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 "lxccontainer.h"
  75 #include "lxclock.h"
  76 #include "lxcseccomp.h"
  77 #include "lxcutmp.h"
  78 #include "mainloop.h"
  79 #include "monitor.h"
  80 #include "namespace.h"
  81 #include "start.h"
  82 #include "storage_utils.h"
  83 #include "sync.h"
  84 #include "utils.h"
  85 #include "lsm/lsm.h"
  86
  87 lxc_log_define(lxc_start, lxc);
  88
  89 extern void mod_all_rdeps(struct lxc_container *c, bool inc);
  90 static bool do_destroy_container(struct lxc_conf *conf);
  91 static int lxc_rmdir_onedev_wrapper(void *data);
  92 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
  93                                             const char *name);
  94
  95 static void print_top_failing_dir(const char *path)
  96 {
  97         size_t len = strlen(path);
  98         char *copy = alloca(len + 1), *p, *e, saved;
  99         strcpy(copy, path);
 100
 101         p = copy;
 102         e = copy + len;
 103         while (p < e) {
 104                 while (p < e && *p == '/')
 105                         p++;
 106                 while (p < e && *p != '/')
 107                         p++;
 108                 saved = *p;
 109                 *p = '\0';
 110                 if (access(copy, X_OK)) {
 111                         SYSERROR("Could not access %s. Please grant it x "
 112                                  "access, or add an ACL for the container "
 113                                  "root.", copy);
 114                         return;
 115                 }
 116                 *p = saved;
 117         }
 118 }
 119
 120 static void close_ns(int ns_fd[LXC_NS_MAX])
 121 {
 122         int i;
 123
 124         for (i = 0; i < LXC_NS_MAX; i++) {
 125                 if (ns_fd[i] > -1) {
 126                         close(ns_fd[i]);
 127                         ns_fd[i] = -1;
 128                 }
 129         }
 130 }
 131
 132 /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
 133  * in clone_flags.
 134  * Return true on success, false on failure.
 135  */
 136 static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid)
 137 {
 138         int i, ret;
 139
 140         for (i = 0; i < LXC_NS_MAX; i++)
 141                 ns_fd[i] = -1;
 142
 143         ret = lxc_preserve_ns(pid, "");
 144         if (ret < 0) {
 145                 SYSERROR("Kernel does not support attaching to namespaces.");
 146                 return false;
 147         } else {
 148                 close(ret);
 149         }
 150
 151         for (i = 0; i < LXC_NS_MAX; i++) {
 152                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 153                         continue;
 154                 ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name);
 155                 if (ns_fd[i] < 0)
 156                         goto error;
 157         }
 158
 159         return true;
 160
 161 error:
 162         if (errno == ENOENT)
 163                 SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name);
 164         else
 165                 SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno));
 166         close_ns(ns_fd);
 167         return false;
 168 }
 169
 170 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 171         int i;
 172
 173         for (i = 0; i < LXC_NS_MAX; i++) {
 174                 if (ns_fd[i] < 0)
 175                         continue;
 176
 177                 if (setns(ns_fd[i], 0) != 0)
 178                         goto error;
 179         }
 180         return 0;
 181
 182 error:
 183         SYSERROR("Failed to attach %s namespace.", ns_info[i].proc_name);
 184         return -1;
 185 }
 186
 187 static int match_fd(int fd)
 188 {
 189         return (fd == 0 || fd == 1 || fd == 2);
 190 }
 191
 192 int lxc_check_inherited(struct lxc_conf *conf, bool closeall,
 193                         int *fds_to_ignore, size_t len_fds)
 194 {
 195         struct dirent *direntp;
 196         int fd, fddir;
 197         size_t i;
 198         DIR *dir;
 199
 200         if (conf && conf->close_all_fds)
 201                 closeall = true;
 202
 203 restart:
 204         dir = opendir("/proc/self/fd");
 205         if (!dir) {
 206                 WARN("Failed to open directory: %s.", strerror(errno));
 207                 return -1;
 208         }
 209
 210         fddir = dirfd(dir);
 211
 212         while ((direntp = readdir(dir))) {
 213                 if (!direntp)
 214                         break;
 215
 216                 if (!strcmp(direntp->d_name, "."))
 217                         continue;
 218
 219                 if (!strcmp(direntp->d_name, ".."))
 220                         continue;
 221
 222                 if (lxc_safe_int(direntp->d_name, &fd) < 0) {
 223                         INFO("Could not parse file descriptor for: %s", direntp->d_name);
 224                         continue;
 225                 }
 226
 227                 for (i = 0; i < len_fds; i++)
 228                         if (fds_to_ignore[i] == fd)
 229                                 break;
 230
 231                 if (fd == fddir || fd == lxc_log_fd ||
 232                     (i < len_fds && fd == fds_to_ignore[i]))
 233                         continue;
 234
 235                 if (current_config && fd == current_config->logfd)
 236                         continue;
 237
 238                 if (match_fd(fd))
 239                         continue;
 240
 241                 if (closeall) {
 242                         close(fd);
 243                         closedir(dir);
 244                         INFO("Closed inherited fd: %d.", fd);
 245                         goto restart;
 246                 }
 247                 WARN("Inherited fd: %d.", fd);
 248         }
 249
 250         /* Only enable syslog at this point to avoid the above logging function
 251          * to open a new fd and make the check_inherited function enter an
 252          * infinite loop.
