src/lxc/start.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #include "config.h"
  25
  26 #include <stdio.h>
  27 #include <string.h>
  28 #include <stdlib.h>
  29 #include <dirent.h>
  30 #include <errno.h>
  31 #include <unistd.h>
  32 #include <signal.h>
  33 #include <fcntl.h>
  34 #include <grp.h>
  35 #include <poll.h>
  36 #include <sys/param.h>
  37 #include <sys/file.h>
  38 #include <sys/mount.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/socket.h>
  42 #include <sys/prctl.h>
  43 #include <sys/types.h>
  44 #include <sys/wait.h>
  45 #include <sys/un.h>
  46 #include <sys/syscall.h>
  47
  48 #if HAVE_SYS_CAPABILITY_H
  49 #include <sys/capability.h>
  50 #endif
  51
  52 #if !HAVE_DECL_PR_CAPBSET_DROP
  53 #define PR_CAPBSET_DROP 24
  54 #endif
  55
  56 #include "start.h"
  57 #include "conf.h"
  58 #include "log.h"
  59 #include "cgroup.h"
  60 #include "error.h"
  61 #include "af_unix.h"
  62 #include "mainloop.h"
  63 #include "utils.h"
  64 #include "lxcutmp.h"
  65 #include "monitor.h"
  66 #include "commands.h"
  67 #include "console.h"
  68 #include "sync.h"
  69 #include "namespace.h"
  70 #include "lxcseccomp.h"
  71 #include "caps.h"
  72 #include "bdev/bdev.h"
  73 #include "lsm/lsm.h"
  74 #include "lxclock.h"
  75
  76 lxc_log_define(lxc_start, lxc);
  77
  78 const struct ns_info ns_info[LXC_NS_MAX] = {
  79         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  80         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  81         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  82         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  83         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  84         [LXC_NS_NET] = {"net", CLONE_NEWNET}
  85 };
  86
  87 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 == '/') p++;
 103                 while (p < e && *p != '/') p++;
 104                 saved = *p;
 105                 *p = '\0';
 106                 if (access(copy, X_OK)) {
 107                         SYSERROR("could not access %s.  Please grant it 'x' " \
 108                               "access, or add an ACL for the container root.",
 109                               copy);
 110                         return;
 111                 }
 112                 *p = saved;
 113         }
 114 }
 115
 116 static void close_ns(int ns_fd[LXC_NS_MAX]) {
 117         int i;
 118
 119         for (i = 0; i < LXC_NS_MAX; i++) {
 120                 if (ns_fd[i] > -1) {
 121                         close(ns_fd[i]);
 122                         ns_fd[i] = -1;
 123                 }
 124         }
 125 }
 126
 127 /*
 128  * preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
 129  * in clone_flags.
 130  * Return true on success, false on failure.  On failure, leave an error
 131  * message in *errmsg, which caller must free.
 132  */
 133 static
 134 bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid, char **errmsg) {
 135         int i, ret;
 136         char path[MAXPATHLEN];
 137
 138         for (i = 0; i < LXC_NS_MAX; i++)
 139                 ns_fd[i] = -1;
 140
 141         snprintf(path, MAXPATHLEN, "/proc/%d/ns", pid);
 142         if (access(path, X_OK)) {
 143                 if (asprintf(errmsg, "Kernel does not support setns.") == -1)
 144                         *errmsg = NULL;
 145                 return false;
 146         }
 147
 148         for (i = 0; i < LXC_NS_MAX; i++) {
 149                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 150                         continue;
 151                 snprintf(path, MAXPATHLEN, "/proc/%d/ns/%s", pid,
 152                          ns_info[i].proc_name);
 153                 ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC);
 154                 if (ns_fd[i] < 0)
 155                         goto error;
 156         }
 157
 158         return true;
 159
 160 error:
 161         if (errno == ENOENT) {
 162                 ret = asprintf(errmsg, "Kernel does not support setns for %s",
 163                         ns_info[i].proc_name);
 164         } else {
 165                 ret = asprintf(errmsg, "Failed to open %s: %s",
 166                         path, strerror(errno));
 167         }
 168         if (ret == -1)
 169                 *errmsg = NULL;
 170         close_ns(ns_fd);
 171         return false;
 172 }
 173
 174 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 175         int i;
 176
 177         for (i = 0; i < LXC_NS_MAX; i++) {
 178                 if (ns_fd[i] < 0)
 179                         continue;
 180
 181                 if (setns(ns_fd[i], 0) != 0)
 182                         goto error;
 183         }
 184         return 0;
 185
 186 error:
 187         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 188         return -1;
 189 }
 190
 191 static int match_fd(int fd)
 192 {
 193         return (fd == 0 || fd == 1 || fd == 2);
 194 }
 195
 196 /*
 197  * Check for any fds we need to close
 198  * * if fd_to_ignore != -1, then if we find that fd open we will ignore it.
 199  * * By default we warn about open fds we find.
 200  * * If closeall is true, we will close open fds.
 201  * * If lxc-start was passed "-C", then conf->close_all_fds will be true,
 202  *     in which case we also close all open fds.
 203  * * A daemonized container will always pass closeall=true.
 204  */
 205 int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore)
 206 {
 207         struct dirent dirent, *direntp;
 208         int fd, fddir;
 209         DIR *dir;
 210
 211         if (conf && conf->close_all_fds)
 212                 closeall = true;
 213
 214 restart:
 215         dir = opendir("/proc/self/fd");
 216         if (!dir) {
 217                 WARN("failed to open directory: %m");
 218                 return -1;
 219         }
 220
 221         fddir = dirfd(dir);
 222
 223         while (!readdir_r(dir, &dirent, &direntp)) {
 224                 if (!direntp)
 225                         break;
 226
 227                 if (!strcmp(direntp->d_name, "."))
