src/lxc/start.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #include "config.h"
  25
  26 #include <stdio.h>
  27 #include <string.h>
  28 #include <stdlib.h>
  29 #include <dirent.h>
  30 #include <errno.h>
  31 #include <unistd.h>
  32 #include <signal.h>
  33 #include <fcntl.h>
  34 #include <grp.h>
  35 #include <poll.h>
  36 #include <sys/param.h>
  37 #include <sys/file.h>
  38 #include <sys/mount.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/socket.h>
  42 #include <sys/prctl.h>
  43 #include <sys/types.h>
  44 #include <sys/wait.h>
  45 #include <sys/un.h>
  46 #include <sys/syscall.h>
  47
  48 #if HAVE_SYS_CAPABILITY_H
  49 #include <sys/capability.h>
  50 #endif
  51
  52 #if !HAVE_DECL_PR_CAPBSET_DROP
  53 #define PR_CAPBSET_DROP 24
  54 #endif
  55
  56 #include "start.h"
  57 #include "conf.h"
  58 #include "log.h"
  59 #include "cgroup.h"
  60 #include "error.h"
  61 #include "af_unix.h"
  62 #include "mainloop.h"
  63 #include "utils.h"
  64 #include "lxcutmp.h"
  65 #include "monitor.h"
  66 #include "commands.h"
  67 #include "console.h"
  68 #include "sync.h"
  69 #include "namespace.h"
  70 #include "lxcseccomp.h"
  71 #include "caps.h"
  72 #include "bdev.h"
  73 #include "lsm/lsm.h"
  74
  75 lxc_log_define(lxc_start, lxc);
  76
  77 const struct ns_info ns_info[LXC_NS_MAX] = {
  78         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  79         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  80         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  81         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  82         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  83         [LXC_NS_NET] = {"net", CLONE_NEWNET}
  84 };
  85
  86 static void print_top_failing_dir(const char *path)
  87 {
  88         size_t len = strlen(path);
  89         char *copy = alloca(len+1), *p, *e, saved;
  90         strcpy(copy, path);
  91
  92         p = copy;
  93         e = copy + len;
  94         while (p < e) {
  95                 while (p < e && *p == '/') p++;
  96                 while (p < e && *p != '/') p++;
  97                 saved = *p;
  98                 *p = '\0';
  99                 if (access(copy, X_OK)) {
 100                         SYSERROR("could not access %s.  Please grant it 'x' " \
 101                               "access, or add an ACL for the container root.",
 102                               copy);
 103                         return;
 104                 }
 105                 *p = saved;
 106         }
 107 }
 108
 109 static void close_ns(int ns_fd[LXC_NS_MAX]) {
 110         int i;
 111
 112         for (i = 0; i < LXC_NS_MAX; i++) {
 113                 if (ns_fd[i] > -1) {
 114                         close(ns_fd[i]);
 115                         ns_fd[i] = -1;
 116                 }
 117         }
 118 }
 119
 120 static int preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags) {
 121         int i, saved_errno;
 122         char path[MAXPATHLEN];
 123
 124         for (i = 0; i < LXC_NS_MAX; i++)
 125                 ns_fd[i] = -1;
 126
 127         if (access("/proc/self/ns", X_OK)) {
 128                 WARN("Kernel does not support attach; preserve_ns ignored");
 129                 return 0;
 130         }
 131
 132         for (i = 0; i < LXC_NS_MAX; i++) {
 133                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 134                         continue;
 135                 snprintf(path, MAXPATHLEN, "/proc/self/ns/%s", ns_info[i].proc_name);
 136                 ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC);
 137                 if (ns_fd[i] < 0)
 138                         goto error;
 139         }
 140
 141         return 0;
 142
 143 error:
 144         saved_errno = errno;
 145         close_ns(ns_fd);
 146         errno = saved_errno;
 147         SYSERROR("failed to open '%s'", path);
 148         return -1;
 149 }
 150
 151 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 152         int i;
 153
 154         for (i = 0; i < LXC_NS_MAX; i++) {
 155                 if (ns_fd[i] < 0)
 156                         continue;
 157
 158                 if (setns(ns_fd[i], 0) != 0)
 159                         goto error;
 160         }
 161         return 0;
 162
 163 error:
 164         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 165         return -1;
 166 }
 167
 168 static int match_fd(int fd)
 169 {
 170         return (fd == 0 || fd == 1 || fd == 2);
 171 }
 172
 173 /*
 174  * Check for any fds we need to close
 175  * * if fd_to_ignore != -1, then if we find that fd open we will ignore it.
 176  * * By default we warn about open fds we find.
 177  * * If closeall is true, we will close open fds.
 178  * * If lxc-start was passed "-C", then conf->close_all_fds will be true,
 179  *     in which case we also close all open fds.
 180  * * A daemonized container will always pass closeall=true.