 253          */
 254         lxc_log_enable_syslog();
 255
 256         closedir(dir); /* cannot fail */
 257         return 0;
 258 }
 259
 260 static int setup_signal_fd(sigset_t *oldmask)
 261 {
 262         sigset_t mask;
 263         int fd;
 264
 265         /* Block everything except serious error signals. */
 266         if (sigfillset(&mask) ||
 267             sigdelset(&mask, SIGILL) ||
 268             sigdelset(&mask, SIGSEGV) ||
 269             sigdelset(&mask, SIGBUS) ||
 270             sigdelset(&mask, SIGWINCH) ||
 271             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 272                 SYSERROR("Failed to set signal mask.");
 273                 return -1;
 274         }
 275
 276         fd = signalfd(-1, &mask, 0);
 277         if (fd < 0) {
 278                 SYSERROR("Failed to create signal file descriptor.");
 279                 return -1;
 280         }
 281
 282         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 283                 SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd);
 284                 close(fd);
 285                 return -1;
 286         }
 287
 288         DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd);
 289
 290         return fd;
 291 }
 292
 293 static int signal_handler(int fd, uint32_t events, void *data,
 294                           struct lxc_epoll_descr *descr)
 295 {
 296         struct signalfd_siginfo siginfo;
 297         siginfo_t info;
 298         int ret;
 299         pid_t *pid = data;
 300         bool init_died = false;
 301
 302         ret = read(fd, &siginfo, sizeof(siginfo));
 303         if (ret < 0) {
 304                 ERROR("Failed to read signal info from signal file descriptor: %d.", fd);
 305                 return -1;
 306         }
 307
 308         if (ret != sizeof(siginfo)) {
 309                 ERROR("Unexpected size for siginfo struct.");
 310                 return -1;
 311         }
 312
 313         /* Check whether init is running. */
 314         info.si_pid = 0;
 315         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 316         if (ret == 0 && info.si_pid == *pid)
 317                 init_died = true;
 318
 319         if (siginfo.ssi_signo != SIGCHLD) {
 320                 kill(*pid, siginfo.ssi_signo);
 321                 INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid);
 322                 return init_died ? 1 : 0;
 323         }
 324
 325         if (siginfo.ssi_code == CLD_STOPPED) {
 326                 INFO("Container init process was stopped.");
 327                 return init_died ? 1 : 0;
 328         } else if (siginfo.ssi_code == CLD_CONTINUED) {
 329                 INFO("Container init process was continued.");
 330                 return init_died ? 1 : 0;
 331         }
 332
 333         /* More robustness, protect ourself from a SIGCHLD sent
 334          * by a process different from the container init.
 335          */
 336         if (siginfo.ssi_pid != *pid) {
 337                 NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid);
 338                 return init_died ? 1 : 0;
 339         }
 340
 341         DEBUG("Container init process %d exited.", *pid);
 342         return 1;
 343 }
 344
 345 int lxc_set_state(const char *name, struct lxc_handler *handler,
 346                   lxc_state_t state)
 347 {
 348         ssize_t ret;
 349         struct lxc_list *cur, *next;
 350         struct state_client *client;
 351         struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 352
 353         process_lock();
 354         /* Only set state under process lock held so that we don't cause
 355          * lxc_cmd_state_server() to miss a state.
 356          */
 357         handler->state = state;
 358         TRACE("set container state to %s", lxc_state2str(state));
 359
 360         if (lxc_list_empty(&handler->state_clients)) {
 361                 TRACE("no state clients registered");
 362                 process_unlock();
 363                 lxc_monitor_send_state(name, state, handler->lxcpath);
 364                 return 0;
 365         }
 366
 367         strncpy(msg.name, name, sizeof(msg.name));
 368         msg.name[sizeof(msg.name) - 1] = 0;
 369
 370         lxc_list_for_each_safe(cur, &handler->state_clients, next) {
 371                 client = cur->elem;
 372
 373                 if (!client->states[state]) {
 374                         TRACE("state %s not registered for state client %d",
 375                               lxc_state2str(state), client->clientfd);
 376                         continue;
 377                 }
 378
 379                 TRACE("sending state %s to state client %d",
 380                       lxc_state2str(state), client->clientfd);
 381
 382         again:
 383                 ret = send(client->clientfd, &msg, sizeof(msg), 0);
 384                 if (ret < 0) {
 385                         if (errno == EINTR)
 386                                 goto again;
 387
 388                         ERROR("failed to send message to client");
 389                 }
 390
 391                 /* kick client from list */
 392                 close(client->clientfd);
 393                 lxc_list_del(cur);
 394                 free(cur->elem);
 395                 free(cur);
 396         }
 397         process_unlock();
 398
 399         /* This function will try to connect to the legacy lxc-monitord state
 400          * server and only exists for backwards compatibility.
 401          */
 402         lxc_monitor_send_state(name, state, handler->lxcpath);
 403
 404         return 0;
 405 }
 406
 407 int lxc_poll(const char *name, struct lxc_handler *handler)
 408 {
 409         int sigfd = handler->sigfd;
 410         int pid = handler->pid;
 411         struct lxc_epoll_descr descr;
 412
 413         if (lxc_mainloop_open(&descr)) {
 414                 ERROR("Failed to create LXC mainloop.");
 415                 goto out_sigfd;
 416         }
 417
 418         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 419                 ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd);
 420                 goto out_mainloop_open;
 421         }
 422
 423         if (lxc_console_mainloop_add(&descr, handler->conf)) {
 424                 ERROR("Failed to add console handler to LXC mainloop.");
 425                 goto out_mainloop_open;
 426         }
 427
 428         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 429                 ERROR("Failed to add command handler to LXC mainloop.");
 430                 goto out_mainloop_open;
 431         }
 432
 433         if (handler->conf->need_utmp_watch) {
 434                 #if HAVE_LIBCAP
 435                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 436                         ERROR("Failed to add utmp handler to LXC mainloop.");
 437                         goto out_mainloop_open;
 438                 }
 439                 #else
 440                         DEBUG("Not starting utmp handler as CAP_SYS_BOOT cannot be dropped without capabilities support.");
 441                 #endif
 442         }
 443         TRACE("lxc mainloop is ready");
 444
 445         return lxc_mainloop(&descr, -1);
 446
 447 out_mainloop_open:
 448         lxc_mainloop_close(&descr);
 449
 450 out_sigfd:
 451         close(sigfd);
 452
 453         return -1;
 454 }
 455
 456 void lxc_free_handler(struct lxc_handler *handler)
 457 {
 458         if (handler->conf && handler->conf->maincmd_fd)
 459                 close(handler->conf->maincmd_fd);
 460
 461         if (handler->name)
 462                 free(handler->name);
 463
 464         handler->conf = NULL;
 465         free(handler);
 466 }
 467
 468 struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf,
 469                                      const char *lxcpath)
 470 {
 471         int i;
 472         struct lxc_handler *handler;
 473
 474         handler = malloc(sizeof(*handler));
 475         if (!handler) {
 476                 ERROR("failed to allocate memory");