 228                         continue;
 229
 230                 if (!strcmp(direntp->d_name, ".."))
 231                         continue;
 232
 233                 fd = atoi(direntp->d_name);
 234
 235                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 236                         continue;
 237
 238                 if (current_config && fd == current_config->logfd)
 239                         continue;
 240
 241                 if (match_fd(fd))
 242                         continue;
 243
 244                 if (closeall) {
 245                         close(fd);
 246                         closedir(dir);
 247                         INFO("closed inherited fd %d", fd);
 248                         goto restart;
 249                 }
 250                 WARN("inherited fd %d", fd);
 251         }
 252
 253         closedir(dir); /* cannot fail */
 254         return 0;
 255 }
 256
 257 static int setup_signal_fd(sigset_t *oldmask)
 258 {
 259         sigset_t mask;
 260         int fd;
 261
 262         /* Block everything except serious error signals */
 263         if (sigfillset(&mask) ||
 264             sigdelset(&mask, SIGILL) ||
 265             sigdelset(&mask, SIGSEGV) ||
 266             sigdelset(&mask, SIGBUS) ||
 267             sigdelset(&mask, SIGWINCH) ||
 268             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 269                 SYSERROR("failed to set signal mask");
 270                 return -1;
 271         }
 272
 273         fd = signalfd(-1, &mask, 0);
 274         if (fd < 0) {
 275                 SYSERROR("failed to create the signal fd");
 276                 return -1;
 277         }
 278
 279         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 280                 SYSERROR("failed to set sigfd to close-on-exec");
 281                 close(fd);
 282                 return -1;
 283         }
 284
 285         DEBUG("sigchild handler set");
 286
 287         return fd;
 288 }
 289
 290 static int signal_handler(int fd, uint32_t events, void *data,
 291                            struct lxc_epoll_descr *descr)
 292 {
 293         struct signalfd_siginfo siginfo;
 294         siginfo_t info;
 295         int ret;
 296         pid_t *pid = data;
 297         bool init_died = false;
 298
 299         ret = read(fd, &siginfo, sizeof(siginfo));
 300         if (ret < 0) {
 301                 ERROR("failed to read signal info");
 302                 return -1;
 303         }
 304
 305         if (ret != sizeof(siginfo)) {
 306                 ERROR("unexpected siginfo size");
 307                 return -1;
 308         }
 309
 310         // check whether init is running
 311         info.si_pid = 0;
 312         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 313         if (ret == 0 && info.si_pid == *pid) {
 314                 init_died = true;
 315         }
 316
 317         if (siginfo.ssi_signo != SIGCHLD) {
 318                 kill(*pid, siginfo.ssi_signo);
 319                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 320                 return init_died ? 1 : 0;
 321         }
 322
 323         if (siginfo.ssi_code == CLD_STOPPED ||
 324             siginfo.ssi_code == CLD_CONTINUED) {
 325                 INFO("container init process was stopped/continued");
 326                 return init_died ? 1 : 0;
 327         }
 328
 329         /* more robustness, protect ourself from a SIGCHLD sent
 330          * by a process different from the container init
 331          */
 332         if (siginfo.ssi_pid != *pid) {
 333                 WARN("invalid pid for SIGCHLD");
 334                 return init_died ? 1 : 0;
 335         }
 336
 337         DEBUG("container init process exited");
 338         return 1;
 339 }
 340
 341 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 342 {
 343         handler->state = state;
 344         lxc_monitor_send_state(name, state, handler->lxcpath);
 345         return 0;
 346 }
 347
 348 int lxc_poll(const char *name, struct lxc_handler *handler)
 349 {
 350         int sigfd = handler->sigfd;
 351         int pid = handler->pid;
 352         struct lxc_epoll_descr descr;
 353
 354         if (lxc_mainloop_open(&descr)) {
 355                 ERROR("failed to create mainloop");
 356                 goto out_sigfd;
 357         }
 358
 359         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 360                 ERROR("failed to add handler for the signal");
 361                 goto out_mainloop_open;
 362         }
 363
 364         if (lxc_console_mainloop_add(&descr, handler)) {
 365                 ERROR("failed to add console handler to mainloop");
 366                 goto out_mainloop_open;
 367         }
 368
 369         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 370                 ERROR("failed to add command handler to mainloop");
 371                 goto out_mainloop_open;
 372         }
 373
 374         if (handler->conf->need_utmp_watch) {
 375                 #if HAVE_SYS_CAPABILITY_H
 376                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 377                         ERROR("failed to add utmp handler to mainloop");
 378                         goto out_mainloop_open;
 379                 }
 380                 #else
 381                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
 382                 #endif
 383         }
 384
 385         return lxc_mainloop(&descr, -1);
 386
 387 out_mainloop_open:
 388         lxc_mainloop_close(&descr);
 389 out_sigfd:
 390         close(sigfd);
 391         return -1;
 392 }
 393
 394 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 395 {
 396         int i;
 397         struct lxc_handler *handler;
 398
 399         handler = malloc(sizeof(*handler));
 400         if (!handler)
 401                 return NULL;
 402
 403         memset(handler, 0, sizeof(*handler));
 404
 405         handler->ttysock[0] = handler->ttysock[1] = -1;
 406         handler->conf = conf;
 407         handler->lxcpath = lxcpath;
 408         handler->pinfd = -1;
 409
 410         for (i = 0; i < LXC_NS_MAX; i++)
 411                 handler->nsfd[i] = -1;
 412
 413         lsm_init();
 414
 415         handler->name = strdup(name);
 416         if (!handler->name) {
 417                 ERROR("failed to allocate memory");
 418                 goto out_free;
 419         }
 420
 421         if (lxc_cmd_init(name, handler, lxcpath))
 422                 goto out_free_name;
 423
 424         if (lxc_read_seccomp_config(conf) != 0) {
 425                 ERROR("failed loading seccomp policy");