 181  */
 182 int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore)
 183 {
 184         struct dirent dirent, *direntp;
 185         int fd, fddir;
 186         DIR *dir;
 187
 188         if (conf && conf->close_all_fds)
 189                 closeall = true;
 190
 191 restart:
 192         dir = opendir("/proc/self/fd");
 193         if (!dir) {
 194                 WARN("failed to open directory: %m");
 195                 return -1;
 196         }
 197
 198         fddir = dirfd(dir);
 199
 200         while (!readdir_r(dir, &dirent, &direntp)) {
 201                 if (!direntp)
 202                         break;
 203
 204                 if (!strcmp(direntp->d_name, "."))
 205                         continue;
 206
 207                 if (!strcmp(direntp->d_name, ".."))
 208                         continue;
 209
 210                 fd = atoi(direntp->d_name);
 211
 212                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 213                         continue;
 214
 215                 if (current_config && fd == current_config->logfd)
 216                         continue;
 217
 218                 if (match_fd(fd))
 219                         continue;
 220
 221                 if (closeall) {
 222                         close(fd);
 223                         closedir(dir);
 224                         INFO("closed inherited fd %d", fd);
 225                         goto restart;
 226                 }
 227                 WARN("inherited fd %d", fd);
 228         }
 229
 230         closedir(dir); /* cannot fail */
 231         return 0;
 232 }
 233
 234 static int setup_signal_fd(sigset_t *oldmask)
 235 {
 236         sigset_t mask;
 237         int fd;
 238
 239         /* Block everything except serious error signals */
 240         if (sigfillset(&mask) ||
 241             sigdelset(&mask, SIGILL) ||
 242             sigdelset(&mask, SIGSEGV) ||
 243             sigdelset(&mask, SIGBUS) ||
 244             sigdelset(&mask, SIGWINCH) ||
 245             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 246                 SYSERROR("failed to set signal mask");
 247                 return -1;
 248         }
 249
 250         fd = signalfd(-1, &mask, 0);
 251         if (fd < 0) {
 252                 SYSERROR("failed to create the signal fd");
 253                 return -1;
 254         }
 255
 256         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 257                 SYSERROR("failed to set sigfd to close-on-exec");
 258                 close(fd);
 259                 return -1;
 260         }
 261
 262         DEBUG("sigchild handler set");
 263
 264         return fd;
 265 }
 266
 267 static int signal_handler(int fd, uint32_t events, void *data,
 268                            struct lxc_epoll_descr *descr)
 269 {
 270         struct signalfd_siginfo siginfo;
 271         siginfo_t info;
 272         int ret;
 273         pid_t *pid = data;
 274         bool init_died = false;
 275
 276         ret = read(fd, &siginfo, sizeof(siginfo));
 277         if (ret < 0) {
 278                 ERROR("failed to read signal info");
 279                 return -1;
 280         }
 281
 282         if (ret != sizeof(siginfo)) {
 283                 ERROR("unexpected siginfo size");
 284                 return -1;
 285         }
 286
 287         // check whether init is running
 288         info.si_pid = 0;
 289         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 290         if (ret == 0 && info.si_pid == *pid) {
 291                 init_died = true;
 292         }
 293
 294         if (siginfo.ssi_signo != SIGCHLD) {
 295                 kill(*pid, siginfo.ssi_signo);
 296                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 297                 return init_died ? 1 : 0;
 298         }
 299
 300         if (siginfo.ssi_code == CLD_STOPPED ||
 301             siginfo.ssi_code == CLD_CONTINUED) {
 302                 INFO("container init process was stopped/continued");
 303                 return init_died ? 1 : 0;
 304         }
 305
 306         /* more robustness, protect ourself from a SIGCHLD sent
 307          * by a process different from the container init
 308          */
 309         if (siginfo.ssi_pid != *pid) {
 310                 WARN("invalid pid for SIGCHLD");
 311                 return init_died ? 1 : 0;
 312         }
 313
 314         DEBUG("container init process exited");
 315         return 1;
 316 }
 317
 318 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 319 {
 320         handler->state = state;
 321         lxc_monitor_send_state(name, state, handler->lxcpath);
 322         return 0;
 323 }
 324
 325 int lxc_poll(const char *name, struct lxc_handler *handler)
 326 {
 327         int sigfd = handler->sigfd;
 328         int pid = handler->pid;
 329         struct lxc_epoll_descr descr;
 330
 331         if (lxc_mainloop_open(&descr)) {
 332                 ERROR("failed to create mainloop");
 333                 goto out_sigfd;
 334         }
 335
 336         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 337                 ERROR("failed to add handler for the signal");
 338                 goto out_mainloop_open;
 339         }
 340
 341         if (lxc_console_mainloop_add(&descr, handler)) {
 342                 ERROR("failed to add console handler to mainloop");
 343                 goto out_mainloop_open;
 344         }
 345
 346         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 347                 ERROR("failed to add command handler to mainloop");
 348                 goto out_mainloop_open;
 349         }
 350
 351         if (handler->conf->need_utmp_watch) {
 352                 #if HAVE_SYS_CAPABILITY_H
 353                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 354                         ERROR("failed to add utmp handler to mainloop");
 355                         goto out_mainloop_open;
 356                 }