 477                 return NULL;
 478         }
 479
 480         memset(handler, 0, sizeof(*handler));
 481
 482         handler->ttysock[0] = handler->ttysock[1] = -1;
 483         handler->conf = conf;
 484         handler->lxcpath = lxcpath;
 485         handler->pinfd = -1;
 486         lxc_list_init(&handler->state_clients);
 487
 488         for (i = 0; i < LXC_NS_MAX; i++)
 489                 handler->nsfd[i] = -1;
 490
 491         handler->name = strdup(name);
 492         if (!handler->name) {
 493                 ERROR("failed to allocate memory");
 494                 goto on_error;
 495         }
 496
 497         if (lxc_cmd_init(name, handler, lxcpath)) {
 498                 ERROR("failed to set up command socket");
 499                 goto on_error;
 500         }
 501
 502         TRACE("unix domain socket %d for command server is ready",
 503               handler->conf->maincmd_fd);
 504
 505         return handler;
 506
 507 on_error:
 508         lxc_free_handler(handler);
 509
 510         return NULL;
 511 }
 512
 513 int lxc_init(const char *name, struct lxc_handler *handler)
 514 {
 515         struct lxc_conf *conf = handler->conf;
 516
 517         lsm_init();
 518         TRACE("initialized LSM");
 519
 520         if (lxc_read_seccomp_config(conf) != 0) {
 521                 ERROR("Failed loading seccomp policy.");
 522                 goto out_close_maincmd_fd;
 523         }
 524         TRACE("read seccomp policy");
 525
 526         /* Begin by setting the state to STARTING. */
 527         if (lxc_set_state(name, handler, STARTING)) {
 528                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING));
 529                 goto out_close_maincmd_fd;
 530         }
 531         TRACE("set container state to \"STARTING\"");
 532
 533         /* Start of environment variable setup for hooks. */
 534         if (name && setenv("LXC_NAME", name, 1))
 535                 SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name);
 536
 537         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1))
 538                 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile);
 539
 540         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1))
 541                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount);
 542
 543         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1))
 544                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path);
 545
 546         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1))
 547                 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path);
 548
 549         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1))
 550                 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path);
 551
 552         if (setenv("LXC_CGNS_AWARE", "1", 1))
 553                 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1.");
 554         /* End of environment variable setup for hooks. */
 555
 556         TRACE("set environment variables");
 557
 558         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 559                 ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name);
 560                 goto out_aborting;
 561         }
 562         TRACE("ran pre-start hooks");
 563
 564         /* The signal fd has to be created before forking otherwise if the child
 565          * process exits before we setup the signal fd, the event will be lost
 566          * and the command will be stuck.
 567          */
 568         handler->sigfd = setup_signal_fd(&handler->oldmask);
 569         if (handler->sigfd < 0) {
 570                 ERROR("Failed to setup SIGCHLD fd handler.");
 571                 goto out_delete_tty;
 572         }
 573         TRACE("set up signal fd");
 574
 575         /* Do this after setting up signals since it might unblock SIGWINCH. */
 576         if (lxc_console_create(conf)) {
 577                 ERROR("Failed to create console for container \"%s\".", name);
 578                 goto out_restore_sigmask;
 579         }
 580         TRACE("created console");
 581
 582         if (lxc_ttys_shift_ids(conf) < 0) {
 583                 ERROR("Failed to shift tty into container.");
 584                 goto out_restore_sigmask;
 585         }
 586         TRACE("shifted tty ids");
 587
 588         INFO("container \"%s\" is initialized", name);
 589         return 0;
 590
 591 out_restore_sigmask:
 592         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 593 out_delete_tty:
 594         lxc_delete_tty(&conf->tty_info);
 595 out_aborting:
 596         lxc_set_state(name, handler, ABORTING);
 597 out_close_maincmd_fd:
 598         close(conf->maincmd_fd);
 599         conf->maincmd_fd = -1;
 600         return -1;
 601 }
 602
 603 void lxc_fini(const char *name, struct lxc_handler *handler)
 604 {
 605         int i, rc;
 606         struct lxc_list *cur, *next;
 607         pid_t self = getpid();
 608         char *namespaces[LXC_NS_MAX + 1];
 609         size_t namespace_count = 0;
 610
 611         /* The STOPPING state is there for future cleanup code which can take
 612          * awhile.
 613          */
 614         lxc_set_state(name, handler, STOPPING);
 615
 616         for (i = 0; i < LXC_NS_MAX; i++) {
 617                 if (handler->nsfd[i] != -1) {
 618                         rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d",
 619                                       ns_info[i].proc_name, self, handler->nsfd[i]);
 620                         if (rc == -1) {
 621                                 SYSERROR("Failed to allocate memory.");
 622                                 break;
 623                         }
 624                         ++namespace_count;
 625                 }
 626         }
 627         namespaces[namespace_count] = NULL;
 628
 629         if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1))
 630                 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot.");
 631
 632         if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1))
 633                 SYSERROR("Failed to set environment variable: LXC_TARGET=stop.");
 634
 635         if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces))
 636                 ERROR("Failed to run lxc.hook.stop for container \"%s\".", name);
 637
 638         while (namespace_count--)
 639                 free(namespaces[namespace_count]);
 640         for (i = 0; i < LXC_NS_MAX; i++) {
 641                 if (handler->nsfd[i] != -1) {
 642                         close(handler->nsfd[i]);
 643                         handler->nsfd[i] = -1;
 644                 }
 645         }
 646
 647         if (handler->netnsfd >= 0) {
 648                 close(handler->netnsfd);
 649                 handler->netnsfd = -1;
 650         }
 651
 652         lxc_set_state(name, handler, STOPPED);
 653
 654         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) {
 655                 ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name);
 656                 if (handler->conf->reboot) {
 657                         WARN("Container will be stopped instead of rebooted.");
 658                         handler->conf->reboot = 0;
 659                         if (setenv("LXC_TARGET", "stop", 1))
 660                                 WARN("Failed to set environment variable: LXC_TARGET=stop.");
 661                 }
 662         }
 663
 664         /* Reset mask set by setup_signal_fd. */
 665         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 666                 WARN("Failed to restore signal mask.");
 667
 668         lxc_console_delete(&handler->conf->console);
 669         lxc_delete_tty(&handler->conf->tty_info);
 670
 671         /* close the command socket */
 672         close(handler->conf->maincmd_fd);
 673         handler->conf->maincmd_fd = -1;
 674
 675         /* The command socket is now closed, no more state clients can register
 676          * themselves from now on. So free the list of state clients.