 426                 goto out_close_maincmd_fd;
 427         }
 428
 429         /* Begin by setting the state to STARTING */
 430         if (lxc_set_state(name, handler, STARTING)) {
 431                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 432                 goto out_close_maincmd_fd;
 433         }
 434
 435         /* Start of environment variable setup for hooks */
 436         if (name && setenv("LXC_NAME", name, 1)) {
 437                 SYSERROR("failed to set environment variable for container name");
 438         }
 439         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 440                 SYSERROR("failed to set environment variable for config path");
 441         }
 442         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 443                 SYSERROR("failed to set environment variable for rootfs mount");
 444         }
 445         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 446                 SYSERROR("failed to set environment variable for rootfs mount");
 447         }
 448         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 449                 SYSERROR("failed to set environment variable for console path");
 450         }
 451         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 452                 SYSERROR("failed to set environment variable for console log");
 453         }
 454         if (setenv("LXC_CGNS_AWARE", "1", 1)) {
 455                 SYSERROR("failed to set LXC_CGNS_AWARE environment variable");
 456         }
 457         /* End of environment variable setup for hooks */
 458
 459         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 460                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 461                 goto out_aborting;
 462         }
 463
 464         /* the signal fd has to be created before forking otherwise
 465          * if the child process exits before we setup the signal fd,
 466          * the event will be lost and the command will be stuck */
 467         handler->sigfd = setup_signal_fd(&handler->oldmask);
 468         if (handler->sigfd < 0) {
 469                 ERROR("failed to set sigchild fd handler");
 470                 goto out_delete_tty;
 471         }
 472
 473         /* do this after setting up signals since it might unblock SIGWINCH */
 474         if (lxc_console_create(conf)) {
 475                 ERROR("failed to create console");
 476                 goto out_restore_sigmask;
 477         }
 478
 479         if (ttys_shift_ids(conf) < 0) {
 480                 ERROR("Failed to shift tty into container");
 481                 goto out_restore_sigmask;
 482         }
 483
 484         INFO("'%s' is initialized", name);
 485         return handler;
 486
 487 out_restore_sigmask:
 488         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 489 out_delete_tty:
 490         lxc_delete_tty(&conf->tty_info);
 491 out_aborting:
 492         lxc_set_state(name, handler, ABORTING);
 493 out_close_maincmd_fd:
 494         close(conf->maincmd_fd);
 495         conf->maincmd_fd = -1;
 496 out_free_name:
 497         free(handler->name);
 498         handler->name = NULL;
 499 out_free:
 500         free(handler);
 501         return NULL;
 502 }
 503
 504 void lxc_fini(const char *name, struct lxc_handler *handler)
 505 {
 506         int i, rc;
 507         pid_t self = getpid();
 508         char *namespaces[LXC_NS_MAX+1];
 509         size_t namespace_count = 0;
 510
 511         /* The STOPPING state is there for future cleanup code
 512          * which can take awhile
 513          */
 514         lxc_set_state(name, handler, STOPPING);
 515
 516         for (i = 0; i < LXC_NS_MAX; i++) {
 517                 if (handler->nsfd[i] != -1) {
 518                         rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d",
 519                                       ns_info[i].proc_name, self, handler->nsfd[i]);
 520                         if (rc == -1) {
 521                                 SYSERROR("failed to allocate memory");
 522                                 break;
 523                         }
 524                         ++namespace_count;
 525                 }
 526         }
 527         namespaces[namespace_count] = NULL;
 528
 529         if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) {
 530                 SYSERROR("failed to set environment variable for stop target");
 531         }
 532         if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) {
 533                 SYSERROR("failed to set environment variable for stop target");
 534         }
 535
 536         if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces))
 537                 ERROR("failed to run stop hooks for container '%s'.", name);
 538
 539         while (namespace_count--)
 540                 free(namespaces[namespace_count]);
 541         for (i = 0; i < LXC_NS_MAX; i++) {
 542                 if (handler->nsfd[i] != -1) {
 543                         close(handler->nsfd[i]);
 544                         handler->nsfd[i] = -1;
 545                 }
 546         }
 547         lxc_set_state(name, handler, STOPPED);
 548
 549         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 550                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 551
 552         /* reset mask set by setup_signal_fd */
 553         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 554                 WARN("failed to restore sigprocmask");
 555
 556         lxc_console_delete(&handler->conf->console);
 557         lxc_delete_tty(&handler->conf->tty_info);
 558         close(handler->conf->maincmd_fd);
 559         handler->conf->maincmd_fd = -1;
 560         free(handler->name);
 561         if (handler->ttysock[0] != -1) {
 562                 close(handler->ttysock[0]);
 563                 close(handler->ttysock[1]);
 564         }
 565
 566         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1)
 567                 lxc_destroy_container_on_signal(handler, name);
 568
 569         cgroup_destroy(handler);
 570         free(handler);
 571 }
 572
 573 void lxc_abort(const char *name, struct lxc_handler *handler)
 574 {
 575         int ret, status;
 576
 577         lxc_set_state(name, handler, ABORTING);
 578         if (handler->pid > 0)
 579                 kill(handler->pid, SIGKILL);
 580         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 581 }
 582
 583 #include <sys/reboot.h>
 584 #include <linux/reboot.h>
 585
 586 /*
 587  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 588  * in a child pid namespace if container reboot support exists.