 357                 #else
 358                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
 359                 #endif
 360         }
 361
 362         return lxc_mainloop(&descr, -1);
 363
 364 out_mainloop_open:
 365         lxc_mainloop_close(&descr);
 366 out_sigfd:
 367         close(sigfd);
 368         return -1;
 369 }
 370
 371 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 372 {
 373         struct lxc_handler *handler;
 374
 375         handler = malloc(sizeof(*handler));
 376         if (!handler)
 377                 return NULL;
 378
 379         memset(handler, 0, sizeof(*handler));
 380
 381         handler->ttysock[0] = handler->ttysock[1] = -1;
 382         handler->conf = conf;
 383         handler->lxcpath = lxcpath;
 384         handler->pinfd = -1;
 385
 386         lsm_init();
 387
 388         handler->name = strdup(name);
 389         if (!handler->name) {
 390                 ERROR("failed to allocate memory");
 391                 goto out_free;
 392         }
 393
 394         if (lxc_cmd_init(name, handler, lxcpath))
 395                 goto out_free_name;
 396
 397         if (lxc_read_seccomp_config(conf) != 0) {
 398                 ERROR("failed loading seccomp policy");
 399                 goto out_close_maincmd_fd;
 400         }
 401
 402         /* Begin by setting the state to STARTING */
 403         if (lxc_set_state(name, handler, STARTING)) {
 404                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 405                 goto out_close_maincmd_fd;
 406         }
 407
 408         /* Start of environment variable setup for hooks */
 409         if (setenv("LXC_NAME", name, 1)) {
 410                 SYSERROR("failed to set environment variable for container name");
 411         }
 412         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 413                 SYSERROR("failed to set environment variable for config path");
 414         }
 415         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 416                 SYSERROR("failed to set environment variable for rootfs mount");
 417         }
 418         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 419                 SYSERROR("failed to set environment variable for rootfs mount");
 420         }
 421         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 422                 SYSERROR("failed to set environment variable for console path");
 423         }
 424         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 425                 SYSERROR("failed to set environment variable for console log");
 426         }
 427         /* End of environment variable setup for hooks */
 428
 429         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 430                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 431                 goto out_aborting;
 432         }
 433
 434         /* the signal fd has to be created before forking otherwise
 435          * if the child process exits before we setup the signal fd,
 436          * the event will be lost and the command will be stuck */
 437         handler->sigfd = setup_signal_fd(&handler->oldmask);
 438         if (handler->sigfd < 0) {
 439                 ERROR("failed to set sigchild fd handler");
 440                 goto out_delete_tty;
 441         }
 442
 443         /* do this after setting up signals since it might unblock SIGWINCH */
 444         if (lxc_console_create(conf)) {
 445                 ERROR("failed to create console");
 446                 goto out_restore_sigmask;
 447         }
 448
 449         if (ttys_shift_ids(conf) < 0) {
 450                 ERROR("Failed to shift tty into container");
 451                 goto out_restore_sigmask;
 452         }
 453
 454         INFO("'%s' is initialized", name);
 455         return handler;
 456
 457 out_restore_sigmask:
 458         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 459 out_delete_tty:
 460         lxc_delete_tty(&conf->tty_info);
 461 out_aborting:
 462         lxc_set_state(name, handler, ABORTING);
 463 out_close_maincmd_fd:
 464         close(conf->maincmd_fd);
 465         conf->maincmd_fd = -1;
 466 out_free_name:
 467         free(handler->name);
 468         handler->name = NULL;
 469 out_free:
 470         free(handler);
 471         return NULL;
 472 }
 473
 474 void lxc_fini(const char *name, struct lxc_handler *handler)
 475 {
 476         /* The STOPPING state is there for future cleanup code
 477          * which can take awhile
 478          */
 479         lxc_set_state(name, handler, STOPPING);
 480         lxc_set_state(name, handler, STOPPED);
 481
 482         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 483                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 484
 485         /* reset mask set by setup_signal_fd */
 486         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 487                 WARN("failed to restore sigprocmask");
 488
 489         lxc_console_delete(&handler->conf->console);
 490         lxc_delete_tty(&handler->conf->tty_info);
 491         close(handler->conf->maincmd_fd);
 492         handler->conf->maincmd_fd = -1;
 493         free(handler->name);
 494         if (handler->ttysock[0] != -1) {
 495                 close(handler->ttysock[0]);
 496                 close(handler->ttysock[1]);
 497         }
 498         cgroup_destroy(handler);
 499         free(handler);
 500 }
 501
 502 void lxc_abort(const char *name, struct lxc_handler *handler)
 503 {
 504         int ret, status;
 505
 506         lxc_set_state(name, handler, ABORTING);
 507         if (handler->pid > 0)
 508                 kill(handler->pid, SIGKILL);
 509         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 510 }
 511
 512 #include <sys/reboot.h>
 513 #include <linux/reboot.h>
 514
 515 /*
 516  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 517  * in a child pid namespace if container reboot support exists.