 677          */
 678         lxc_list_for_each_safe(cur, &handler->state_clients, next) {
 679                 struct state_client *client = cur->elem;
 680                 /* close state client socket */
 681                 close(client->clientfd);
 682                 lxc_list_del(cur);
 683                 free(cur->elem);
 684                 free(cur);
 685         }
 686
 687         free(handler->name);
 688         if (handler->ttysock[0] != -1) {
 689                 close(handler->ttysock[0]);
 690                 close(handler->ttysock[1]);
 691         }
 692
 693         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1)
 694                 lxc_destroy_container_on_signal(handler, name);
 695
 696         cgroup_destroy(handler);
 697         free(handler);
 698 }
 699
 700 void lxc_abort(const char *name, struct lxc_handler *handler)
 701 {
 702         int ret, status;
 703
 704         lxc_set_state(name, handler, ABORTING);
 705         if (handler->pid > 0)
 706                 kill(handler->pid, SIGKILL);
 707         while ((ret = waitpid(-1, &status, 0)) > 0) {
 708                 ;
 709         }
 710 }
 711
 712 #include <sys/reboot.h>
 713 #include <linux/reboot.h>
 714
 715 /* reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL in a child pid namespace
 716  * if container reboot support exists.  Otherwise, it will either succeed or
 717  * return -EPERM.
 718  */
 719 static int container_reboot_supported(void *arg)
 720 {
 721         int *cmd = arg;
 722         int ret;
 723
 724         ret = reboot(*cmd);
 725         if (ret == -1 && errno == EINVAL)
 726                 return 1;
 727         return 0;
 728 }
 729
 730 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 731 {
 732         FILE *f;
 733         int ret, cmd, v, flags;
 734         long stack_size = 4096;
 735         void *stack = alloca(stack_size);
 736         int status;
 737         pid_t pid;
 738
 739         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 740         if (!f) {
 741                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 742                 return 1;
 743         }
 744
 745         ret = fscanf(f, "%d", &v);
 746         fclose(f);
 747         if (ret != 1) {
 748                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del.");
 749                 return 1;
 750         }
 751         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 752
 753         flags = CLONE_NEWPID | SIGCHLD;
 754         if (!lxc_list_empty(&conf->id_map))
 755                 flags |= CLONE_NEWUSER;
 756
 757 #ifdef __ia64__
 758         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 759 #else
 760         stack += stack_size;
 761         pid = clone(container_reboot_supported, stack, flags, &cmd);
 762 #endif
 763         if (pid < 0) {
 764                 if (flags & CLONE_NEWUSER)
 765                         ERROR("Failed to clone (%#x): %s (includes CLONE_NEWUSER).", flags, strerror(errno));
 766                 else
 767                         ERROR("Failed to clone (%#x): %s.", flags, strerror(errno));
 768                 return -1;
 769         }
 770         if (wait(&status) < 0) {
 771                 SYSERROR("Unexpected wait error: %s.", strerror(errno));
 772                 return -1;
 773         }
 774
 775         if (WEXITSTATUS(status) != 1)
 776                 return 1;
 777
 778         return 0;
 779 }
 780
 781 /* netpipe is used in the unprivileged case to transfer the ifindexes from
 782  * parent to child
 783  */
 784 static int netpipe = -1;
 785
 786 static inline int count_veths(struct lxc_list *network)
 787 {
 788         struct lxc_list *iterator;
 789         struct lxc_netdev *netdev;
 790         int count = 0;
 791
 792         lxc_list_for_each(iterator, network) {
 793                 netdev = iterator->elem;
 794                 if (netdev->type != LXC_NET_VETH)
 795                         continue;
 796                 count++;
 797         }
 798         return count;
 799 }
 800
 801 static int read_unpriv_netifindex(struct lxc_list *network)
 802 {
 803         struct lxc_list *iterator;
 804         struct lxc_netdev *netdev;
 805
 806         if (netpipe == -1)
 807                 return 0;
 808         lxc_list_for_each(iterator, network) {
 809                 netdev = iterator->elem;
 810                 if (netdev->type != LXC_NET_VETH)
 811                         continue;
 812                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 813                         ERROR("Out of memory.");
 814                         close(netpipe);
 815                         return -1;
 816                 }
 817                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 818                         close(netpipe);
 819                         return -1;
 820                 }
 821         }
 822         close(netpipe);
 823         return 0;
 824 }
 825
 826 static int do_start(void *data)
 827 {
 828         struct lxc_list *iterator;
 829         struct lxc_handler *handler = data;
 830         int devnull_fd = -1, ret;
 831         char path[PATH_MAX];
 832
 833         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 834                 SYSERROR("Failed to set signal mask.");
 835                 return -1;
 836         }
 837
 838         /* This prctl must be before the synchro, so if the parent dies before
 839          * we set the parent death signal, we will detect its death with the
 840          * synchro right after, otherwise we have a window where the parent can
 841          * exit before we set the pdeath signal leading to a unsupervized
 842          * container.
 843          */
 844         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 845                 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL.");
 846                 return -1;
 847         }
 848
 849         lxc_sync_fini_parent(handler);
 850
 851         /* Don't leak the pinfd to the container. */
 852         if (handler->pinfd >= 0) {
 853                 close(handler->pinfd);
 854         }
 855
 856         if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP))
 857                 return -1;
 858
 859         /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
 860          * https://github.com/lxc/lxd/issues/1978.
 861          */
 862         if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) ==
 863             (CLONE_NEWNET | CLONE_NEWUSER)) {
 864                 ret = unshare(CLONE_NEWNET);
 865                 if (ret < 0) {
 866                         SYSERROR("Failed to unshare CLONE_NEWNET.");
 867                         goto out_warn_father;
 868                 }
 869                 INFO("Unshared CLONE_NEWNET.");
 870         }
 871
 872         /* Tell the parent task it can begin to configure the container and wait
 873          * for it to finish.
 874          */
 875         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 876                 return -1;
 877
 878         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 879                 goto out_warn_father;
 880
 881         /* If we are in a new user namespace, become root there to have
 882          * privilege over our namespace.
 883          */
 884         if (!lxc_list_empty(&handler->conf->id_map)) {
 885                 if (lxc_switch_uid_gid(0, 0) < 0)
 886                         goto out_warn_father;
 887
 888                 /* Drop groups only after we switched to a valid gid in the new
 889                  * user namespace.
 890                  */
 891                 if (lxc_setgroups(0, NULL) < 0)
 892                         goto out_warn_father;
 893         }
 894
 895         if (access(handler->lxcpath, X_OK)) {
 896                 print_top_failing_dir(handler->lxcpath);
 897                 goto out_warn_father;
 898         }
 899
 900         #if HAVE_LIBCAP
 901         if (handler->conf->need_utmp_watch) {
 902                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 903                         SYSERROR("Failed to remove the CAP_SYS_BOOT capability.");
 904                         goto out_warn_father;
 905                 }
 906                 DEBUG("Dropped the CAP_SYS_BOOT capability.");
 907         }
 908         #endif
 909
 910         ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount);
 911         if (ret < 0 || ret >= sizeof(path))
 912                 goto out_warn_father;
 913
 914         /* In order to checkpoint restore, we need to have everything in the
 915          * same mount namespace. However, some containers may not have a
 916          * reasonable /dev (in particular, they may not have /dev/null), so we
 917          * can't set init's std fds to /dev/null by opening it from inside the
 918          * container.