 589  * Otherwise, it will either succeed or return -EPERM.
 590  */
 591 static int container_reboot_supported(void *arg)
 592 {
 593         int *cmd = arg;
 594         int ret;
 595
 596         ret = reboot(*cmd);
 597         if (ret == -1 && errno == EINVAL)
 598                 return 1;
 599         return 0;
 600 }
 601
 602 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 603 {
 604         FILE *f;
 605         int ret, cmd, v, flags;
 606         long stack_size = 4096;
 607         void *stack = alloca(stack_size);
 608         int status;
 609         pid_t pid;
 610
 611         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 612         if (!f) {
 613                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 614                 return 1;
 615         }
 616
 617         ret = fscanf(f, "%d", &v);
 618         fclose(f);
 619         if (ret != 1) {
 620                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 621                 return 1;
 622         }
 623         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 624
 625         flags = CLONE_NEWPID | SIGCHLD;
 626         if (!lxc_list_empty(&conf->id_map))
 627                 flags |= CLONE_NEWUSER;
 628
 629 #ifdef __ia64__
 630         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 631 #else
 632         stack += stack_size;
 633         pid = clone(container_reboot_supported, stack, flags, &cmd);
 634 #endif
 635         if (pid < 0) {
 636                 if (flags & CLONE_NEWUSER)
 637                         ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno));
 638                 else
 639                         ERROR("failed to clone (%#x): %s", flags, strerror(errno));
 640                 return -1;
 641         }
 642         if (wait(&status) < 0) {
 643                 SYSERROR("unexpected wait error: %m");
 644                 return -1;
 645         }
 646
 647         if (WEXITSTATUS(status) != 1)
 648                 return 1;
 649
 650         return 0;
 651 }
 652
 653 /*
 654  * netpipe is used in the unprivileged case to transfer the ifindexes
 655  * from parent to child
 656  */
 657 static int netpipe = -1;
 658
 659 static inline int count_veths(struct lxc_list *network)
 660 {
 661         struct lxc_list *iterator;
 662         struct lxc_netdev *netdev;
 663         int count = 0;
 664
 665         lxc_list_for_each(iterator, network) {
 666                 netdev = iterator->elem;
 667                 if (netdev->type != LXC_NET_VETH)
 668                         continue;
 669                 count++;
 670         }
 671         return count;
 672 }
 673
 674 static int read_unpriv_netifindex(struct lxc_list *network)
 675 {
 676         struct lxc_list *iterator;
 677         struct lxc_netdev *netdev;
 678
 679         if (netpipe == -1)
 680                 return 0;
 681         lxc_list_for_each(iterator, network) {
 682                 netdev = iterator->elem;
 683                 if (netdev->type != LXC_NET_VETH)
 684                         continue;
 685                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 686                         ERROR("Out of memory");
 687                         close(netpipe);
 688                         return -1;
 689                 }
 690                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 691                         close(netpipe);
 692                         return -1;
 693                 }
 694         }
 695         close(netpipe);
 696         return 0;
 697 }
 698
 699 static int do_start(void *data)
 700 {
 701         struct lxc_list *iterator;
 702         struct lxc_handler *handler = data;
 703
 704         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 705                 SYSERROR("failed to set sigprocmask");
 706                 return -1;
 707         }
 708
 709         /* This prctl must be before the synchro, so if the parent
 710          * dies before we set the parent death signal, we will detect
 711          * its death with the synchro right after, otherwise we have
 712          * a window where the parent can exit before we set the pdeath
 713          * signal leading to a unsupervized container.
 714          */
 715         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 716                 SYSERROR("failed to set pdeath signal");
 717                 return -1;
 718         }
 719
 720         lxc_sync_fini_parent(handler);
 721
 722         /* don't leak the pinfd to the container */
 723         if (handler->pinfd >= 0) {
 724                 close(handler->pinfd);
 725         }
 726
 727         /* Tell the parent task it can begin to configure the
 728          * container and wait for it to finish
 729          */
 730         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 731                 return -1;
 732
 733         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 734                 goto out_warn_father;
 735
 736         /*
 737          * if we are in a new user namespace, become root there to have
 738          * privilege over our namespace. When using lxc-execute we default to root,
 739          * but this can be overriden using the lxc.init_uid and lxc.init_gid
 740          * configuration options.