 518  * Otherwise, it will either succeed or return -EPERM.
 519  */
 520 static int container_reboot_supported(void *arg)
 521 {
 522         int *cmd = arg;
 523         int ret;
 524
 525         ret = reboot(*cmd);
 526         if (ret == -1 && errno == EINVAL)
 527                 return 1;
 528         return 0;
 529 }
 530
 531 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 532 {
 533         FILE *f;
 534         int ret, cmd, v, flags;
 535         long stack_size = 4096;
 536         void *stack = alloca(stack_size);
 537         int status;
 538         pid_t pid;
 539
 540         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 541         if (!f) {
 542                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 543                 return 1;
 544         }
 545
 546         ret = fscanf(f, "%d", &v);
 547         fclose(f);
 548         if (ret != 1) {
 549                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 550                 return 1;
 551         }
 552         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 553
 554         flags = CLONE_NEWPID | SIGCHLD;
 555         if (!lxc_list_empty(&conf->id_map))
 556                 flags |= CLONE_NEWUSER;
 557
 558 #ifdef __ia64__
 559         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 560 #else
 561         stack += stack_size;
 562         pid = clone(container_reboot_supported, stack, flags, &cmd);
 563 #endif
 564         if (pid < 0) {
 565                 if (flags & CLONE_NEWUSER)
 566                         ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno));
 567                 else
 568                         ERROR("failed to clone (%#x): %s", flags, strerror(errno));
 569                 return -1;
 570         }
 571         if (wait(&status) < 0) {
 572                 SYSERROR("unexpected wait error: %m");
 573                 return -1;
 574         }
 575
 576         if (WEXITSTATUS(status) != 1)
 577                 return 1;
 578
 579         return 0;
 580 }
 581
 582 /*
 583  * netpipe is used in the unprivileged case to transfer the ifindexes
 584  * from parent to child
 585  */
 586 static int netpipe = -1;
 587
 588 static inline int count_veths(struct lxc_list *network)
 589 {
 590         struct lxc_list *iterator;
 591         struct lxc_netdev *netdev;
 592         int count = 0;
 593
 594         lxc_list_for_each(iterator, network) {
 595                 netdev = iterator->elem;
 596                 if (netdev->type != LXC_NET_VETH)
 597                         continue;
 598                 count++;
 599         }
 600         return count;
 601 }
 602
 603 static int read_unpriv_netifindex(struct lxc_list *network)
 604 {
 605         struct lxc_list *iterator;
 606         struct lxc_netdev *netdev;
 607
 608         if (netpipe == -1)
 609                 return 0;
 610         lxc_list_for_each(iterator, network) {
 611                 netdev = iterator->elem;
 612                 if (netdev->type != LXC_NET_VETH)
 613                         continue;
 614                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 615                         ERROR("Out of memory");
 616                         close(netpipe);
 617                         return -1;
 618                 }
 619                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 620                         close(netpipe);
 621                         return -1;
 622                 }
 623         }
 624         close(netpipe);
 625         return 0;
 626 }
 627
 628 static int do_start(void *data)
 629 {
 630         struct lxc_list *iterator;
 631         struct lxc_handler *handler = data;
 632
 633         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 634                 SYSERROR("failed to set sigprocmask");
 635                 return -1;
 636         }
 637
 638         /* This prctl must be before the synchro, so if the parent
 639          * dies before we set the parent death signal, we will detect
 640          * its death with the synchro right after, otherwise we have
 641          * a window where the parent can exit before we set the pdeath
 642          * signal leading to a unsupervized container.