 919          *
 920          * If that's the case, fall back to using the host's /dev/null. This
 921          * means that migration won't work, but at least we won't spew output
 922          * where it isn't wanted.
 923          */
 924         if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) {
 925                 devnull_fd = open_devnull();
 926
 927                 if (devnull_fd < 0)
 928                         goto out_warn_father;
 929                 WARN("Using /dev/null from the host for container init's "
 930                      "standard file descriptors. Migration will not work.");
 931         }
 932
 933         /* Setup the container, ip, names, utsname, ... */
 934         if (lxc_setup(handler)) {
 935                 ERROR("Failed to setup container \"%s\".", handler->name);
 936                 goto out_warn_father;
 937         }
 938
 939         /* Ask father to setup cgroups and wait for him to finish. */
 940         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 941                 goto out_error;
 942
 943         /* Unshare cgroup namespace after we have setup our cgroups. If we do it
 944          * earlier we end up with a wrong view of /proc/self/cgroup. For
 945          * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
 946          * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
 947          * /proc/self/cgroup would show us:
 948          *
 949          *      8:cpuset:/lxc/c
 950          *
 951          * whereas it should actually show
 952          *
 953          *      8:cpuset:/
 954          */
 955         if (cgns_supported()) {
 956                 if (unshare(CLONE_NEWCGROUP) < 0) {
 957                         INFO("Failed to unshare CLONE_NEWCGROUP.");
 958                         goto out_warn_father;
 959                 }
 960                 INFO("Unshared CLONE_NEWCGROUP.");
 961         }
 962
 963         /* Set the label to change to when we exec(2) the container's init. */
 964         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 965                 goto out_warn_father;
 966
 967         /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
 968          * before we aren't allowed anymore.
 969          */
 970         if (handler->conf->no_new_privs) {
 971                 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
 972                         SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 973                         goto out_warn_father;
 974                 }
 975                 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 976         }
 977
 978         /* Some init's such as busybox will set sane tty settings on stdin,
 979          * stdout, stderr which it thinks is the console. We already set them
 980          * the way we wanted on the real terminal, and we want init to do its
 981          * setup on its console ie. the pty allocated in lxc_console_create() so
 982          * make sure that that pty is stdin,stdout,stderr.
 983          */
 984         if (lxc_console_set_stdfds(handler->conf->console.slave) < 0)
 985                 goto out_warn_father;
 986
 987         /* If we mounted a temporary proc, then unmount it now. */
 988         tmp_proc_unmount(handler->conf);
 989
 990         if (lxc_seccomp_load(handler->conf) != 0)
 991                 goto out_warn_father;
 992
 993         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 994                 ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name);
 995                 goto out_warn_father;
 996         }
 997
 998         /* The container has been setup. We can now switch to an unprivileged
 999          * uid/gid.
1000          */
1001         if (handler->conf->is_execute) {
1002                 bool have_cap_setgid;
1003                 uid_t new_uid = handler->conf->init_uid;
1004                 gid_t new_gid = handler->conf->init_gid;
1005
1006                 /* If we are in a new user namespace we already dropped all
1007                  * groups when we switched to root in the new user namespace
1008                  * further above. Only drop groups if we can, so ensure that we
1009                  * have necessary privilege.
1010                  */
1011                 #if HAVE_LIBCAP
1012                 have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE);
1013                 #else
1014                 have_cap_setgid = false;
1015                 #endif
1016                 if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) {
1017                         if (lxc_setgroups(0, NULL) < 0)
1018                                 goto out_warn_father;
1019                 }
1020
1021                 if (lxc_switch_uid_gid(new_uid, new_gid) < 0)
1022                         goto out_warn_father;
1023         }
1024
1025         /* The clearenv() and putenv() calls have been moved here to allow us to
1026          * use environment variables passed to the various hooks, such as the
1027          * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS
1028          * are valid in this context but others are.
1029          */
1030         if (clearenv()) {
1031                 SYSERROR("Failed to clear environment.");
1032                 /* Don't error out though. */
1033         }
1034
1035         lxc_list_for_each(iterator, &handler->conf->environment) {
1036                 if (putenv((char *)iterator->elem)) {
1037                         SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem);
1038                         goto out_warn_father;
1039                 }
1040         }
1041
1042         if (putenv("container=lxc")) {
1043                 SYSERROR("Failed to set environment variable: container=lxc.");
1044                 goto out_warn_father;
1045         }
1046
1047         if (handler->conf->pty_names) {
1048                 if (putenv(handler->conf->pty_names)) {
1049                         SYSERROR("Failed to set environment variable for container ptys.");
1050                         goto out_warn_father;
1051                 }
1052         }
1053
1054         close(handler->sigfd);
1055
1056         if (devnull_fd < 0) {
1057                 devnull_fd = open_devnull();
1058
1059                 if (devnull_fd < 0)
1060                         goto out_warn_father;
1061         }
1062
1063         if (handler->backgrounded && set_stdfds(devnull_fd))
1064                 goto out_warn_father;
1065
1066         if (devnull_fd >= 0) {
1067                 close(devnull_fd);
1068                 devnull_fd = -1;
1069         }
1070
1071         setsid();
1072
1073         /* After this call, we are in error because this ops should not return
1074          * as it execs.
1075          */
1076         handler->ops->start(handler, handler->data);
1077
1078 out_warn_father:
1079         /* We want the parent to know something went wrong, so we return a
1080          * special error code.