 741          */
 742         if (!lxc_list_empty(&handler->conf->id_map)) {
 743                 gid_t new_gid = 0;
 744                 if (handler->conf->is_execute && handler->conf->init_gid)
 745                         new_gid = handler->conf->init_gid;
 746
 747                 uid_t new_uid = 0;
 748                 if (handler->conf->is_execute && handler->conf->init_uid)
 749                         new_uid = handler->conf->init_uid;
 750
 751                 NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid);
 752                 if (setgid(new_gid)) {
 753                         SYSERROR("setgid");
 754                         goto out_warn_father;
 755                 }
 756                 if (setuid(new_uid)) {
 757                         SYSERROR("setuid");
 758                         goto out_warn_father;
 759                 }
 760                 if (setgroups(0, NULL)) {
 761                         SYSERROR("setgroups");
 762                         goto out_warn_father;
 763                 }
 764         }
 765
 766         if (access(handler->lxcpath, X_OK)) {
 767                 print_top_failing_dir(handler->lxcpath);
 768                 goto out_warn_father;
 769         }
 770
 771         #if HAVE_SYS_CAPABILITY_H
 772         if (handler->conf->need_utmp_watch) {
 773                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 774                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 775                         goto out_warn_father;
 776                 }
 777                 DEBUG("Dropped cap_sys_boot");
 778         }
 779         #endif
 780
 781         /* Setup the container, ip, names, utsname, ... */
 782         if (lxc_setup(handler)) {
 783                 ERROR("failed to setup the container");
 784                 goto out_warn_father;
 785         }
 786
 787         /* ask father to setup cgroups and wait for him to finish */
 788         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 789                 return -1;
 790
 791         /* Set the label to change to when we exec(2) the container's init */
 792         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 793                 goto out_warn_father;
 794
 795         /* Some init's such as busybox will set sane tty settings on stdin,
 796          * stdout, stderr which it thinks is the console. We already set them
 797          * the way we wanted on the real terminal, and we want init to do its
 798          * setup on its console ie. the pty allocated in lxc_console_create()
 799          * so make sure that that pty is stdin,stdout,stderr.
 800          */
 801         if (lxc_console_set_stdfds(handler->conf->console.slave) < 0)
 802                 goto out_warn_father;
 803
 804         /* If we mounted a temporary proc, then unmount it now */
 805         tmp_proc_unmount(handler->conf);
 806
 807         if (lxc_seccomp_load(handler->conf) != 0)
 808                 goto out_warn_father;
 809
 810         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 811                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 812                 goto out_warn_father;
 813         }
 814
 815         /* The clearenv() and putenv() calls have been moved here
 816          * to allow us to use environment variables passed to the various
 817          * hooks, such as the start hook above.  Not all of the
 818          * variables like CONFIG_PATH or ROOTFS are valid in this
 819          * context but others are. */
 820         if (clearenv()) {
 821                 SYSERROR("failed to clear environment");
 822                 /* don't error out though */
 823         }
 824
 825         lxc_list_for_each(iterator, &handler->conf->environment) {
 826                 if (putenv((char *)iterator->elem)) {
 827                         SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem);
 828                         goto out_warn_father;
 829                 }
 830         }
 831
 832         if (putenv("container=lxc")) {
 833                 SYSERROR("failed to set environment variable 'container=lxc'");
 834                 goto out_warn_father;
 835         }
 836
 837         if (handler->conf->pty_names) {
 838                 if (putenv(handler->conf->pty_names)) {
 839                         SYSERROR("failed to set environment variable for container ptys");
 840                         goto out_warn_father;
 841                 }
 842         }
 843
 844         close(handler->sigfd);
 845
 846         if (handler->backgrounded && null_stdfds() < 0)
 847                 goto out_warn_father;
 848
 849         if (cgns_supported() && unshare(CLONE_NEWCGROUP) != 0) {
 850                 SYSERROR("Failed to unshare cgroup namespace");
 851                 goto out_warn_father;
 852         }
 853
 854         setsid();
 855
 856         /* after this call, we are in error because this
 857          * ops should not return as it execs */
 858         handler->ops->start(handler, handler->data);
 859
 860 out_warn_father:
 861         /* we want the parent to know something went wrong, so any
 862          * value other than what it expects is ok. */
 863         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 864         return -1;
 865 }
 866
 867 static int save_phys_nics(struct lxc_conf *conf)
 868 {
 869         struct lxc_list *iterator;
 870         int am_root = (getuid() == 0);
 871
 872         if (!am_root)
 873                 return 0;
 874
 875         lxc_list_for_each(iterator, &conf->network) {
 876                 struct lxc_netdev *netdev = iterator->elem;
 877
 878                 if (netdev->type != LXC_NET_PHYS)
 879                         continue;
 880                 conf->saved_nics = realloc(conf->saved_nics,
 881                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 882                 if (!conf->saved_nics) {
 883                         SYSERROR("failed to allocate memory");
 884                         return -1;
 885                 }
 886                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 887                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 888                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 889                         SYSERROR("failed to allocate memory");
 890                         return -1;
 891                 }
 892                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 893                         conf->saved_nics[conf->num_savednics].ifindex,
 894                         conf->saved_nics[conf->num_savednics].orig_name);
 895                 conf->num_savednics++;
 896         }
 897
 898         return 0;
 899 }
 900
 901 static int recv_fd(int sock, int *fd)
 902 {
 903         if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) {
 904                 SYSERROR("Error receiving tty fd from child");