 643          */
 644         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 645                 SYSERROR("failed to set pdeath signal");
 646                 return -1;
 647         }
 648
 649         lxc_sync_fini_parent(handler);
 650
 651         /* don't leak the pinfd to the container */
 652         if (handler->pinfd >= 0) {
 653                 close(handler->pinfd);
 654         }
 655
 656         /* Tell the parent task it can begin to configure the
 657          * container and wait for it to finish
 658          */
 659         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 660                 return -1;
 661
 662         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 663                 goto out_warn_father;
 664
 665         /*
 666          * if we are in a new user namespace, become root there to have
 667          * privilege over our namespace
 668          */
 669         if (!lxc_list_empty(&handler->conf->id_map)) {
 670                 NOTICE("switching to gid/uid 0 in new user namespace");
 671                 if (setgid(0)) {
 672                         SYSERROR("setgid");
 673                         goto out_warn_father;
 674                 }
 675                 if (setuid(0)) {
 676                         SYSERROR("setuid");
 677                         goto out_warn_father;
 678                 }
 679                 if (setgroups(0, NULL)) {
 680                         SYSERROR("setgroups");
 681                         goto out_warn_father;
 682                 }
 683         }
 684
 685         if (access(handler->lxcpath, X_OK)) {
 686                 print_top_failing_dir(handler->lxcpath);
 687                 goto out_warn_father;
 688         }
 689
 690         #if HAVE_SYS_CAPABILITY_H
 691         if (handler->conf->need_utmp_watch) {
 692                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 693                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 694                         goto out_warn_father;
 695                 }
 696                 DEBUG("Dropped cap_sys_boot");
 697         }
 698         #endif
 699
 700         /* Setup the container, ip, names, utsname, ... */
 701         if (lxc_setup(handler)) {
 702                 ERROR("failed to setup the container");
 703                 goto out_warn_father;
 704         }
 705
 706         /* ask father to setup cgroups and wait for him to finish */
 707         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 708                 return -1;
 709
 710         /* Set the label to change to when we exec(2) the container's init */
 711         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 712                 goto out_warn_father;
 713
 714         /* Some init's such as busybox will set sane tty settings on stdin,
 715          * stdout, stderr which it thinks is the console. We already set them
 716          * the way we wanted on the real terminal, and we want init to do its
 717          * setup on its console ie. the pty allocated in lxc_console_create()
 718          * so make sure that that pty is stdin,stdout,stderr.
 719          */
 720         if (lxc_console_set_stdfds(handler) < 0)
 721                 goto out_warn_father;
 722
 723         /* If we mounted a temporary proc, then unmount it now */
 724         tmp_proc_unmount(handler->conf);
 725
 726         if (lxc_seccomp_load(handler->conf) != 0)
 727                 goto out_warn_father;
 728
 729         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 730                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 731                 goto out_warn_father;
 732         }
 733
 734         /* The clearenv() and putenv() calls have been moved here
 735          * to allow us to use environment variables passed to the various
 736          * hooks, such as the start hook above.  Not all of the
 737          * variables like CONFIG_PATH or ROOTFS are valid in this
 738          * context but others are. */
 739         if (clearenv()) {
 740                 SYSERROR("failed to clear environment");
 741                 /* don't error out though */
 742         }
 743
 744         lxc_list_for_each(iterator, &handler->conf->environment) {
 745                 if (putenv((char *)iterator->elem)) {
 746                         SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem);
 747                         goto out_warn_father;
 748                 }
 749         }
 750
 751         if (putenv("container=lxc")) {
 752                 SYSERROR("failed to set environment variable 'container=lxc'");
 753                 goto out_warn_father;
 754         }
 755
 756         if (handler->conf->pty_names) {
 757                 if (putenv(handler->conf->pty_names)) {
 758                         SYSERROR("failed to set environment variable for container ptys");
 759                         goto out_warn_father;
 760                 }
 761         }
 762
 763         close(handler->sigfd);
 764
 765         if (handler->backgrounded && null_stdfds() < 0)
 766                 goto out_warn_father;
 767
 768         /* after this call, we are in error because this
 769          * ops should not return as it execs */
 770         handler->ops->start(handler, handler->data);
 771
 772 out_warn_father:
 773         /* we want the parent to know something went wrong, so any
 774          * value other than what it expects is ok. */
 775         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 776         return -1;
 777 }
 778
 779 static int save_phys_nics(struct lxc_conf *conf)
 780 {
 781         struct lxc_list *iterator;
 782
 783         lxc_list_for_each(iterator, &conf->network) {
 784                 struct lxc_netdev *netdev = iterator->elem;
 785
 786                 if (netdev->type != LXC_NET_PHYS)
 787                         continue;
 788                 conf->saved_nics = realloc(conf->saved_nics,
 789                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 790                 if (!conf->saved_nics) {
