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  * Serge Hallyn <serge@hallyn.com>
   9  * Christian Brauner <christian.brauner@ubuntu.com>
  10  *
  11  * This library is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2.1 of the License, or (at your option) any later version.
  15  *
  16  * This library is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with this library; if not, write to the Free Software
  23  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  24  */
  25
  26 #define _GNU_SOURCE
  27 #include "config.h"
  28
  29 #include <alloca.h>
  30 #include <dirent.h>
  31 #include <errno.h>
  32 #include <fcntl.h>
  33 #include <grp.h>
  34 #include <poll.h>
  35 #include <signal.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <string.h>
  39 #include <unistd.h>
  40 #include <sys/file.h>
  41 #include <sys/mount.h>
  42 #include <sys/param.h>
  43 #include <sys/prctl.h>
  44 #include <sys/socket.h>
  45 #include <sys/stat.h>
  46 #include <sys/syscall.h>
  47 #include <sys/types.h>
  48 #include <sys/un.h>
  49 #include <sys/wait.h>
  50
  51 #if HAVE_LIBCAP
  52 #include <sys/capability.h>
  53 #endif
  54
  55 #if !HAVE_DECL_PR_CAPBSET_DROP
  56 #define PR_CAPBSET_DROP 24
  57 #endif
  58
  59 #if !HAVE_DECL_PR_SET_NO_NEW_PRIVS
  60 #define PR_SET_NO_NEW_PRIVS 38
  61 #endif
  62
  63 #if !HAVE_DECL_PR_GET_NO_NEW_PRIVS
  64 #define PR_GET_NO_NEW_PRIVS 39
  65 #endif
  66
  67 #include "af_unix.h"
  68 #include "caps.h"
  69 #include "cgroup.h"
  70 #include "commands.h"
  71 #include "commands_utils.h"
  72 #include "conf.h"
  73 #include "confile_utils.h"
  74 #include "console.h"
  75 #include "error.h"
  76 #include "log.h"
  77 #include "lxccontainer.h"
  78 #include "lxclock.h"
  79 #include "lxcseccomp.h"
  80 #include "mainloop.h"
  81 #include "monitor.h"
  82 #include "namespace.h"
  83 #include "network.h"
  84 #include "start.h"
  85 #include "storage.h"
  86 #include "storage_utils.h"
  87 #include "sync.h"
  88 #include "utils.h"
  89 #include "lsm/lsm.h"
  90
  91 lxc_log_define(lxc_start, lxc);
  92
  93 extern void mod_all_rdeps(struct lxc_container *c, bool inc);
  94 static bool do_destroy_container(struct lxc_handler *handler);
  95 static int lxc_rmdir_onedev_wrapper(void *data);
  96 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
  97                                             const char *name);
  98
  99 static void print_top_failing_dir(const char *path)
 100 {
 101         size_t len = strlen(path);
 102         char *copy = alloca(len + 1), *p, *e, saved;
 103         strcpy(copy, path);
 104
 105         p = copy;
 106         e = copy + len;
 107         while (p < e) {
 108                 while (p < e && *p == '/')
 109                         p++;
 110                 while (p < e && *p != '/')
 111                         p++;
 112                 saved = *p;
 113                 *p = '\0';
 114                 if (access(copy, X_OK)) {
 115                         SYSERROR("Could not access %s. Please grant it x "
 116                                  "access, or add an ACL for the container "
 117                                  "root.", copy);
 118                         return;
 119                 }
 120                 *p = saved;
 121         }
 122 }
 123
 124 static void close_ns(int ns_fd[LXC_NS_MAX])
 125 {
 126         int i;
 127
 128         for (i = 0; i < LXC_NS_MAX; i++) {
 129                 if (ns_fd[i] > -1) {
 130                         close(ns_fd[i]);
 131                         ns_fd[i] = -1;
 132                 }
 133         }
 134 }
 135
 136 /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
 137  * in clone_flags.
 138  * Return true on success, false on failure.
 139  */
 140 static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid)
 141 {
 142         int i, ret;
 143
 144         for (i = 0; i < LXC_NS_MAX; i++)
 145                 ns_fd[i] = -1;
 146
 147         ret = lxc_preserve_ns(pid, "");
 148         if (ret < 0) {
 149                 SYSERROR("Kernel does not support attaching to namespaces.");
 150                 return false;
 151         } else {
 152                 close(ret);
 153         }
 154
 155         for (i = 0; i < LXC_NS_MAX; i++) {
 156                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 157                         continue;
 158
 159                 ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name);
 160                 if (ns_fd[i] < 0)
 161                         goto error;
 162
 163                 DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, ns_fd[i]);
 164         }
 165
 166         return true;
 167
 168 error:
 169         if (errno == ENOENT)
 170                 SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name);
 171         else
 172                 SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno));
 173         close_ns(ns_fd);
 174         return false;
 175 }
 176
 177 static int match_fd(int fd)
 178 {
 179         return (fd == 0 || fd == 1 || fd == 2);
 180 }
 181
 182 int lxc_check_inherited(struct lxc_conf *conf, bool closeall,
 183                         int *fds_to_ignore, size_t len_fds)
 184 {
 185         struct dirent *direntp;
 186         int fd, fddir;
 187         size_t i;
 188         DIR *dir;
 189
 190         if (conf && conf->close_all_fds)
 191                 closeall = true;
 192
 193 restart:
 194         dir = opendir("/proc/self/fd");
 195         if (!dir) {
 196                 WARN("Failed to open directory: %s.", strerror(errno));
 197                 return -1;
 198         }
 199
 200         fddir = dirfd(dir);
 201
 202         while ((direntp = readdir(dir))) {
 203                 struct lxc_list *cur;
 204                 bool matched = false;
 205
 206                 if (!direntp)
 207                         break;
 208
 209                 if (!strcmp(direntp->d_name, "."))
 210                         continue;
 211
 212                 if (!strcmp(direntp->d_name, ".."))
 213                         continue;
 214
 215                 if (lxc_safe_int(direntp->d_name, &fd) < 0) {
 216                         INFO("Could not parse file descriptor for: %s", direntp->d_name);
 217                         continue;
 218                 }
 219
 220                 for (i = 0; i < len_fds; i++)
 221                         if (fds_to_ignore[i] == fd)
 222                                 break;
 223
 224                 if (fd == fddir || fd == lxc_log_fd ||
 225                     (i < len_fds && fd == fds_to_ignore[i]))
 226                         continue;
 227
 228                 /* Keep state clients that wait on reboots. */
 229                 if (conf) {
 230                         lxc_list_for_each(cur, &conf->state_clients) {
 231                                 struct lxc_state_client *client = cur->elem;
 232
 233                                 if (client->clientfd != fd)
 234                                         continue;
 235
 236                                 matched = true;
 237                                 break;
 238                         }
 239                 }
 240
 241                 if (matched)
 242                         continue;
 243
 244                 if (current_config && fd == current_config->logfd)
 245                         continue;
 246
 247                 if (match_fd(fd))
 248                         continue;
 249
 250                 if (closeall) {
 251                         close(fd);
 252                         closedir(dir);
 253                         INFO("Closed inherited fd %d", fd);
 254                         goto restart;
 255                 }
 256                 WARN("Inherited fd %d", fd);
 257         }
 258
 259         /* Only enable syslog at this point to avoid the above logging function
 260          * to open a new fd and make the check_inherited function enter an
 261          * infinite loop.