 905                 return -1;
 906         }
 907         if (*fd == -1)
 908                 return -1;
 909         return 0;
 910 }
 911
 912 static int recv_ttys_from_child(struct lxc_handler *handler)
 913 {
 914         struct lxc_conf *conf = handler->conf;
 915         int i, sock = handler->ttysock[1];
 916         struct lxc_tty_info *tty_info = &conf->tty_info;
 917
 918         if (!conf->tty)
 919                 return 0;
 920
 921         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty);
 922         if (!tty_info->pty_info) {
 923                 SYSERROR("failed to allocate pty_info");
 924                 return -1;
 925         }
 926
 927         for (i = 0; i < conf->tty; i++) {
 928                 struct lxc_pty_info *pty_info = &tty_info->pty_info[i];
 929                 pty_info->busy = 0;
 930                 if (recv_fd(sock, &pty_info->slave) < 0 ||
 931                                 recv_fd(sock, &pty_info->master) < 0) {
 932                         ERROR("Error receiving tty info from child");
 933                         return -1;
 934                 }
 935         }
 936         tty_info->nbtty = conf->tty;
 937
 938         return 0;
 939 }
 940
 941 void resolve_clone_flags(struct lxc_handler *handler)
 942 {
 943         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
 944
 945         if (!lxc_list_empty(&handler->conf->id_map)) {
 946                 INFO("Cloning a new user namespace");
 947                 handler->clone_flags |= CLONE_NEWUSER;
 948         }
 949
 950         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 951                 if (!lxc_requests_empty_network(handler))
 952                         handler->clone_flags |= CLONE_NEWNET;
 953         } else {
 954                 INFO("Inheriting a net namespace");
 955         }
 956
 957         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 958                 handler->clone_flags |= CLONE_NEWIPC;
 959         } else {
 960                 INFO("Inheriting an IPC namespace");
 961         }
 962
 963         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 964                 handler->clone_flags |= CLONE_NEWUTS;
 965         } else {
 966                 INFO("Inheriting a UTS namespace");
 967         }
 968 }
 969
 970 static int lxc_spawn(struct lxc_handler *handler)
 971 {
 972         int failed_before_rename = 0;
 973         const char *name = handler->name;
 974         char *errmsg = NULL;
 975         bool cgroups_connected = false;
 976         int saved_ns_fd[LXC_NS_MAX];
 977         int preserve_mask = 0, i;
 978         int netpipepair[2], nveths;
 979
 980         netpipe = -1;
 981
 982         for (i = 0; i < LXC_NS_MAX; i++)
 983                 if (handler->conf->inherit_ns_fd[i] != -1)
 984                         preserve_mask |= ns_info[i].clone_flag;
 985
 986         if (lxc_sync_init(handler))
 987                 return -1;
 988
 989         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
 990                 lxc_sync_fini(handler);
 991                 return -1;
 992         }
 993
 994         resolve_clone_flags(handler);
 995
 996         if (handler->clone_flags & CLONE_NEWNET) {
 997                 if (!lxc_list_empty(&handler->conf->network)) {
 998
 999                         /* Find gateway addresses from the link device, which is
1000                          * no longer accessible inside the container. Do this
1001                          * before creating network interfaces, since goto
1002                          * out_delete_net does not work before lxc_clone. */
1003                         if (lxc_find_gateway_addresses(handler)) {
1004                                 ERROR("failed to find gateway addresses");
1005                                 lxc_sync_fini(handler);
1006                                 return -1;
1007                         }
1008
1009                         /* that should be done before the clone because we will
1010                          * fill the netdev index and use them in the child
1011                          */
1012                         if (lxc_create_network(handler)) {
1013                                 ERROR("failed to create the network");
1014                                 lxc_sync_fini(handler);
1015                                 return -1;
1016                         }
1017                 }
1018
1019                 if (save_phys_nics(handler->conf)) {
1020                         ERROR("failed to save physical nic info");
1021                         goto out_abort;
1022                 }
1023         }
1024
1025         if (!cgroup_init(handler)) {
1026                 ERROR("failed initializing cgroup support");
1027                 goto out_delete_net;
1028         }
1029
1030         cgroups_connected = true;
1031
1032         if (!cgroup_create(handler)) {
1033                 ERROR("failed creating cgroups");
1034                 goto out_delete_net;
1035         }
1036
1037         /*
1038          * if the rootfs is not a blockdev, prevent the container from
1039          * marking it readonly.
1040          *
1041          * if the container is unprivileged then skip rootfs pinning
1042          */
1043         if (lxc_list_empty(&handler->conf->id_map)) {
1044                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
1045                 if (handler->pinfd == -1)
1046                         INFO("failed to pin the container's rootfs");
1047         }
1048
1049         if (!preserve_ns(saved_ns_fd, preserve_mask, getpid(), &errmsg)) {
1050                 SYSERROR("Failed to preserve requested namespaces: %s",
1051                         errmsg ? errmsg : "(Out of memory)");
1052                 free(errmsg);
1053                 goto out_delete_net;
1054         }
1055         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
1056                 goto out_delete_net;
1057
1058         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
1059                 if (pipe(netpipepair) < 0) {
1060                         SYSERROR("Error creating pipe");
1061                         goto out_delete_net;
1062                 }
1063                 /* store netpipe in the global var for do_start's use */
1064                 netpipe = netpipepair[0];
1065         }
1066
1067         /* Create a process in a new set of namespaces */
1068         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
1069         if (handler->pid < 0) {
1070                 SYSERROR("failed to fork into a new namespace");
1071                 goto out_delete_net;
1072         }
1073
1074         if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid, &errmsg)) {
1075                 INFO("Failed to store namespace references for stop hook: %s",
1076                         errmsg ? errmsg : "(Out of memory)");