 791                         SYSERROR("failed to allocate memory");
 792                         return -1;
 793                 }
 794                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 795                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 796                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 797                         SYSERROR("failed to allocate memory");
 798                         return -1;
 799                 }
 800                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 801                         conf->saved_nics[conf->num_savednics].ifindex,
 802                         conf->saved_nics[conf->num_savednics].orig_name);
 803                 conf->num_savednics++;
 804         }
 805
 806         return 0;
 807 }
 808
 809 static int recv_fd(int sock, int *fd)
 810 {
 811         if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) {
 812                 SYSERROR("Error receiving tty fd from child");
 813                 return -1;
 814         }
 815         if (*fd == -1)
 816                 return -1;
 817         return 0;
 818 }
 819
 820 static int recv_ttys_from_child(struct lxc_handler *handler)
 821 {
 822         struct lxc_conf *conf = handler->conf;
 823         int i, sock = handler->ttysock[1];
 824         struct lxc_tty_info *tty_info = &conf->tty_info;
 825
 826         if (!conf->tty)
 827                 return 0;
 828
 829         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty);
 830         if (!tty_info->pty_info) {
 831                 SYSERROR("failed to allocate pty_info");
 832                 return -1;
 833         }
 834
 835         for (i = 0; i < conf->tty; i++) {
 836                 struct lxc_pty_info *pty_info = &tty_info->pty_info[i];
 837                 pty_info->busy = 0;
 838                 if (recv_fd(sock, &pty_info->slave) < 0 ||
 839                                 recv_fd(sock, &pty_info->master) < 0) {
 840                         ERROR("Error receiving tty info from child");
 841                         return -1;
 842                 }
 843         }
 844         tty_info->nbtty = conf->tty;
 845
 846         return 0;
 847 }
 848
 849 void resolve_clone_flags(struct lxc_handler *handler)
 850 {
 851         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
 852
 853         if (!lxc_list_empty(&handler->conf->id_map)) {
 854                 INFO("Cloning a new user namespace");
 855                 handler->clone_flags |= CLONE_NEWUSER;
 856         }
 857
 858         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 859                 if (!lxc_requests_empty_network(handler))
 860                         handler->clone_flags |= CLONE_NEWNET;
 861         } else {
 862                 INFO("Inheriting a net namespace");
 863         }
 864
 865         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 866                 handler->clone_flags |= CLONE_NEWIPC;
 867         } else {
 868                 INFO("Inheriting an IPC namespace");
 869         }
 870
 871         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 872                 handler->clone_flags |= CLONE_NEWUTS;
 873         } else {
 874                 INFO("Inheriting a UTS namespace");
 875         }
 876 }
 877
 878 static int lxc_spawn(struct lxc_handler *handler)
 879 {
 880         int failed_before_rename = 0;
 881         const char *name = handler->name;
 882         bool cgroups_connected = false;
 883         int saved_ns_fd[LXC_NS_MAX];
 884         int preserve_mask = 0, i;
 885         int netpipepair[2], nveths;
 886
 887         netpipe = -1;
 888
 889         for (i = 0; i < LXC_NS_MAX; i++)
 890                 if (handler->conf->inherit_ns_fd[i] != -1)
 891                         preserve_mask |= ns_info[i].clone_flag;
 892
 893         if (lxc_sync_init(handler))
 894                 return -1;
 895
 896         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
 897                 lxc_sync_fini(handler);
 898                 return -1;
 899         }
 900
 901         resolve_clone_flags(handler);
 902
 903         if (handler->clone_flags & CLONE_NEWNET) {
 904                 if (!lxc_list_empty(&handler->conf->network)) {
 905
 906                         /* Find gateway addresses from the link device, which is
 907                          * no longer accessible inside the container. Do this
 908                          * before creating network interfaces, since goto
 909                          * out_delete_net does not work before lxc_clone. */
 910                         if (lxc_find_gateway_addresses(handler)) {
 911                                 ERROR("failed to find gateway addresses");
 912                                 lxc_sync_fini(handler);
 913                                 return -1;
 914                         }
 915
 916                         /* that should be done before the clone because we will
 917                          * fill the netdev index and use them in the child
 918                          */
 919                         if (lxc_create_network(handler)) {
 920                                 ERROR("failed to create the network");
 921                                 lxc_sync_fini(handler);
 922                                 return -1;
 923                         }
 924                 }
 925
 926                 if (save_phys_nics(handler->conf)) {
 927                         ERROR("failed to save physical nic info");
 928                         goto out_abort;
 929                 }
 930         }
 931
 932         if (!cgroup_init(handler)) {
 933                 ERROR("failed initializing cgroup support");
 934                 goto out_delete_net;
 935         }
 936
 937         cgroups_connected = true;
 938
 939         if (!cgroup_create(handler)) {
 940                 ERROR("failed creating cgroups");
 941                 goto out_delete_net;
 942         }
 943
 944         /*
 945          * if the rootfs is not a blockdev, prevent the container from
 946          * marking it readonly.