1081          */
1082         lxc_sync_wake_parent(handler, LXC_SYNC_ERROR);
1083
1084 out_error:
1085         if (devnull_fd >= 0)
1086                 close(devnull_fd);
1087
1088         return -1;
1089 }
1090
1091 static int save_phys_nics(struct lxc_conf *conf)
1092 {
1093         struct lxc_list *iterator;
1094         int am_root = (getuid() == 0);
1095
1096         if (!am_root)
1097                 return 0;
1098
1099         lxc_list_for_each(iterator, &conf->network) {
1100                 struct lxc_netdev *netdev = iterator->elem;
1101
1102                 if (netdev->type != LXC_NET_PHYS)
1103                         continue;
1104                 conf->saved_nics = realloc(conf->saved_nics,
1105                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
1106                 if (!conf->saved_nics)
1107                         return -1;
1108                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
1109                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
1110                 if (!conf->saved_nics[conf->num_savednics].orig_name)
1111                         return -1;
1112                 INFO("Stored saved_nic #%d idx %d name %s.", conf->num_savednics,
1113                         conf->saved_nics[conf->num_savednics].ifindex,
1114                         conf->saved_nics[conf->num_savednics].orig_name);
1115                 conf->num_savednics++;
1116         }
1117
1118         return 0;
1119 }
1120
1121 static int lxc_recv_ttys_from_child(struct lxc_handler *handler)
1122 {
1123         int i;
1124         int *ttyfds;
1125         struct lxc_pty_info *pty_info;
1126         int ret = -1;
1127         int sock = handler->ttysock[1];
1128         struct lxc_conf *conf = handler->conf;
1129         struct lxc_tty_info *tty_info = &conf->tty_info;
1130         size_t num_ttyfds = (2 * conf->tty);
1131
1132         if (!conf->tty)
1133                 return 0;
1134
1135         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty);
1136         if (!tty_info->pty_info)
1137                 return -1;
1138
1139         ttyfds = malloc(num_ttyfds * sizeof(int));
1140         if (!ttyfds)
1141                 return -1;
1142
1143         ret = lxc_abstract_unix_recv_fds(sock, ttyfds, num_ttyfds, NULL, 0);
1144         for (i = 0; (ret >= 0 && *ttyfds != -1) && (i < num_ttyfds); i++) {
1145                 pty_info = &tty_info->pty_info[i / 2];
1146                 pty_info->busy = 0;
1147                 pty_info->slave = ttyfds[i++];
1148                 pty_info->master = ttyfds[i];
1149                 TRACE("received pty with master fd %d and slave fd %d from "
1150                       "parent", pty_info->master, pty_info->slave);
1151         }
1152
1153         tty_info->nbtty = conf->tty;
1154
1155         free(ttyfds);
1156
1157         if (ret < 0)
1158                 ERROR("failed to receive %d ttys from child: %s", conf->tty,
1159                       strerror(errno));
1160         else
1161                 TRACE("received %d ttys from child", conf->tty);
1162
1163         return ret;
1164 }
1165
1166 void resolve_clone_flags(struct lxc_handler *handler)
1167 {
1168         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
1169
1170         if (!lxc_list_empty(&handler->conf->id_map))
1171                 handler->clone_flags |= CLONE_NEWUSER;
1172
1173         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
1174                 if (!lxc_requests_empty_network(handler))
1175                         handler->clone_flags |= CLONE_NEWNET;
1176         } else {
1177                 INFO("Inheriting a NET namespace.");
1178         }
1179
1180         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1)
1181                 handler->clone_flags |= CLONE_NEWIPC;
1182         else
1183                 INFO("Inheriting an IPC namespace.");
1184
1185         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1)
1186                 handler->clone_flags |= CLONE_NEWUTS;
1187         else
1188                 INFO("Inheriting a UTS namespace.");
1189 }
1190
1191 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1192  * exec()s the requested container binary.
1193  * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1194  * right here should be double checked if they'd pose a security risk. (For
1195  * example, any {u}mount() operations performed here will be reflected on the
1196  * host!)
1197  */
1198 static int lxc_spawn(struct lxc_handler *handler)
1199 {
1200         int failed_before_rename = 0;
1201         const char *name = handler->name;
1202         bool cgroups_connected = false;
1203         int saved_ns_fd[LXC_NS_MAX];
1204         int preserve_mask = 0, i, flags;
1205         int netpipepair[2], nveths;
1206         bool wants_to_map_ids;
1207         struct lxc_list *id_map;
1208
1209         netpipe = -1;
1210         id_map = &handler->conf->id_map;
1211         wants_to_map_ids = !lxc_list_empty(id_map);
1212
1213         for (i = 0; i < LXC_NS_MAX; i++)
1214                 if (handler->conf->inherit_ns_fd[i] != -1)
1215                         preserve_mask |= ns_info[i].clone_flag;
1216
1217         if (lxc_sync_init(handler))
1218                 return -1;
1219
1220         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
1221                 lxc_sync_fini(handler);
1222                 return -1;
1223         }
1224
1225         resolve_clone_flags(handler);
1226
1227         if (handler->clone_flags & CLONE_NEWNET) {
1228                 if (!lxc_list_empty(&handler->conf->network)) {
1229
1230                         /* Find gateway addresses from the link device, which is
1231                          * no longer accessible inside the container. Do this
1232                          * before creating network interfaces, since goto
1233                          * out_delete_net does not work before lxc_clone.
1234                          */
1235                         if (lxc_find_gateway_addresses(handler)) {
1236                                 ERROR("Failed to find gateway addresses.");
1237                                 lxc_sync_fini(handler);
1238                                 return -1;
1239                         }
1240
1241                         /* That should be done before the clone because we will
1242                          * fill the netdev index and use them in the child.
1243                          */
1244                         if (lxc_setup_networks_in_parent_namespaces(handler)) {
1245                                 ERROR("Failed to create the network.");
1246                                 lxc_sync_fini(handler);
1247                                 return -1;
1248                         }
1249                 }
1250
1251                 if (save_phys_nics(handler->conf)) {
1252                         ERROR("Failed to save physical nic info.");
1253                         goto out_abort;
1254                 }
1255         }
1256
1257         if (!cgroup_init(handler)) {
1258                 ERROR("Failed initializing cgroup support.");
1259                 goto out_delete_net;
1260         }
1261
1262         cgroups_connected = true;
1263
1264         if (!cgroup_create(handler)) {
1265                 ERROR("Failed creating cgroups.");
1266                 goto out_delete_net;
1267         }
1268
1269         /* If the rootfs is not a blockdev, prevent the container from marking
1270          * it readonly.
1271          * If the container is unprivileged then skip rootfs pinning.
1272          */
1273         if (!wants_to_map_ids) {
1274                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
1275                 if (handler->pinfd == -1)
1276                         INFO("Failed to pin the rootfs for container \"%s\".", handler->name);
1277         }
1278
1279         if (!preserve_ns(saved_ns_fd, preserve_mask, getpid()))
1280                 goto out_delete_net;
1281
1282         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
1283                 goto out_delete_net;
1284
1285         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
1286                 if (pipe(netpipepair) < 0) {
1287                         SYSERROR("Failed to create pipe.");
1288                         goto out_delete_net;
1289                 }
1290                 /* Store netpipe in the global var for do_start's use. */
1291                 netpipe = netpipepair[0];
1292         }
1293
1294         /* Create a process in a new set of namespaces. */
1295         flags = handler->clone_flags;
1296         if (handler->clone_flags & CLONE_NEWUSER) {
1297                 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1298                  * clone a new user namespace first and only later unshare our
1299                  * network namespace to ensure that network devices ownership is
1300                  * set up correctly.