1077                 free(errmsg);
1078         }
1079
1080         if (attach_ns(saved_ns_fd))
1081                 WARN("failed to restore saved namespaces");
1082
1083         lxc_sync_fini_child(handler);
1084
1085         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
1086                 failed_before_rename = 1;
1087
1088         if (!cgroup_create_legacy(handler)) {
1089                 ERROR("failed to setup the legacy cgroups for %s", name);
1090                 goto out_delete_net;
1091         }
1092         if (!cgroup_setup_limits(handler, false)) {
1093                 ERROR("failed to setup the cgroup limits for '%s'", name);
1094                 goto out_delete_net;
1095         }
1096
1097         if (!cgroup_enter(handler))
1098                 goto out_delete_net;
1099
1100         if (!cgroup_chown(handler))
1101                 goto out_delete_net;
1102
1103         if (failed_before_rename)
1104                 goto out_delete_net;
1105
1106         /* Create the network configuration */
1107         if (handler->clone_flags & CLONE_NEWNET) {
1108                 if (lxc_assign_network(handler->lxcpath, handler->name,
1109                                         &handler->conf->network, handler->pid)) {
1110                         ERROR("failed to create the configured network");
1111                         goto out_delete_net;
1112                 }
1113         }
1114
1115         if (netpipe != -1) {
1116                 struct lxc_list *iterator;
1117                 struct lxc_netdev *netdev;
1118
1119                 close(netpipe);
1120                 lxc_list_for_each(iterator, &handler->conf->network) {
1121                         netdev = iterator->elem;
1122                         if (netdev->type != LXC_NET_VETH)
1123                                 continue;
1124                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1125                                 ERROR("Error writing veth name to container");
1126                                 goto out_delete_net;
1127                         }
1128                 }
1129                 close(netpipepair[1]);
1130         }
1131
1132         /* map the container uids - the container became an invalid
1133          * userid the moment it was cloned with CLONE_NEWUSER - this
1134          * call doesn't change anything immediately, but allows the
1135          * container to setuid(0) (0 being mapped to something else on
1136          * the host) later to become a valid uid again */
1137         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
1138                 ERROR("failed to set up id mapping");
1139                 goto out_delete_net;
1140         }
1141
1142         /* Tell the child to continue its initialization.  we'll get
1143          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1144          */
1145         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1146                 goto out_delete_net;
1147
1148         if (!cgroup_setup_limits(handler, true)) {
1149                 ERROR("failed to setup the devices cgroup for '%s'", name);
1150                 goto out_delete_net;
1151         }
1152
1153         cgroup_disconnect();
1154         cgroups_connected = false;
1155
1156         /* read tty fds allocated by child */
1157         if (recv_ttys_from_child(handler) < 0) {
1158                 ERROR("failed to receive tty info from child");
1159                 goto out_delete_net;
1160         }
1161
1162         /* Tell the child to complete its initialization and wait for
1163          * it to exec or return an error.  (the child will never
1164          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
1165          * sync pipe, causing lxc_sync_barrier_child to return
1166          * success, or return a different value, causing us to error
1167          * out).
1168          */
1169         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1170                 return -1;
1171
1172         if (detect_shared_rootfs())
1173                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
1174
1175         if (handler->ops->post_start(handler, handler->data))
1176                 goto out_abort;
1177
1178         if (lxc_set_state(name, handler, RUNNING)) {
1179                 ERROR("failed to set state to %s",
1180                               lxc_state2str(RUNNING));
1181                 goto out_abort;
1182         }
1183
1184         lxc_sync_fini(handler);
1185
1186         return 0;
1187
1188 out_delete_net:
1189         if (cgroups_connected)
1190                 cgroup_disconnect();
1191         if (handler->clone_flags & CLONE_NEWNET)
1192                 lxc_delete_network(handler);
1193 out_abort:
1194         lxc_abort(name, handler);
1195         lxc_sync_fini(handler);
1196         if (handler->pinfd >= 0) {
1197                 close(handler->pinfd);
1198                 handler->pinfd = -1;
1199         }
1200
1201         return -1;
1202 }
1203
1204 int get_netns_fd(int pid)
1205 {
1206         char path[MAXPATHLEN];
1207         int ret, fd;
1208
1209         ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid);
1210         if (ret < 0 || ret >= MAXPATHLEN) {
1211                 WARN("Failed to pin netns file for pid %d", pid);
1212                 return -1;
1213         }
1214
1215         fd = open(path, O_RDONLY);
1216         if (fd < 0) {
1217                 WARN("Failed to pin netns file %s for pid %d: %s",
1218                                 path, pid, strerror(errno));
1219                 return -1;
1220         }
1221         return fd;
1222 }
1223
1224 int __lxc_start(const char *name, struct lxc_conf *conf,
1225                 struct lxc_operations* ops, void *data, const char *lxcpath,
1226                 bool backgrounded)
1227 {
1228         struct lxc_handler *handler;
1229         int err = -1;
1230         int status;
1231         int netnsfd = -1;
1232
1233         handler = lxc_init(name, conf, lxcpath);
1234         if (!handler) {
1235                 ERROR("failed to initialize the container");
1236                 return -1;
1237         }
1238         handler->ops = ops;
1239         handler->data = data;
1240         handler->backgrounded = backgrounded;
1241
1242         if (must_drop_cap_sys_boot(handler->conf)) {
1243                 #if HAVE_SYS_CAPABILITY_H
1244                 DEBUG("Dropping cap_sys_boot");
1245                 #else
1246                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1247                 #endif
1248         } else {
1249                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1250                 handler->conf->need_utmp_watch = 0;
1251         }
1252
1253         if (!attach_block_device(handler->conf)) {
1254                 ERROR("Failure attaching block device");
1255                 goto out_fini_nonet;