 262          */
 263         lxc_log_enable_syslog();
 264
 265         closedir(dir); /* cannot fail */
 266         return 0;
 267 }
 268
 269 static int setup_signal_fd(sigset_t *oldmask)
 270 {
 271         sigset_t mask;
 272         int fd;
 273
 274         /* Block everything except serious error signals. */
 275         if (sigfillset(&mask) ||
 276             sigdelset(&mask, SIGILL) ||
 277             sigdelset(&mask, SIGSEGV) ||
 278             sigdelset(&mask, SIGBUS) ||
 279             sigdelset(&mask, SIGWINCH) ||
 280             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 281                 SYSERROR("Failed to set signal mask.");
 282                 return -1;
 283         }
 284
 285         fd = signalfd(-1, &mask, 0);
 286         if (fd < 0) {
 287                 SYSERROR("Failed to create signal file descriptor.");
 288                 return -1;
 289         }
 290
 291         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 292                 SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd);
 293                 close(fd);
 294                 return -1;
 295         }
 296
 297         DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd);
 298
 299         return fd;
 300 }
 301
 302 static int signal_handler(int fd, uint32_t events, void *data,
 303                           struct lxc_epoll_descr *descr)
 304 {
 305         struct signalfd_siginfo siginfo;
 306         siginfo_t info;
 307         int ret;
 308         pid_t *pid = data;
 309         bool init_died = false;
 310
 311         ret = read(fd, &siginfo, sizeof(siginfo));
 312         if (ret < 0) {
 313                 ERROR("Failed to read signal info from signal file descriptor: %d.", fd);
 314                 return -1;
 315         }
 316
 317         if (ret != sizeof(siginfo)) {
 318                 ERROR("Unexpected size for siginfo struct.");
 319                 return -1;
 320         }
 321
 322         /* Check whether init is running. */
 323         info.si_pid = 0;
 324         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 325         if (ret == 0 && info.si_pid == *pid)
 326                 init_died = true;
 327
 328         if (siginfo.ssi_signo != SIGCHLD) {
 329                 kill(*pid, siginfo.ssi_signo);
 330                 INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid);
 331                 return init_died ? 1 : 0;
 332         }
 333
 334         if (siginfo.ssi_code == CLD_STOPPED) {
 335                 INFO("Container init process was stopped.");
 336                 return init_died ? 1 : 0;
 337         } else if (siginfo.ssi_code == CLD_CONTINUED) {
 338                 INFO("Container init process was continued.");
 339                 return init_died ? 1 : 0;
 340         }
 341
 342         /* More robustness, protect ourself from a SIGCHLD sent
 343          * by a process different from the container init.
 344          */
 345         if (siginfo.ssi_pid != *pid) {
 346                 NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid);
 347                 return init_died ? 1 : 0;
 348         }
 349
 350         DEBUG("Container init process %d exited.", *pid);
 351         return 1;
 352 }
 353
 354 static int lxc_serve_state_clients(const char *name,
 355                                    struct lxc_handler *handler,
 356                                    lxc_state_t state)
 357 {
 358         ssize_t ret;
 359         struct lxc_list *cur, *next;
 360         struct lxc_state_client *client;
 361         struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 362
 363         process_lock();
 364         handler->state = state;
 365         TRACE("Set container state to %s", lxc_state2str(state));
 366
 367         if (lxc_list_empty(&handler->conf->state_clients)) {
 368                 TRACE("No state clients registered");
 369                 process_unlock();
 370                 lxc_monitor_send_state(name, state, handler->lxcpath);
 371                 return 0;
 372         }
 373
 374         strncpy(msg.name, name, sizeof(msg.name));
 375         msg.name[sizeof(msg.name) - 1] = 0;
 376
 377         lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) {
 378                 client = cur->elem;
 379
 380                 if (client->states[state] == 0) {
 381                         TRACE("State %s not registered for state client %d",
 382                               lxc_state2str(state), client->clientfd);
 383                         continue;
 384                 }
 385
 386                 TRACE("Sending state %s to state client %d",
 387                       lxc_state2str(state), client->clientfd);
 388
 389         again:
 390                 ret = send(client->clientfd, &msg, sizeof(msg), 0);
 391                 if (ret <= 0) {
 392                         if (errno == EINTR) {
 393                                 TRACE("Caught EINTR; retrying");
 394                                 goto again;
 395                         }
 396
 397                         ERROR("%s - Failed to send message to client",
 398                               strerror(errno));
 399                 }
 400
 401                 /* kick client from list */
 402                 close(client->clientfd);
 403                 lxc_list_del(cur);
 404                 free(cur->elem);
 405                 free(cur);
 406         }
 407         process_unlock();
 408
 409         return 0;
 410 }
 411
 412 static int lxc_serve_state_socket_pair(const char *name,
 413                                        struct lxc_handler *handler,
 414                                        lxc_state_t state)
 415 {
 416         ssize_t ret;
 417
 418         if (!handler->backgrounded ||
 419             handler->state_socket_pair[1] < 0 ||
 420             state == STARTING)
 421                 return 0;
 422
 423         /* Close read end of the socket pair. */
 424         close(handler->state_socket_pair[0]);
 425         handler->state_socket_pair[0] = -1;
 426
 427 again:
 428         ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1],
 429                                                 &(int){state}, sizeof(int));
 430         if (ret != sizeof(int)) {
 431                 if (errno == EINTR)
 432                         goto again;
 433                 SYSERROR("Failed to send state to %d",
 434                          handler->state_socket_pair[1]);
 435                 return -1;
 436         }
 437
 438         TRACE("Sent container state \"%s\" to %d", lxc_state2str(state),
 439               handler->state_socket_pair[1]);
 440
 441         /* Close write end of the socket pair. */
 442         close(handler->state_socket_pair[1]);
 443         handler->state_socket_pair[1] = -1;
 444
 445         return 0;
 446 }
 447
 448 int lxc_set_state(const char *name, struct lxc_handler *handler,
 449                   lxc_state_t state)
 450 {
 451         int ret;
 452
 453         ret = lxc_serve_state_socket_pair(name, handler, state);
 454         if (ret < 0) {
 455                 ERROR("Failed to synchronize via anonymous pair of unix sockets");
 456                 return -1;
 457         }
 458
 459         ret = lxc_serve_state_clients(name, handler, state);
 460         if (ret < 0)
 461                 return -1;
 462
 463         /* This function will try to connect to the legacy lxc-monitord state
 464          * server and only exists for backwards compatibility.