 947          *
 948          * if the container is unprivileged then skip rootfs pinning
 949          */
 950         if (lxc_list_empty(&handler->conf->id_map)) {
 951                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 952                 if (handler->pinfd == -1)
 953                         INFO("failed to pin the container's rootfs");
 954         }
 955
 956         if (preserve_ns(saved_ns_fd, preserve_mask) < 0)
 957                 goto out_delete_net;
 958         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
 959                 goto out_delete_net;
 960
 961         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
 962                 if (pipe(netpipepair) < 0) {
 963                         SYSERROR("Error creating pipe");
 964                         goto out_delete_net;
 965                 }
 966                 /* store netpipe in the global var for do_start's use */
 967                 netpipe = netpipepair[0];
 968         }
 969
 970         /* Create a process in a new set of namespaces */
 971         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 972         if (handler->pid < 0) {
 973                 SYSERROR("failed to fork into a new namespace");
 974                 goto out_delete_net;
 975         }
 976
 977         if (attach_ns(saved_ns_fd))
 978                 WARN("failed to restore saved namespaces");
 979
 980         lxc_sync_fini_child(handler);
 981
 982         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 983                 failed_before_rename = 1;
 984
 985         if (!cgroup_create_legacy(handler)) {
 986                 ERROR("failed to setup the legacy cgroups for %s", name);
 987                 goto out_delete_net;
 988         }
 989         if (!cgroup_setup_limits(handler, false)) {
 990                 ERROR("failed to setup the cgroup limits for '%s'", name);
 991                 goto out_delete_net;
 992         }
 993
 994         if (!cgroup_enter(handler))
 995                 goto out_delete_net;
 996
 997         if (!cgroup_chown(handler))
 998                 goto out_delete_net;
 999
1000         if (failed_before_rename)
1001                 goto out_delete_net;
1002
1003         /* Create the network configuration */
1004         if (handler->clone_flags & CLONE_NEWNET) {
1005                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
1006                         ERROR("failed to create the configured network");
1007                         goto out_delete_net;
1008                 }
1009         }
1010
1011         if (netpipe != -1) {
1012                 struct lxc_list *iterator;
1013                 struct lxc_netdev *netdev;
1014
1015                 close(netpipe);
1016                 lxc_list_for_each(iterator, &handler->conf->network) {
1017                         netdev = iterator->elem;
1018                         if (netdev->type != LXC_NET_VETH)
1019                                 continue;
1020                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1021                                 ERROR("Error writing veth name to container");
1022                                 goto out_delete_net;
1023                         }
1024                 }
1025                 close(netpipepair[1]);
1026         }
1027
1028         /* map the container uids - the container became an invalid
1029          * userid the moment it was cloned with CLONE_NEWUSER - this
1030          * call doesn't change anything immediately, but allows the
1031          * container to setuid(0) (0 being mapped to something else on
1032          * the host) later to become a valid uid again */
1033         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
1034                 ERROR("failed to set up id mapping");
1035                 goto out_delete_net;
1036         }
1037
1038         /* Tell the child to continue its initialization.  we'll get
1039          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1040          */
1041         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1042                 goto out_delete_net;
1043
1044         if (!cgroup_setup_limits(handler, true)) {
1045                 ERROR("failed to setup the devices cgroup for '%s'", name);
1046                 goto out_delete_net;
1047         }
1048
1049         cgroup_disconnect();
1050         cgroups_connected = false;
1051
1052         /* read tty fds allocated by child */
1053         if (recv_ttys_from_child(handler) < 0) {
1054                 ERROR("failed to receive tty info from child");
1055                 goto out_delete_net;
1056         }
1057
1058         /* Tell the child to complete its initialization and wait for
1059          * it to exec or return an error.  (the child will never
1060          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
1061          * sync pipe, causing lxc_sync_barrier_child to return
1062          * success, or return a different value, causing us to error
1063          * out).
1064          */
1065         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1066                 return -1;
1067
1068         if (detect_shared_rootfs())
1069                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
1070
1071         if (handler->ops->post_start(handler, handler->data))
1072                 goto out_abort;
1073
1074         if (lxc_set_state(name, handler, RUNNING)) {
1075                 ERROR("failed to set state to %s",
1076                               lxc_state2str(RUNNING));
1077                 goto out_abort;
1078         }
1079
1080         lxc_sync_fini(handler);
1081
1082         return 0;
1083
1084 out_delete_net:
1085         if (cgroups_connected)
1086                 cgroup_disconnect();
1087         if (handler->clone_flags & CLONE_NEWNET)
1088                 lxc_delete_network(handler);
1089 out_abort:
1090         lxc_abort(name, handler);
1091         lxc_sync_fini(handler);
1092         if (handler->pinfd >= 0) {
1093                 close(handler->pinfd);
1094                 handler->pinfd = -1;
1095         }
1096
1097         return -1;
1098 }
1099
1100 int get_netns_fd(int pid)
1101 {
1102         char path[MAXPATHLEN];