1256         }
1257
1258         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1259                 /* if the backing store is a device, mount it here and now */
1260                 if (rootfs_is_blockdev(conf)) {
1261                         if (unshare(CLONE_NEWNS) < 0) {
1262                                 ERROR("Error unsharing mounts");
1263                                 goto out_fini_nonet;
1264                         }
1265                         remount_all_slave();
1266                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1267                                 ERROR("Error setting up rootfs mount as root before spawn");
1268                                 goto out_fini_nonet;
1269                         }
1270                         INFO("Set up container rootfs as host root");
1271                 }
1272         }
1273
1274         err = lxc_spawn(handler);
1275         if (err) {
1276                 ERROR("failed to spawn '%s'", name);
1277                 goto out_detach_blockdev;
1278         }
1279
1280         handler->conf->reboot = 0;
1281
1282         netnsfd = get_netns_fd(handler->pid);
1283
1284         err = lxc_poll(name, handler);
1285         if (err) {
1286                 ERROR("mainloop exited with an error");
1287                 if (netnsfd >= 0)
1288                         close(netnsfd);
1289                 goto out_abort;
1290         }
1291
1292         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1293                 continue;
1294
1295         /*
1296          * If the child process exited but was not signaled,
1297          * it didn't call reboot.  This should mean it was an
1298          * lxc-execute which simply exited.  In any case, treat
1299          * it as a 'halt'
1300          */
1301         if (WIFSIGNALED(status)) {
1302                 switch(WTERMSIG(status)) {
1303                 case SIGINT: /* halt */
1304                         DEBUG("Container halting");
1305                         break;
1306                 case SIGHUP: /* reboot */
1307                         DEBUG("Container rebooting");
1308                         handler->conf->reboot = 1;
1309                         break;
1310                 case SIGSYS: /* seccomp */
1311                         DEBUG("Container violated its seccomp policy");
1312                         break;
1313                 default:
1314                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1315                         break;
1316                 }
1317         }
1318
1319         DEBUG("Pushing physical nics back to host namespace");
1320         lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf);
1321
1322         DEBUG("Tearing down virtual network devices used by container");
1323         lxc_delete_network(handler);
1324
1325         if (netnsfd >= 0)
1326                 close(netnsfd);
1327
1328         if (handler->pinfd >= 0) {
1329                 close(handler->pinfd);
1330                 handler->pinfd = -1;
1331         }
1332
1333         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1334         err =  lxc_error_set_and_log(handler->pid, status);
1335 out_fini:
1336         lxc_delete_network(handler);
1337
1338 out_detach_blockdev:
1339         detach_block_device(handler->conf);
1340
1341 out_fini_nonet:
1342         lxc_fini(name, handler);
1343         return err;
1344
1345 out_abort:
1346         lxc_abort(name, handler);
1347         goto out_fini;
1348 }
1349
1350 struct start_args {
1351         char *const *argv;
1352 };
1353
1354 static int start(struct lxc_handler *handler, void* data)
1355 {
1356         struct start_args *arg = data;
1357
1358         NOTICE("exec'ing '%s'", arg->argv[0]);
1359
1360         execvp(arg->argv[0], arg->argv);
1361         SYSERROR("failed to exec %s", arg->argv[0]);
1362         return 0;
1363 }
1364
1365 static int post_start(struct lxc_handler *handler, void* data)
1366 {
1367         struct start_args *arg = data;
1368
1369         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1370         return 0;
1371 }
1372
1373 static struct lxc_operations start_ops = {
1374         .start = start,
1375         .post_start = post_start
1376 };
1377
1378 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1379               const char *lxcpath, bool backgrounded)
1380 {
1381         struct start_args start_arg = {
1382                 .argv = argv,
1383         };
1384
1385         conf->need_utmp_watch = 1;
1386         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded);
1387 }
1388
1389 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1390                                             const char *name)
1391 {
1392         char destroy[MAXPATHLEN];
1393         bool bret = true;
1394         int ret = 0;
1395         struct lxc_container *c;
1396         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1397                 bret = do_destroy_container(handler->conf);
1398                 if (!bret) {
1399                         ERROR("Error destroying rootfs for %s", name);
1400                         return;
1401                 }
1402         }
1403         INFO("Destroyed rootfs for %s", name);
1404
1405         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1406         if (ret < 0 || ret >= MAXPATHLEN) {
1407                 ERROR("Error printing path for %s", name);
1408                 ERROR("Error destroying directory for %s", name);
1409                 return;
1410         }
1411
1412         c = lxc_container_new(name, handler->lxcpath);
1413         if (c) {
1414                 if (container_disk_lock(c)) {
1415                         INFO("Could not update lxc_snapshots file");
1416                         lxc_container_put(c);
1417                 } else {
1418                         mod_all_rdeps(c, false);
1419                         container_disk_unlock(c);
1420                         lxc_container_put(c);
1421                 }
1422         }
1423
1424         if (am_unpriv())
1425                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, destroy);
1426         else
1427                 ret = lxc_rmdir_onedev(destroy, NULL);
1428
1429         if (ret < 0) {
1430                 ERROR("Error destroying directory for %s", name);
1431                 return;
1432         }
1433         INFO("Destroyed directory for %s", name);
1434 }
1435
1436 static int lxc_rmdir_onedev_wrapper(void *data)
1437 {
1438         char *arg = (char *) data;
1439         return lxc_rmdir_onedev(arg, NULL);
1440 }
1441
1442 static bool do_destroy_container(struct lxc_conf *conf) {
1443         if (am_unpriv()) {
1444                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf) < 0)
1445                         return false;
1446                 return true;
1447         }
1448         return bdev_destroy(conf);
1449 }
1450