 465          */
 466         lxc_monitor_send_state(name, state, handler->lxcpath);
 467
 468         return 0;
 469 }
 470
 471 int lxc_poll(const char *name, struct lxc_handler *handler)
 472 {
 473         int sigfd = handler->sigfd;
 474         int pid = handler->pid;
 475         struct lxc_epoll_descr descr;
 476
 477         if (lxc_mainloop_open(&descr)) {
 478                 ERROR("Failed to create LXC mainloop.");
 479                 goto out_sigfd;
 480         }
 481
 482         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 483                 ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd);
 484                 goto out_mainloop_open;
 485         }
 486
 487         if (lxc_console_mainloop_add(&descr, handler->conf)) {
 488                 ERROR("Failed to add console handler to LXC mainloop.");
 489                 goto out_mainloop_open;
 490         }
 491
 492         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 493                 ERROR("Failed to add command handler to LXC mainloop.");
 494                 goto out_mainloop_open;
 495         }
 496
 497         TRACE("lxc mainloop is ready");
 498
 499         return lxc_mainloop(&descr, -1);
 500
 501 out_mainloop_open:
 502         lxc_mainloop_close(&descr);
 503
 504 out_sigfd:
 505         close(sigfd);
 506
 507         return -1;
 508 }
 509
 510 void lxc_free_handler(struct lxc_handler *handler)
 511 {
 512         if (handler->conf && handler->conf->maincmd_fd)
 513                 close(handler->conf->maincmd_fd);
 514
 515         if (handler->state_socket_pair[0] >= 0)
 516                 close(handler->state_socket_pair[0]);
 517
 518         if (handler->state_socket_pair[1] >= 0)
 519                 close(handler->state_socket_pair[1]);
 520
 521         handler->conf = NULL;
 522         free(handler);
 523 }
 524
 525 struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf,
 526                                      const char *lxcpath, bool daemonize)
 527 {
 528         int i, ret;
 529         struct lxc_handler *handler;
 530
 531         handler = malloc(sizeof(*handler));
 532         if (!handler) {
 533                 ERROR("failed to allocate memory");
 534                 return NULL;
 535         }
 536
 537         memset(handler, 0, sizeof(*handler));
 538
 539         /* Note that am_unpriv() checks the effective uid. We probably don't
 540          * care if we are real root only if we are running as root so this
 541          * should be fine.
 542          */
 543         handler->am_root = !am_unpriv();
 544         handler->data_sock[0] = handler->data_sock[1] = -1;
 545         handler->conf = conf;
 546         handler->lxcpath = lxcpath;
 547         handler->pinfd = -1;
 548         handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1;
 549         if (handler->conf->reboot == 0)
 550                 lxc_list_init(&handler->conf->state_clients);
 551
 552         for (i = 0; i < LXC_NS_MAX; i++)
 553                 handler->nsfd[i] = -1;
 554
 555         handler->name = name;
 556
 557         if (daemonize && !handler->conf->reboot) {
 558                 /* Create socketpair() to synchronize on daemonized startup.
 559                  * When the container reboots we don't need to synchronize again
 560                  * currently so don't open another socketpair().
 561                  */
 562                 ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0,
 563                                  handler->state_socket_pair);
 564                 if (ret < 0) {
 565                         ERROR("Failed to create anonymous pair of unix sockets");
 566                         goto on_error;
 567                 }
 568                 TRACE("Created anonymous pair {%d,%d} of unix sockets",
 569                       handler->state_socket_pair[0],
 570                       handler->state_socket_pair[1]);
 571         }
 572
 573         if (handler->conf->reboot == 0) {
 574                 handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command");
 575                 if (handler->conf->maincmd_fd < 0) {
 576                         ERROR("Failed to set up command socket");
 577                         goto on_error;
 578                 }
 579         }
 580         TRACE("Unix domain socket %d for command server is ready",
 581               handler->conf->maincmd_fd);
 582
 583         return handler;
 584
 585 on_error:
 586         lxc_free_handler(handler);
 587
 588         return NULL;
 589 }
 590
 591 int lxc_init(const char *name, struct lxc_handler *handler)
 592 {
 593         const char *loglevel;
 594         struct lxc_conf *conf = handler->conf;
 595
 596         lsm_init();
 597         TRACE("initialized LSM");
 598
 599         if (lxc_read_seccomp_config(conf) != 0) {
 600                 ERROR("Failed loading seccomp policy.");
 601                 goto out_close_maincmd_fd;
 602         }
 603         TRACE("read seccomp policy");
 604
 605         /* Begin by setting the state to STARTING. */
 606         if (lxc_set_state(name, handler, STARTING)) {
 607                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING));
 608                 goto out_close_maincmd_fd;
 609         }
 610         TRACE("set container state to \"STARTING\"");
 611
 612         /* Start of environment variable setup for hooks. */
 613         if (name && setenv("LXC_NAME", name, 1))
 614                 SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name);
 615
 616         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1))
 617                 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile);
 618
 619         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1))
 620                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount);
 621
 622         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1))
 623                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path);
 624
 625         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1))
 626                 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path);
 627
 628         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1))
 629                 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path);
 630
 631         if (setenv("LXC_CGNS_AWARE", "1", 1))
 632                 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1.");
 633
 634         loglevel = lxc_log_priority_to_string(lxc_log_get_level());
 635         if (setenv("LXC_LOG_LEVEL", loglevel, 1))
 636                 SYSERROR("Failed to set environment variable LXC_LOG_LEVEL=%s", loglevel);
 637         /* End of environment variable setup for hooks. */
 638
 639         TRACE("set environment variables");
 640
 641         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 642                 ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name);
 643                 goto out_aborting;
 644         }
 645         TRACE("ran pre-start hooks");
 646
 647         /* The signal fd has to be created before forking otherwise if the child
 648          * process exits before we setup the signal fd, the event will be lost
 649          * and the command will be stuck.
 650          */
 651         handler->sigfd = setup_signal_fd(&handler->oldmask);
 652         if (handler->sigfd < 0) {
 653                 ERROR("Failed to setup SIGCHLD fd handler.");
 654                 goto out_delete_tty;
 655         }
 656         TRACE("set up signal fd");
 657
 658         /* Do this after setting up signals since it might unblock SIGWINCH. */
 659         if (lxc_console_create(conf)) {
 660                 ERROR("Failed to create console for container \"%s\".", name);
 661                 goto out_restore_sigmask;
 662         }
 663         TRACE("created console");
 664
 665         if (lxc_ttys_shift_ids(conf) < 0) {
 666                 ERROR("Failed to shift tty into container.");
 667                 goto out_restore_sigmask;
 668         }
 669         TRACE("shifted tty ids");
 670
 671         INFO("container \"%s\" is initialized", name);
 672         return 0;
 673
 674 out_restore_sigmask:
 675         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 676 out_delete_tty:
 677         lxc_delete_tty(&conf->tty_info);
 678 out_aborting:
 679         lxc_set_state(name, handler, ABORTING);
 680 out_close_maincmd_fd:
 681         close(conf->maincmd_fd);
 682         conf->maincmd_fd = -1;
 683         return -1;
 684 }
 685
 686 void lxc_fini(const char *name, struct lxc_handler *handler)
 687 {
 688         int i, rc;
 689         struct lxc_list *cur, *next;
 690         pid_t self = getpid();
 691         char *namespaces[LXC_NS_MAX + 1];
 692         size_t namespace_count = 0;
 693
 694         /* The STOPPING state is there for future cleanup code which can take
 695          * awhile.