1301                  */
1302                 flags &= ~CLONE_NEWNET;
1303         }
1304         handler->pid = lxc_clone(do_start, handler, flags);
1305         if (handler->pid < 0) {
1306                 SYSERROR("Failed to clone a new set of namespaces.");
1307                 goto out_delete_net;
1308         }
1309         for (i = 0; i < LXC_NS_MAX; i++)
1310                 if (flags & ns_info[i].clone_flag)
1311                         INFO("Cloned %s.", ns_info[i].flag_name);
1312
1313         if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid))
1314                 INFO("Failed to preserve namespace for lxc.hook.stop.");
1315
1316         if (attach_ns(saved_ns_fd))
1317                 WARN("Failed to restore saved namespaces.");
1318
1319         lxc_sync_fini_child(handler);
1320
1321         /* Map the container uids. The container became an invalid userid the
1322          * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1323          * anything immediately, but allows the container to setuid(0) (0 being
1324          * mapped to something else on the host.) later to become a valid uid
1325          * again.
1326          */
1327         if (wants_to_map_ids && lxc_map_ids(id_map, handler->pid)) {
1328                 ERROR("Failed to set up id mapping.");
1329                 goto out_delete_net;
1330         }
1331
1332         if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) {
1333                 failed_before_rename = 1;
1334                 goto out_delete_net;
1335         }
1336
1337         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) {
1338                 failed_before_rename = 1;
1339                 goto out_delete_net;
1340         }
1341
1342         if (!cgroup_create_legacy(handler)) {
1343                 ERROR("Failed to setup legacy cgroups for container \"%s\".", name);
1344                 goto out_delete_net;
1345         }
1346         if (!cgroup_setup_limits(handler, false)) {
1347                 ERROR("Failed to setup cgroup limits for container \"%s\".", name);
1348                 goto out_delete_net;
1349         }
1350
1351         if (!cgroup_enter(handler))
1352                 goto out_delete_net;
1353
1354         if (!cgroup_chown(handler))
1355                 goto out_delete_net;
1356
1357         if (failed_before_rename)
1358                 goto out_delete_net;
1359
1360         /* Create the network configuration. */
1361         if (handler->clone_flags & CLONE_NEWNET) {
1362                 if (lxc_assign_network(handler->lxcpath, handler->name,
1363                                        &handler->conf->network, handler->pid)) {
1364                         ERROR("Failed to create the configured network.");
1365                         goto out_delete_net;
1366                 }
1367         }
1368
1369         if (netpipe != -1) {
1370                 struct lxc_list *iterator;
1371                 struct lxc_netdev *netdev;
1372
1373                 close(netpipe);
1374                 lxc_list_for_each(iterator, &handler->conf->network) {
1375                         netdev = iterator->elem;
1376                         if (netdev->type != LXC_NET_VETH)
1377                                 continue;
1378                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1379                                 ERROR("Error writing veth name to container.");
1380                                 goto out_delete_net;
1381                         }
1382                 }
1383                 close(netpipepair[1]);
1384         }
1385
1386         /* Tell the child to continue its initialization. We'll get
1387          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1388          */
1389         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1390                 goto out_delete_net;
1391
1392         if (!lxc_list_empty(&handler->conf->limits) && setup_resource_limits(&handler->conf->limits, handler->pid)) {
1393                 ERROR("failed to setup resource limits for '%s'", name);
1394                 goto out_delete_net;
1395         }
1396
1397         if (!cgroup_setup_limits(handler, true)) {
1398                 ERROR("Failed to setup the devices cgroup for container \"%s\".", name);
1399                 goto out_delete_net;
1400         }
1401
1402         cgroup_disconnect();
1403         cgroups_connected = false;
1404
1405         /* Read tty fds allocated by child. */
1406         if (lxc_recv_ttys_from_child(handler) < 0) {
1407                 ERROR("Failed to receive tty info from child process.");
1408                 goto out_delete_net;
1409         }
1410
1411         /* Tell the child to complete its initialization and wait for it to exec
1412          * or return an error. (The child will never return
1413          * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing
1414          * lxc_sync_barrier_child to return success, or return a different
1415          * value, causing us to error out).
1416          */
1417         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1418                 return -1;
1419
1420         if (handler->ops->post_start(handler, handler->data))
1421                 goto out_abort;
1422
1423         if (lxc_set_state(name, handler, RUNNING)) {
1424                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name,
1425                       lxc_state2str(RUNNING));
1426                 goto out_abort;
1427         }
1428
1429         lxc_sync_fini(handler);
1430         handler->netnsfd = lxc_preserve_ns(handler->pid, "net");
1431
1432         return 0;
1433
1434 out_delete_net:
1435         if (cgroups_connected)
1436                 cgroup_disconnect();
1437         if (handler->clone_flags & CLONE_NEWNET)
1438                 lxc_delete_network(handler);
1439 out_abort:
1440         lxc_abort(name, handler);
1441         lxc_sync_fini(handler);
1442         if (handler->pinfd >= 0) {
1443                 close(handler->pinfd);
1444                 handler->pinfd = -1;
1445         }
1446
1447         return -1;
1448 }
1449
1450 int __lxc_start(const char *name, struct lxc_handler *handler,
1451                 struct lxc_operations* ops, void *data, const char *lxcpath,
1452                 bool backgrounded)
1453 {
1454         int status;
1455         int err = -1;
1456         bool removed_all_netdevs = true;
1457         struct lxc_conf *conf = handler->conf;
1458
1459         if (lxc_init(name, handler) < 0) {
1460                 ERROR("Failed to initialize container \"%s\".", name);
1461                 return -1;
1462         }
1463         handler->ops = ops;
1464         handler->data = data;
1465         handler->backgrounded = backgrounded;
1466         handler->netnsfd = -1;
1467
1468         if (must_drop_cap_sys_boot(handler->conf)) {
1469                 #if HAVE_LIBCAP
1470                 DEBUG("Dropping CAP_SYS_BOOT capability.");
1471                 #else
1472                 DEBUG("Not dropping CAP_SYS_BOOT capability as capabilities aren't supported.");
1473                 #endif
1474         } else {
1475                 DEBUG("Not dropping CAP_SYS_BOOT or watching utmp.");
1476                 handler->conf->need_utmp_watch = 0;
1477         }
1478
1479         if (!attach_block_device(handler->conf)) {
1480                 ERROR("Failed to attach block device.");
1481                 goto out_fini_nonet;
1482         }
1483
1484         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1485                 /* If the backing store is a device, mount it here and now. */
1486                 if (rootfs_is_blockdev(conf)) {
1487                         if (unshare(CLONE_NEWNS) < 0) {
1488                                 ERROR("Failed to unshare CLONE_NEWNS.");
1489                                 goto out_fini_nonet;
1490                         }
1491                         INFO("Unshared CLONE_NEWNS.");
1492
1493                         remount_all_slave();
1494                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1495                                 ERROR("Error setting up rootfs mount as root before spawn.");
1496                                 goto out_fini_nonet;
1497                         }
1498                         INFO("Set up container rootfs as host root.");
1499                 }
1500         }
1501
1502         err = lxc_spawn(handler);
1503         if (err) {
1504                 ERROR("Failed to spawn container \"%s\".", name);
1505                 goto out_detach_blockdev;
1506         }
1507
1508         handler->conf->reboot = 0;
1509
1510         err = lxc_poll(name, handler);
1511         if (err) {
1512                 ERROR("LXC mainloop exited with error: %d.", err);
1513                 if (handler->netnsfd >= 0) {
1514                         close(handler->netnsfd);
1515                         handler->netnsfd = -1;
1516                 }
1517                 goto out_abort;
1518         }
1519
1520         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1521                 continue;
1522
1523         /* If the child process exited but was not signaled, it didn't call
1524          * reboot. This should mean it was an lxc-execute which simply exited.