1103         int ret, fd;
1104
1105         ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid);
1106         if (ret < 0 || ret >= MAXPATHLEN) {
1107                 WARN("Failed to pin netns file for pid %d", pid);
1108                 return -1;
1109         }
1110
1111         fd = open(path, O_RDONLY);
1112         if (fd < 0) {
1113                 WARN("Failed to pin netns file %s for pid %d: %s",
1114                                 path, pid, strerror(errno));
1115                 return -1;
1116         }
1117         return fd;
1118 }
1119
1120 int __lxc_start(const char *name, struct lxc_conf *conf,
1121                 struct lxc_operations* ops, void *data, const char *lxcpath,
1122                 bool backgrounded)
1123 {
1124         struct lxc_handler *handler;
1125         int err = -1;
1126         int status;
1127         int netnsfd = -1;
1128
1129         handler = lxc_init(name, conf, lxcpath);
1130         if (!handler) {
1131                 ERROR("failed to initialize the container");
1132                 return -1;
1133         }
1134         handler->ops = ops;
1135         handler->data = data;
1136         handler->backgrounded = backgrounded;
1137
1138         if (must_drop_cap_sys_boot(handler->conf)) {
1139                 #if HAVE_SYS_CAPABILITY_H
1140                 DEBUG("Dropping cap_sys_boot");
1141                 #else
1142                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1143                 #endif
1144         } else {
1145                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1146                 handler->conf->need_utmp_watch = 0;
1147         }
1148
1149         if (!attach_block_device(handler->conf)) {
1150                 ERROR("Failure attaching block device");
1151                 goto out_fini_nonet;
1152         }
1153
1154         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1155                 /* if the backing store is a device, mount it here and now */
1156                 if (rootfs_is_blockdev(conf)) {
1157                         if (unshare(CLONE_NEWNS) < 0) {
1158                                 ERROR("Error unsharing mounts");
1159                                 goto out_fini_nonet;
1160                         }
1161                         remount_all_slave();
1162                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1163                                 ERROR("Error setting up rootfs mount as root before spawn");
1164                                 goto out_fini_nonet;
1165                         }
1166                         INFO("Set up container rootfs as host root");
1167                 }
1168         }
1169
1170         err = lxc_spawn(handler);
1171         if (err) {
1172                 ERROR("failed to spawn '%s'", name);
1173                 goto out_detach_blockdev;
1174         }
1175
1176         handler->conf->reboot = 0;
1177
1178         netnsfd = get_netns_fd(handler->pid);
1179
1180         err = lxc_poll(name, handler);
1181         if (err) {
1182                 ERROR("mainloop exited with an error");
1183                 if (netnsfd >= 0)
1184                         close(netnsfd);
1185                 goto out_abort;
1186         }
1187
1188         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1189                 continue;
1190
1191         /*
1192          * If the child process exited but was not signaled,
1193          * it didn't call reboot.  This should mean it was an
1194          * lxc-execute which simply exited.  In any case, treat
1195          * it as a 'halt'
1196          */
1197         if (WIFSIGNALED(status)) {
1198                 switch(WTERMSIG(status)) {
1199                 case SIGINT: /* halt */
1200                         DEBUG("Container halting");
1201                         break;
1202                 case SIGHUP: /* reboot */
1203                         DEBUG("Container rebooting");
1204                         handler->conf->reboot = 1;
1205                         break;
1206                 case SIGSYS: /* seccomp */
1207                         DEBUG("Container violated its seccomp policy");
1208                         break;
1209                 default:
1210                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1211                         break;
1212                 }
1213         }
1214
1215         lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf);
1216         if (netnsfd >= 0)
1217                 close(netnsfd);
1218
1219         if (handler->pinfd >= 0) {
1220                 close(handler->pinfd);
1221                 handler->pinfd = -1;
1222         }
1223
1224         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1225         err =  lxc_error_set_and_log(handler->pid, status);
1226 out_fini:
1227         lxc_delete_network(handler);
1228
1229 out_detach_blockdev:
1230         detach_block_device(handler->conf);
1231
1232 out_fini_nonet:
1233         lxc_fini(name, handler);
1234         return err;
1235
1236 out_abort:
1237         lxc_abort(name, handler);
1238         goto out_fini;
1239 }
1240
1241 struct start_args {
1242         char *const *argv;
1243 };
1244
1245 static int start(struct lxc_handler *handler, void* data)
1246 {
1247         struct start_args *arg = data;
1248
1249         NOTICE("exec'ing '%s'", arg->argv[0]);
1250
1251         execvp(arg->argv[0], arg->argv);
1252         SYSERROR("failed to exec %s", arg->argv[0]);
1253         return 0;
1254 }
1255
1256 static int post_start(struct lxc_handler *handler, void* data)
1257 {
1258         struct start_args *arg = data;
1259
1260         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1261         return 0;
1262 }
1263
1264 static struct lxc_operations start_ops = {
1265         .start = start,
1266         .post_start = post_start
1267 };
1268
1269 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1270               const char *lxcpath, bool backgrounded)
1271 {
1272         struct start_args start_arg = {
1273                 .argv = argv,
1274         };
1275
1276         conf->need_utmp_watch = 1;
1277         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded);
1278 }