 696          */
 697         lxc_set_state(name, handler, STOPPING);
 698
 699         for (i = 0; i < LXC_NS_MAX; i++) {
 700                 if (handler->nsfd[i] != -1) {
 701                         rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d",
 702                                       ns_info[i].proc_name, self, handler->nsfd[i]);
 703                         if (rc == -1) {
 704                                 SYSERROR("Failed to allocate memory.");
 705                                 break;
 706                         }
 707                         ++namespace_count;
 708                 }
 709         }
 710         namespaces[namespace_count] = NULL;
 711
 712         if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1))
 713                 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot.");
 714
 715         if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1))
 716                 SYSERROR("Failed to set environment variable: LXC_TARGET=stop.");
 717
 718         if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces))
 719                 ERROR("Failed to run lxc.hook.stop for container \"%s\".", name);
 720
 721         while (namespace_count--)
 722                 free(namespaces[namespace_count]);
 723
 724         for (i = 0; i < LXC_NS_MAX; i++) {
 725                 if (handler->nsfd[i] < 0)
 726                         continue;
 727
 728                 close(handler->nsfd[i]);
 729                 handler->nsfd[i] = -1;
 730         }
 731
 732         cgroup_destroy(handler);
 733
 734         lxc_set_state(name, handler, STOPPED);
 735
 736         if (handler->conf->reboot == 0) {
 737                 /* close command socket */
 738                 close(handler->conf->maincmd_fd);
 739                 handler->conf->maincmd_fd = -1;
 740         }
 741
 742         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) {
 743                 ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name);
 744                 if (handler->conf->reboot) {
 745                         WARN("Container will be stopped instead of rebooted.");
 746                         handler->conf->reboot = 0;
 747                         if (setenv("LXC_TARGET", "stop", 1))
 748                                 WARN("Failed to set environment variable: LXC_TARGET=stop.");
 749                 }
 750         }
 751
 752         /* Reset mask set by setup_signal_fd. */
 753         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 754                 WARN("Failed to restore signal mask.");
 755
 756         lxc_console_delete(&handler->conf->console);
 757         lxc_delete_tty(&handler->conf->tty_info);
 758
 759         /* The command socket is now closed, no more state clients can register
 760          * themselves from now on. So free the list of state clients.
 761          */
 762         lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) {
 763                 struct lxc_state_client *client = cur->elem;
 764
 765                 /* Keep state clients that want to be notified about reboots. */
 766                 if ((handler->conf->reboot > 0) && (client->states[RUNNING] == 2))
 767                         continue;
 768
 769                 /* close state client socket */
 770                 close(client->clientfd);
 771                 lxc_list_del(cur);
 772                 free(cur->elem);
 773                 free(cur);
 774         }
 775
 776         if (handler->data_sock[0] != -1) {
 777                 close(handler->data_sock[0]);
 778                 close(handler->data_sock[1]);
 779         }
 780
 781         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1)
 782                 lxc_destroy_container_on_signal(handler, name);
 783
 784         free(handler);
 785 }
 786
 787 void lxc_abort(const char *name, struct lxc_handler *handler)
 788 {
 789         int ret, status;
 790
 791         lxc_set_state(name, handler, ABORTING);
 792         if (handler->pid > 0)
 793                 kill(handler->pid, SIGKILL);
 794         while ((ret = waitpid(-1, &status, 0)) > 0) {
 795                 ;
 796         }
 797 }
 798
 799 static int do_start(void *data)
 800 {
 801         int ret;
 802         struct lxc_list *iterator;
 803         char path[PATH_MAX];
 804         int devnull_fd = -1;
 805         struct lxc_handler *handler = data;
 806         bool have_cap_setgid;
 807         uid_t new_uid;
 808         gid_t new_gid;
 809
 810         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 811                 SYSERROR("Failed to set signal mask.");
 812                 return -1;
 813         }
 814
 815         /* This prctl must be before the synchro, so if the parent dies before
 816          * we set the parent death signal, we will detect its death with the
 817          * synchro right after, otherwise we have a window where the parent can
 818          * exit before we set the pdeath signal leading to a unsupervized
 819          * container.
 820          */
 821         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 822                 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL.");
 823                 return -1;
 824         }
 825
 826         lxc_sync_fini_parent(handler);
 827
 828         /* Don't leak the pinfd to the container. */
 829         if (handler->pinfd >= 0)
 830                 close(handler->pinfd);
 831
 832         if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP))
 833                 return -1;
 834
 835         /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
 836          * https://github.com/lxc/lxd/issues/1978.
 837          */
 838         if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) ==
 839             (CLONE_NEWNET | CLONE_NEWUSER)) {
 840                 ret = unshare(CLONE_NEWNET);
 841                 if (ret < 0) {
 842                         SYSERROR("Failed to unshare CLONE_NEWNET.");
 843                         goto out_warn_father;
 844                 }
 845                 INFO("Unshared CLONE_NEWNET.");
 846         }
 847
 848         /* Tell the parent task it can begin to configure the container and wait
 849          * for it to finish.
 850          */
 851         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 852                 return -1;
 853
 854         if (lxc_network_recv_veth_names_from_parent(handler) < 0) {
 855                 ERROR("Failed to receive veth names from parent");
 856                 goto out_warn_father;
 857         }
 858
 859         /* If we are in a new user namespace, become root there to have
 860          * privilege over our namespace.
 861          */
 862         if (!lxc_list_empty(&handler->conf->id_map)) {
 863                 if (lxc_switch_uid_gid(0, 0) < 0)
 864                         goto out_warn_father;
 865
 866                 /* Drop groups only after we switched to a valid gid in the new
 867                  * user namespace.
 868                  */
 869                 if (lxc_setgroups(0, NULL) < 0)
 870                         goto out_warn_father;
 871         }
 872
 873         if (access(handler->lxcpath, X_OK)) {
 874                 print_top_failing_dir(handler->lxcpath);
 875                 goto out_warn_father;
 876         }
 877
 878         ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount);
 879         if (ret < 0 || ret >= sizeof(path))
 880                 goto out_warn_father;
 881
 882         /* In order to checkpoint restore, we need to have everything in the
 883          * same mount namespace. However, some containers may not have a
 884          * reasonable /dev (in particular, they may not have /dev/null), so we
 885          * can't set init's std fds to /dev/null by opening it from inside the
 886          * container.
