goto out_fini_nonet;
1399                         }
1400                         INFO("Unshared CLONE_NEWNS.");
1401
1402                         remount_all_slave();
1403                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1404                                 ERROR("Error setting up rootfs mount as root before spawn.");
1405                                 goto out_fini_nonet;
1406                         }
1407                         INFO("Set up container rootfs as host root.");
1408                 }
1409         }
1410
1411         err = lxc_spawn(handler);
1412         if (err) {
1413                 ERROR("Failed to spawn container \"%s\".", name);
1414                 goto out_detach_blockdev;
1415         }
1416
1417         handler->conf->reboot = 0;
1418
1419         err = lxc_poll(name, handler);
1420         if (err) {
1421                 ERROR("LXC mainloop exited with error: %d.", err);
1422                 if (handler->netnsfd >= 0) {
1423                         close(handler->netnsfd);
1424                         handler->netnsfd = -1;
1425                 }
1426                 goto out_abort;
1427         }
1428
1429         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1430                 continue;
1431
1432         /* If the child process exited but was not signaled, it didn't call
1433          * reboot. This should mean it was an lxc-execute which simply exited.
1434          * In any case, treat it as a 'halt'.
1435          */
1436         if (WIFSIGNALED(status)) {
1437                 switch(WTERMSIG(status)) {
1438                 case SIGINT: /* halt */
1439                         DEBUG("Container \"%s\" is halting.", name);
1440                         break;
1441                 case SIGHUP: /* reboot */
1442                         DEBUG("Container \"%s\" is rebooting.", name);
1443                         handler->conf->reboot = 1;
1444                         break;
1445                 case SIGSYS: /* seccomp */
1446                         DEBUG("Container \"%s\" violated its seccomp policy.", name);
1447                         break;
1448                 default:
1449                         DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status));
1450                         break;
1451                 }
1452         }
1453
1454         DEBUG("Pushing physical nics back to host namespace");
1455         lxc_restore_phys_nics_to_netns(handler->netnsfd, handler->conf);
1456
1457         DEBUG("Tearing down virtual network devices used by container \"%s\".", name);
1458         removed_all_netdevs = lxc_delete_network(handler);
1459
1460         if (handler->pinfd >= 0) {
1461                 close(handler->pinfd);
1462                 handler->pinfd = -1;
1463         }
1464
1465         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1466         err =  lxc_error_set_and_log(handler->pid, status);
1467 out_fini:
1468         if (!removed_all_netdevs) {
1469                 DEBUG("Failed tearing down network devices used by container. Trying again!");
1470                 removed_all_netdevs = lxc_delete_network(handler);
1471                 if (!removed_all_netdevs)
1472                         DEBUG("Failed tearing down network devices used by container. Not trying again!");
1473         }
1474
1475 out_detach_blockdev:
1476         detach_block_device(handler->conf);
1477
1478 out_fini_nonet:
1479         lxc_fini(name, handler);
1480         return err;
1481
1482 out_abort:
1483         lxc_abort(name, handler);
1484         goto out_fini;
1485 }
1486
1487 struct start_args {
1488         char *const *argv;
1489 };
1490
1491 static int start(struct lxc_handler *handler, void* data)
1492 {
1493         struct start_args *arg = data;
1494
1495         NOTICE("Exec'ing \"%s\".", arg->argv[0]);
1496
1497         execvp(arg->argv[0], arg->argv);
1498         SYSERROR("Failed to exec \"%s\".", arg->argv[0]);
1499         return 0;
1500 }
1501
1502 static int post_start(struct lxc_handler *handler, void* data)
1503 {
1504         struct start_args *arg = data;
1505
1506         NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid);
1507         return 0;
1508 }
1509
1510 static struct lxc_operations start_ops = {
1511         .start = start,
1512         .post_start = post_start
1513 };
1514
1515 int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler,
1516               const char *lxcpath, bool backgrounded)
1517 {
1518         struct start_args start_arg = {
1519                 .argv = argv,
1520         };
1521
1522         handler->conf->need_utmp_watch = 1;
1523         return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded);
1524 }
1525
1526 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1527                                             const char *name)
1528 {
1529         char destroy[MAXPATHLEN];
1530         bool bret = true;
1531         int ret = 0;
1532         struct lxc_container *c;
1533         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1534                 bret = do_destroy_container(handler->conf);
1535                 if (!bret) {
1536                         ERROR("Error destroying rootfs for container \"%s\".", name);
1537                         return;
1538                 }
1539         }
1540         INFO("Destroyed rootfs for container \"%s\".", name);
1541
1542         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1543         if (ret < 0 || ret >= MAXPATHLEN) {
1544                 ERROR("Error destroying directory for container \"%s\".", name);
1545                 return;
1546         }
1547
1548         c = lxc_container_new(name, handler->lxcpath);
1549         if (c) {
1550                 if (container_disk_lock(c)) {
1551                         INFO("Could not update lxc_snapshots file.");
1552                         lxc_container_put(c);
1553                 } else {
1554                         mod_all_rdeps(c, false);
1555                         container_disk_unlock(c);
1556                         lxc_container_put(c);
1557                 }
1558         }
1559
1560         if (am_unpriv())
1561                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper,
1562                                     destroy, "lxc_rmdir_onedev_wrapper");
1563         else
1564                 ret = lxc_rmdir_onedev(destroy, NULL);
1565
1566         if (ret < 0) {
1567                 ERROR("Error destroying directory for container \"%s\".", name);
1568                 return;
1569         }
1570         INFO("Destroyed directory for container \"%s\".", name);
1571 }
1572
1573 static int lxc_rmdir_onedev_wrapper(void *data)
1574 {
1575         char *arg = (char *) data;
1576         return lxc_rmdir_onedev(arg, NULL);
1577 }
1578
1579 static bool do_destroy_container(struct lxc_conf *conf) {
1580         if (am_unpriv()) {
1581                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf,
1582                                   "bdev_destroy_wrapper") < 0)
1583                         return false;
1584                 return true;
1585         }
1586         return bdev_destroy(conf);
1587 }