1525          * In any case, treat it as a 'halt'.
1526          */
1527         if (WIFSIGNALED(status)) {
1528                 switch(WTERMSIG(status)) {
1529                 case SIGINT: /* halt */
1530                         DEBUG("Container \"%s\" is halting.", name);
1531                         break;
1532                 case SIGHUP: /* reboot */
1533                         DEBUG("Container \"%s\" is rebooting.", name);
1534                         handler->conf->reboot = 1;
1535                         break;
1536                 case SIGSYS: /* seccomp */
1537                         DEBUG("Container \"%s\" violated its seccomp policy.", name);
1538                         break;
1539                 default:
1540                         DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status));
1541                         break;
1542                 }
1543         }
1544
1545         DEBUG("Pushing physical nics back to host namespace");
1546         lxc_restore_phys_nics_to_netns(handler->netnsfd, handler->conf);
1547
1548         DEBUG("Tearing down virtual network devices used by container \"%s\".", name);
1549         removed_all_netdevs = lxc_delete_network(handler);
1550
1551         if (handler->pinfd >= 0) {
1552                 close(handler->pinfd);
1553                 handler->pinfd = -1;
1554         }
1555
1556         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1557         err =  lxc_error_set_and_log(handler->pid, status);
1558 out_fini:
1559         if (!removed_all_netdevs) {
1560                 DEBUG("Failed tearing down network devices used by container. Trying again!");
1561                 removed_all_netdevs = lxc_delete_network(handler);
1562                 if (!removed_all_netdevs)
1563                         DEBUG("Failed tearing down network devices used by container. Not trying again!");
1564         }
1565
1566 out_detach_blockdev:
1567         detach_block_device(handler->conf);
1568
1569 out_fini_nonet:
1570         lxc_fini(name, handler);
1571         return err;
1572
1573 out_abort:
1574         lxc_abort(name, handler);
1575         goto out_fini;
1576 }
1577
1578 struct start_args {
1579         char *const *argv;
1580 };
1581
1582 static int start(struct lxc_handler *handler, void* data)
1583 {
1584         struct start_args *arg = data;
1585
1586         NOTICE("Exec'ing \"%s\".", arg->argv[0]);
1587
1588         execvp(arg->argv[0], arg->argv);
1589         SYSERROR("Failed to exec \"%s\".", arg->argv[0]);
1590         return 0;
1591 }
1592
1593 static int post_start(struct lxc_handler *handler, void* data)
1594 {
1595         struct start_args *arg = data;
1596
1597         NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid);
1598         return 0;
1599 }
1600
1601 static struct lxc_operations start_ops = {
1602         .start = start,
1603         .post_start = post_start
1604 };
1605
1606 int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler,
1607               const char *lxcpath, bool backgrounded)
1608 {
1609         struct start_args start_arg = {
1610                 .argv = argv,
1611         };
1612
1613         handler->conf->need_utmp_watch = 1;
1614         return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded);
1615 }
1616
1617 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1618                                             const char *name)
1619 {
1620         char destroy[MAXPATHLEN];
1621         bool bret = true;
1622         int ret = 0;
1623         struct lxc_container *c;
1624         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1625                 bret = do_destroy_container(handler->conf);
1626                 if (!bret) {
1627                         ERROR("Error destroying rootfs for container \"%s\".", name);
1628                         return;
1629                 }
1630         }
1631         INFO("Destroyed rootfs for container \"%s\".", name);
1632
1633         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1634         if (ret < 0 || ret >= MAXPATHLEN) {
1635                 ERROR("Error destroying directory for container \"%s\".", name);
1636                 return;
1637         }
1638
1639         c = lxc_container_new(name, handler->lxcpath);
1640         if (c) {
1641                 if (container_disk_lock(c)) {
1642                         INFO("Could not update lxc_snapshots file.");
1643                         lxc_container_put(c);
1644                 } else {
1645                         mod_all_rdeps(c, false);
1646                         container_disk_unlock(c);
1647                         lxc_container_put(c);
1648                 }
1649         }
1650
1651         if (am_unpriv())
1652                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper,
1653                                     destroy, "lxc_rmdir_onedev_wrapper");
1654         else
1655                 ret = lxc_rmdir_onedev(destroy, NULL);
1656
1657         if (ret < 0) {
1658                 ERROR("Error destroying directory for container \"%s\".", name);
1659                 return;
1660         }
1661         INFO("Destroyed directory for container \"%s\".", name);
1662 }
1663
1664 static int lxc_rmdir_onedev_wrapper(void *data)
1665 {
1666         char *arg = (char *) data;
1667         return lxc_rmdir_onedev(arg, NULL);
1668 }
1669
1670 static bool do_destroy_container(struct lxc_conf *conf) {
1671         if (am_unpriv()) {
1672                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf,
1673                                   "bdev_destroy_wrapper") < 0)
1674                         return false;
1675                 return true;
1676         }
1677         return bdev_destroy(conf);
1678 }