 887          *
 888          * If that's the case, fall back to using the host's /dev/null. This
 889          * means that migration won't work, but at least we won't spew output
 890          * where it isn't wanted.
 891          */
 892         if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) {
 893                 devnull_fd = open_devnull();
 894
 895                 if (devnull_fd < 0)
 896                         goto out_warn_father;
 897                 WARN("Using /dev/null from the host for container init's "
 898                      "standard file descriptors. Migration will not work.");
 899         }
 900
 901         /* Ask father to setup cgroups and wait for him to finish. */
 902         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 903                 goto out_error;
 904
 905         /* Unshare cgroup namespace after we have setup our cgroups. If we do it
 906          * earlier we end up with a wrong view of /proc/self/cgroup. For
 907          * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
 908          * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
 909          * /proc/self/cgroup would show us:
 910          *
 911          *      8:cpuset:/lxc/c
 912          *
 913          * whereas it should actually show
 914          *
 915          *      8:cpuset:/
 916          */
 917         if (cgns_supported()) {
 918                 if (unshare(CLONE_NEWCGROUP) < 0) {
 919                         INFO("Failed to unshare CLONE_NEWCGROUP.");
 920                         goto out_warn_father;
 921                 }
 922                 INFO("Unshared CLONE_NEWCGROUP.");
 923         }
 924
 925         /* Add the requested environment variables to the current environment to
 926          * allow them to be used by the various hooks, such as the start hook
 927          * above.
 928          */
 929         lxc_list_for_each(iterator, &handler->conf->environment) {
 930                 if (putenv((char *)iterator->elem)) {
 931                         SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem);
 932                         goto out_warn_father;
 933                 }
 934         }
 935
 936         /* Setup the container, ip, names, utsname, ... */
 937         ret = lxc_setup(handler);
 938         close(handler->data_sock[0]);
 939         close(handler->data_sock[1]);
 940         if (ret < 0) {
 941                 ERROR("Failed to setup container \"%s\".", handler->name);
 942                 goto out_warn_father;
 943         }
 944
 945         /* Set the label to change to when we exec(2) the container's init. */
 946         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 947                 goto out_warn_father;
 948
 949         /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
 950          * before we aren't allowed anymore.
 951          */
 952         if (handler->conf->no_new_privs) {
 953                 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
 954                         SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 955                         goto out_warn_father;
 956                 }
 957                 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 958         }
 959
 960         /* Some init's such as busybox will set sane tty settings on stdin,
 961          * stdout, stderr which it thinks is the console. We already set them
 962          * the way we wanted on the real terminal, and we want init to do its
 963          * setup on its console ie. the pty allocated in lxc_console_create() so
 964          * make sure that that pty is stdin,stdout,stderr.
 965          */
 966          if (handler->conf->console.slave >= 0)
 967                  if (set_stdfds(handler->conf->console.slave) < 0) {
 968                         ERROR("Failed to redirect std{in,out,err} to pty file "
 969                               "descriptor %d",
 970                               handler->conf->console.slave);
 971                         goto out_warn_father;
 972                  }
 973
 974         /* If we mounted a temporary proc, then unmount it now. */
 975         tmp_proc_unmount(handler->conf);
 976
 977         if (lxc_seccomp_load(handler->conf) != 0)
 978                 goto out_warn_father;
 979
 980         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 981                 ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name);
 982                 goto out_warn_father;
 983         }
 984
 985         close(handler->sigfd);
 986
 987         if (devnull_fd < 0) {
 988                 devnull_fd = open_devnull();
 989
 990                 if (devnull_fd < 0)
 991                         goto out_warn_father;
 992         }
 993
 994         if (handler->conf->console.slave < 0 && handler->backgrounded)
 995                 if (set_stdfds(devnull_fd) < 0) {
 996                         ERROR("Failed to redirect std{in,out,err} to "
 997                               "\"/dev/null\"");
 998                         goto out_warn_father;
 999                 }
1000
1001         if (devnull_fd >= 0) {
1002                 close(devnull_fd);
1003                 devnull_fd = -1;
1004         }
1005
1006         setsid();
1007
1008         if (handler->conf->init_cwd && chdir(handler->conf->init_cwd)) {
1009                 SYSERROR("Could not change directory to \"%s\"", handler->conf->init_cwd);
1010                 goto out_warn_father;
1011         }
1012
1013         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS))
1014                 goto out_warn_father;
1015
1016         /* Reset the environment variables the user requested in a clear
1017          * environment.
1018          */
1019         if (clearenv()) {
1020                 SYSERROR("Failed to clear environment.");
1021                 /* Don't error out though. */
1022         }
1023
1024         lxc_list_for_each(iterator, &handler->conf->environment) {
1025                 if (putenv((char *)iterator->elem)) {
1026                         SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem);
1027                         goto out_warn_father;
1028                 }
1029         }
1030
1031         if (putenv("container=lxc")) {
1032                 SYSERROR("Failed to set environment variable: container=lxc.");
1033                 goto out_warn_father;
1034         }
1035
1036         if (handler->conf->pty_names) {
1037                 if (putenv(handler->conf->pty_names)) {
1038                         SYSERROR("Failed to set environment variable for container ptys.");
1039                         goto out_warn_father;
1040                 }
1041         }
1042
1043         /* The container has been setup. We can now switch to an unprivileged
1044          * uid/gid.
1045          */
1046         new_uid = handler->conf->init_uid;
1047         new_gid = handler->conf->init_gid;
1048
1049         /* If we are in a new user namespace we already dropped all
1050          * groups when we switched to root in the new user namespace
1051          * further above. Only drop groups if we can, so ensure that we
1052          * have necessary privilege.
1053          */
1054         #if HAVE_LIBCAP
1055         have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE);
1056         #else
1057         have_cap_setgid = false;
1058         #endif
1059         if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) {
1060                 if (lxc_setgroups(0, NULL) < 0)
1061                         goto out_warn_father;
1062         }
1063
1064         if (lxc_switch_uid_gid(new_uid, new_gid) < 0)
1065                 goto out_warn_father;
1066
1067         /* After this call, we are in error because this ops should not return
1068          * as it execs.
1069          */
1070         handler->ops->start(handler, handler->data);
1071
1072 out_warn_father:
1073         /* We want the parent to know something went wrong, so we return a
1074          * special error code.
