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