1075          */
1076         lxc_sync_wake_parent(handler, LXC_SYNC_ERROR);
1077
1078 out_error:
1079         if (devnull_fd >= 0)
1080                 close(devnull_fd);
1081
1082         return -1;
1083 }
1084
1085 static int lxc_recv_ttys_from_child(struct lxc_handler *handler)
1086 {
1087         int i;
1088         struct lxc_pty_info *pty_info;
1089         int ret = -1;
1090         int sock = handler->data_sock[1];
1091         struct lxc_conf *conf = handler->conf;
1092         struct lxc_tty_info *tty_info = &conf->tty_info;
1093
1094         if (!conf->tty)
1095                 return 0;
1096
1097         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty);
1098         if (!tty_info->pty_info)
1099                 return -1;
1100
1101         for (i = 0; i < conf->tty; i++) {
1102                 int ttyfds[2];
1103
1104                 ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0);
1105                 if (ret < 0)
1106                         break;
1107
1108                 pty_info = &tty_info->pty_info[i];
1109                 pty_info->busy = 0;
1110                 pty_info->master = ttyfds[0];
1111                 pty_info->slave = ttyfds[1];
1112                 TRACE("Received pty with master fd %d and slave fd %d from "
1113                       "parent", pty_info->master, pty_info->slave);
1114         }
1115         if (ret < 0)
1116                 ERROR("Failed to receive %d ttys from child: %s", conf->tty,
1117                       strerror(errno));
1118         else
1119                 TRACE("Received %d ttys from child", conf->tty);
1120
1121         tty_info->nbtty = conf->tty;
1122
1123         return ret;
1124 }
1125
1126 void resolve_clone_flags(struct lxc_handler *handler)
1127 {
1128         handler->clone_flags = CLONE_NEWNS;
1129
1130         if (!handler->conf->inherit_ns[LXC_NS_USER]) {
1131                 if (!lxc_list_empty(&handler->conf->id_map))
1132                         handler->clone_flags |= CLONE_NEWUSER;
1133         } else {
1134                 INFO("Inheriting user namespace");
1135         }
1136
1137         if (!handler->conf->inherit_ns[LXC_NS_NET]) {
1138                 if (!lxc_requests_empty_network(handler))
1139                         handler->clone_flags |= CLONE_NEWNET;
1140         } else {
1141                 INFO("Inheriting net namespace");
1142         }
1143
1144         if (!handler->conf->inherit_ns[LXC_NS_IPC])
1145                 handler->clone_flags |= CLONE_NEWIPC;
1146         else
1147                 INFO("Inheriting ipc namespace");
1148
1149         if (!handler->conf->inherit_ns[LXC_NS_UTS])
1150                 handler->clone_flags |= CLONE_NEWUTS;
1151         else
1152                 INFO("Inheriting uts namespace");
1153
1154         if (!handler->conf->inherit_ns[LXC_NS_PID])
1155                 handler->clone_flags |= CLONE_NEWPID;
1156         else
1157                 INFO("Inheriting pid namespace");
1158 }
1159
1160 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1161  * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1162  * But since the memory between parent and child is shared on CLONE_VM this
1163  * would invalidate the getpid() cache that glibc used to maintain and so
1164  * getpid() in the child would return the parent's pid. This is all fixed in
1165  * newer glibc versions where the getpid() cache is removed and the pid/tid is
1166  * not reset anymore.
1167  * However, if for whatever reason you - dear commiter - somehow need to get the
1168  * pid of the dummy intermediate process for do_share_ns() you need to call
1169  * syscall(__NR_getpid) directly. The next lxc_clone() call does not employ
1170  * CLONE_VM and will be fine.
1171  */
1172 static inline int do_share_ns(void *arg)
1173 {
1174         int i, flags, ret;
1175         struct lxc_handler *handler = arg;
1176
1177         for (i = 0; i < LXC_NS_MAX; i++) {
1178                 if (handler->nsfd[i] < 0)
1179                         continue;
1180
1181                 ret = setns(handler->nsfd[i], 0);
1182                 if (ret < 0)
1183                         return -1;
1184
1185                 DEBUG("Inherited %s namespace", ns_info[i].proc_name);
1186         }
1187
1188         flags = handler->on_clone_flags;
1189         flags |= CLONE_PARENT;
1190         handler->pid = lxc_clone(do_start, handler, flags);
1191         if (handler->pid < 0)
1192                 return -1;
1193
1194         return 0;
1195 }
1196
1197 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1198  * exec()s the requested container binary.
1199  * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1200  * right here should be double checked if they'd pose a security risk. (For
1201  * example, any {u}mount() operations performed here will be reflected on the
1202  * host!)
1203  */
1204 static int lxc_spawn(struct lxc_handler *handler)
1205 {
1206         int i, ret;
1207         char pidstr[20];
1208         bool wants_to_map_ids;
1209         struct lxc_list *id_map;
1210         const char *name = handler->name;
1211         const char *lxcpath = handler->lxcpath;
1212         bool cgroups_connected = false, share_ns = false;
1213         struct lxc_conf *conf = handler->conf;
1214
1215         id_map = &conf->id_map;
1216         wants_to_map_ids = !lxc_list_empty(id_map);
1217
1218         for (i = 0; i < LXC_NS_MAX; i++) {
1219                 if (!conf->inherit_ns[i])
1220                         continue;
1221
1222                 handler->nsfd[i] = lxc_inherit_namespace(conf->inherit_ns[i], lxcpath, ns_info[i].proc_name);
1223                 if (handler->nsfd[i] < 0)
1224                         return -1;
1225
1226                 share_ns = true;
1227         }
1228
1229         if (lxc_sync_init(handler))
1230                 return -1;
1231
1232         ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0,
1233                          handler->data_sock);
1234         if (ret < 0) {
1235                 lxc_sync_fini(handler);
1236                 return -1;
1237         }
1238
1239         resolve_clone_flags(handler);
1240
1241         if (handler->clone_flags & CLONE_NEWNET) {
1242                 if (!lxc_list_empty(&conf->network)) {
1243
1244                         /* Find gateway addresses from the link device, which is
1245                          * no longer accessible inside the container. Do this
1246                          * before creating network interfaces, since goto
1247                          * out_delete_net does not work before lxc_clone.
1248                          */
1249                         if (lxc_find_gateway_addresses(handler)) {
1250                                 ERROR("Failed to find gateway addresses.");
1251                                 lxc_sync_fini(handler);
1252                                 return -1;
1253                         }
1254
1255                         /* That should be done before the clone because we will
1256                          * fill the netdev index and use them in the child.
1257                          */
1258                         if (lxc_create_network_priv(handler)) {
1259                                 ERROR("Failed to create the network.");
1260                                 lxc_sync_fini(handler);
1261                                 return -1;
1262                         }
1263                 }
1264         }
1265
1266         if (!cgroup_init(handler)) {
1267                 ERROR("Failed initializing cgroup support.");
1268                 goto out_delete_net;
1269         }
1270
1271         cgroups_connected = true;
1272
1273         if (!cgroup_create(handler)) {
1274                 ERROR("Failed creating cgroups.");
1275                 goto out_delete_net;
1276         }
1277
1278         /* If the rootfs is not a blockdev, prevent the container from marking
1279          * it readonly.
1280          * If the container is unprivileged then skip rootfs pinning.
1281          */
1282         if (!wants_to_map_ids) {
1283                 handler->pinfd = pin_rootfs(conf->rootfs.path);
1284                 if (handler->pinfd == -1)
1285                         INFO("Failed to pin the rootfs for container \"%s\".", handler->name);
1286         }
1287
1288         /* Create a process in a new set of namespaces. */
1289         handler->on_clone_flags = handler->clone_flags;
1290         if (handler->clone_flags & CLONE_NEWUSER) {
1291                 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1292                  * clone a new user namespace first and only later unshare our
1293                  * network namespace to ensure that network devices ownership is
1294                  * set up correctly.
1295                  */
1296                 handler->on_clone_flags &= ~CLONE_NEWNET;
1297         }
1298
1299         if (share_ns)
1300                 ret = lxc_clone(do_share_ns, handler, CLONE_VFORK | CLONE_VM | CLONE_FILES);
1301         else
1302                 handler->pid = lxc_clone(do_start, handler, handler->on_clone_flags);
1303         if (handler->pid < 0 || ret < 0) {
1304                 SYSERROR("Failed to clone a new set of namespaces.");
1305                 goto out_delete_net;
1306         }
1307         TRACE("Cloned child process %d", handler->pid);
1308
1309         for (i = 0; i < LXC_NS_MAX; i++)
1310                 if (handler->on_clone_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 & ~CLONE_NEWNET, handler->pid)) {
1314                 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1315                 goto out_delete_net;
1316         }
1317
1318         lxc_sync_fini_child(handler);
1319
1320         /* Map the container uids. The container became an invalid userid the
1321          * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1322          * anything immediately, but allows the container to setuid(0) (0 being
1323          * mapped to something else on the host.) later to become a valid uid
1324          * again.
1325          */
1326         if (wants_to_map_ids) {
1327                 if (!handler->conf->inherit_ns[LXC_NS_USER]) {
1328                         ret = lxc_map_ids(id_map, handler->pid);
1329                         if (ret < 0) {
1330                                 ERROR("Failed to set up id mapping.");
1331                                 goto out_delete_net;
1332                         }
1333                 }
1334         }
1335
1336         if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP))
1337                 goto out_delete_net;
1338
1339         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
1340                 goto out_delete_net;
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         /* Now we're ready to preserve the network namespace */
1358         ret = lxc_preserve_ns(handler->pid, "net");
1359         if (ret < 0) {
1360                 ERROR("%s - Failed to preserve net namespace", strerror(errno));
1361                 goto out_delete_net;
1362         }
1363         handler->nsfd[LXC_NS_NET] = ret;
1364         DEBUG("Preserved net namespace via fd %d", ret);
1365
1366         /* Create the network configuration. */
1367         if (handler->clone_flags & CLONE_NEWNET) {
1368                 if (lxc_network_move_created_netdev_priv(handler->lxcpath,
1369                                                          handler->name,
1370                                                          &conf->network,
1371                                                          handler->pid)) {
1372                         ERROR("Failed to create the configured network.");
1373                         goto out_delete_net;
1374                 }
1375
1376                 if (lxc_create_network_unpriv(handler->lxcpath, handler->name,
1377                                               &conf->network,
1378                                               handler->pid)) {
1379                         ERROR("Failed to create the configured network.");
1380                         goto out_delete_net;
1381                 }
1382         }
1383
1384         if (lxc_network_send_veth_names_to_child(handler) < 0) {
1385                 ERROR("Failed to send veth names to child");
1386                 goto out_delete_net;
1387         }
1388
1389         if (!lxc_list_empty(&conf->procs)) {
1390                 ret = setup_proc_filesystem(&conf->procs, handler->pid);
1391                 if (ret < 0)
1392                         goto out_delete_net;
1393         }
1394
1395         /* Tell the child to continue its initialization. We'll get
1396          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1397          */
1398         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1399                 goto out_delete_net;
1400
1401         if (!lxc_list_empty(&conf->limits) && setup_resource_limits(&conf->limits, handler->pid)) {
1402                 ERROR("failed to setup resource limits for '%s'", name);
1403                 goto out_delete_net;
1404         }
1405
1406         if (lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE))
1407                 goto out_delete_net;
1408
1409         if (!cgroup_setup_limits(handler, true)) {
1410                 ERROR("Failed to setup the devices cgroup for container \"%s\".", name);
1411                 goto out_delete_net;
1412         }
1413         TRACE("Set up cgroup device limits");
1414
1415         cgroup_disconnect();
1416         cgroups_connected = false;
1417
1418         snprintf(pidstr, 20, "%d", handler->pid);
1419         if (setenv("LXC_PID", pidstr, 1))
1420                 SYSERROR("Failed to set environment variable: LXC_PID=%s.", pidstr);
1421
1422         /* Run any host-side start hooks */
1423         if (run_lxc_hooks(name, "start-host", conf, handler->lxcpath, NULL)) {
1424                 ERROR("Failed to run lxc.hook.start-host for container \"%s\".", name);
1425                 return -1;
1426         }
1427
1428         /* Tell the child to complete its initialization and wait for it to exec
1429          * or return an error. (The child will never return
1430          * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1431          * lxc_sync_barrier_child to return success, or return a different
1432          * value, causing us to error out).
1433          */
1434         if (lxc_sync_barrier_child(handler, LXC_SYNC_READY_START))
1435                 return -1;
1436
1437         if (cgns_supported()) {
1438                 ret = lxc_preserve_ns(handler->pid, "cgroup");
1439                 if (ret < 0) {
1440                         ERROR("%s - Failed to preserve cgroup namespace", strerror(errno));
1441                         goto out_delete_net;
1442                 }
1443                 handler->nsfd[LXC_NS_CGROUP] = ret;
1444                 DEBUG("Preserved cgroup namespace via fd %d", ret);
1445         }
1446
1447         if (lxc_network_recv_name_and_ifindex_from_child(handler) < 0) {
1448                 ERROR("Failed to receive names and ifindices for network "
1449                       "devices from child");
1450                 goto out_delete_net;
1451         }
1452
1453         /* Now all networks are created, network devices are moved into place,
1454          * and the correct names and ifindeces in the respective namespaces have
1455          * been recorded. The corresponding structs have now all been filled. So
1456          * log them for debugging purposes.
1457          */
1458         lxc_log_configured_netdevs(conf);
1459
1460         /* Read tty fds allocated by child. */
1461         if (lxc_recv_ttys_from_child(handler) < 0) {
1462                 ERROR("Failed to receive tty info from child process.");
1463                 goto out_delete_net;
1464         }
1465
1466         if (handler->ops->post_start(handler, handler->data))
1467                 goto out_abort;
1468
1469         if (lxc_set_state(name, handler, RUNNING)) {
1470                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name,
1471                       lxc_state2str(RUNNING));
1472                 goto out_abort;
1473         }
1474
1475         lxc_sync_fini(handler);
1476
1477         return 0;
1478
1479 out_delete_net:
1480         if (cgroups_connected)
1481                 cgroup_disconnect();
1482
1483         if (handler->clone_flags & CLONE_NEWNET)
1484                 lxc_delete_network(handler);
1485
1486 out_abort:
1487         lxc_abort(name, handler);
1488         lxc_sync_fini(handler);
1489         if (handler->pinfd >= 0) {
1490                 close(handler->pinfd);
1491                 handler->pinfd = -1;
1492         }
1493
1494         return -1;
1495 }
1496
1497 int __lxc_start(const char *name, struct lxc_handler *handler,
1498                 struct lxc_operations* ops, void *data, const char *lxcpath,
1499                 bool backgrounded)
1500 {
1501         int status;
1502         int err = -1;
1503         struct lxc_conf *conf = handler->conf;
1504
1505         if (lxc_init(name, handler) < 0) {
1506                 ERROR("Failed to initialize container \"%s\".", name);
1507                 return -1;
1508         }
1509         handler->ops = ops;
1510         handler->data = data;
1511         handler->backgrounded = backgrounded;
1512
1513         if (!attach_block_device(handler->conf)) {
1514                 ERROR("Failed to attach block device.");
1515                 goto out_fini_nonet;
1516         }
1517
1518         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1519                 /* If the backing store is a device, mount it here and now. */
1520                 if (rootfs_is_blockdev(conf)) {
1521                         if (unshare(CLONE_NEWNS) < 0) {
1522                                 ERROR("Failed to unshare CLONE_NEWNS.");
1523                                 goto out_fini_nonet;
1524                         }
1525                         INFO("Unshared CLONE_NEWNS.");
1526
1527                         remount_all_slave();
1528                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1529                                 ERROR("Error setting up rootfs mount as root before spawn.");
1530                                 goto out_fini_nonet;
1531                         }
1532                         INFO("Set up container rootfs as host root.");
1533                 }
1534         }
1535
1536         err = lxc_spawn(handler);
1537         if (err) {
1538                 ERROR("Failed to spawn container \"%s\".", name);
1539                 goto out_detach_blockdev;
1540         }
1541         /* close parent side of data socket */
1542         close(handler->data_sock[0]);
1543         handler->data_sock[0] = -1;
1544         close(handler->data_sock[1]);
1545         handler->data_sock[1] = -1;
1546
1547         handler->conf->reboot = 0;
1548
1549         err = lxc_poll(name, handler);
1550         if (err) {
1551                 ERROR("LXC mainloop exited with error: %d.", err);
1552                 goto out_abort;
1553         }
1554
1555         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1556                 continue;
1557
1558         /* If the child process exited but was not signaled, it didn't call
1559          * reboot. This should mean it was an lxc-execute which simply exited.
1560          * In any case, treat it as a 'halt'.
1561          */
1562         if (WIFSIGNALED(status)) {
1563                 switch(WTERMSIG(status)) {
1564                 case SIGINT: /* halt */
1565                         DEBUG("Container \"%s\" is halting.", name);
1566                         break;
1567                 case SIGHUP: /* reboot */
1568                         DEBUG("Container \"%s\" is rebooting.", name);
1569                         handler->conf->reboot = 1;
1570                         break;
1571                 case SIGSYS: /* seccomp */
1572                         DEBUG("Container \"%s\" violated its seccomp policy.", name);
1573                         break;
1574                 default:
1575                         DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status));
1576                         break;
1577                 }
1578         }
1579
1580         err = lxc_restore_phys_nics_to_netns(handler);
1581         if (err < 0)
1582                 ERROR("Failed to move physical network devices back to parent "
1583                       "network namespace");
1584
1585         if (handler->pinfd >= 0) {
1586                 close(handler->pinfd);
1587                 handler->pinfd = -1;
1588         }
1589
1590         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1591         err =  lxc_error_set_and_log(handler->pid, status);
1592
1593 out_fini:
1594         lxc_delete_network(handler);
1595
1596 out_detach_blockdev:
1597         detach_block_device(handler->conf);
1598
1599 out_fini_nonet:
1600         lxc_fini(name, handler);
1601         return err;
1602
1603 out_abort:
1604         lxc_abort(name, handler);
1605         goto out_fini;
1606 }
1607
1608 struct start_args {
1609         char *const *argv;
1610 };
1611
1612 static int start(struct lxc_handler *handler, void* data)
1613 {
1614         struct start_args *arg = data;
1615
1616         NOTICE("Exec'ing \"%s\".", arg->argv[0]);
1617
1618         execvp(arg->argv[0], arg->argv);
1619         SYSERROR("Failed to exec \"%s\".", arg->argv[0]);
1620         return 0;
1621 }
1622
1623 static int post_start(struct lxc_handler *handler, void* data)
1624 {
1625         struct start_args *arg = data;
1626
1627         NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid);
1628         return 0;
1629 }
1630
1631 static struct lxc_operations start_ops = {
1632         .start = start,
1633         .post_start = post_start
1634 };
1635
1636 int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler,
1637               const char *lxcpath, bool backgrounded)
1638 {
1639         struct start_args start_arg = {
1640                 .argv = argv,
1641         };
1642
1643         return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded);
1644 }
1645
1646 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1647                                             const char *name)
1648 {
1649         char destroy[MAXPATHLEN];
1650         bool bret = true;
1651         int ret = 0;
1652         struct lxc_container *c;
1653         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1654                 bret = do_destroy_container(handler);
1655                 if (!bret) {
1656                         ERROR("Error destroying rootfs for container \"%s\".", name);
1657                         return;
1658                 }
1659         }
1660         INFO("Destroyed rootfs for container \"%s\".", name);
1661
1662         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1663         if (ret < 0 || ret >= MAXPATHLEN) {
1664                 ERROR("Error destroying directory for container \"%s\".", name);
1665                 return;
1666         }
1667
1668         c = lxc_container_new(name, handler->lxcpath);
1669         if (c) {
1670                 if (container_disk_lock(c)) {
1671                         INFO("Could not update lxc_snapshots file.");
1672                         lxc_container_put(c);
1673                 } else {
1674                         mod_all_rdeps(c, false);
1675                         container_disk_unlock(c);
1676                         lxc_container_put(c);
1677                 }
1678         }
1679
1680         if (!handler->am_root)
1681                 ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper,
1682                                        destroy, "lxc_rmdir_onedev_wrapper");
1683         else
1684                 ret = lxc_rmdir_onedev(destroy, NULL);
1685
1686         if (ret < 0) {
1687                 ERROR("Error destroying directory for container \"%s\".", name);
1688                 return;
1689         }
1690         INFO("Destroyed directory for container \"%s\".", name);
1691 }
1692
1693 static int lxc_rmdir_onedev_wrapper(void *data)
1694 {
1695         char *arg = (char *) data;
1696         return lxc_rmdir_onedev(arg, NULL);
1697 }
1698
1699 static bool do_destroy_container(struct lxc_handler *handler) {
1700         int ret;
1701
1702         if (!handler->am_root) {
1703                 ret = userns_exec_full(handler->conf, storage_destroy_wrapper,
1704                                        handler->conf, "storage_destroy_wrapper");
1705                 if (ret < 0)
1706                         return false;
1707
1708                 return true;
1709         }
1710
1711         return storage_destroy(handler->conf);
1712 }