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 (name && setenv("LXC_NAME", name, 1)) {
 410                 SYSERROR("failed to set environment variable for container name");
 411         }
 412         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 413                 SYSERROR("failed to set environment variable for config path");
 414         }
 415         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 416                 SYSERROR("failed to set environment variable for rootfs mount");
 417         }
 418         if (conf->rootfs.path && 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. When using lxc-execute we default to root,
 668          * but this can be overriden using the lxc.init_uid and lxc.init_gid
 669          * configuration options.
 670          */
 671         if (!lxc_list_empty(&handler->conf->id_map)) {
 672                 gid_t new_gid = 0;
 673                 if (handler->conf->is_execute && handler->conf->init_gid)
 674                         new_gid = handler->conf->init_gid;
 675
 676                 uid_t new_uid = 0;
 677                 if (handler->conf->is_execute && handler->conf->init_uid)
 678                         new_uid = handler->conf->init_uid;
 679
 680                 NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid);
 681                 if (setgid(new_gid)) {
 682                         SYSERROR("setgid");
 683                         goto out_warn_father;
 684                 }
 685                 if (setuid(new_uid)) {
 686                         SYSERROR("setuid");
 687                         goto out_warn_father;
 688                 }
 689                 if (setgroups(0, NULL)) {
 690                         SYSERROR("setgroups");
 691                         goto out_warn_father;
 692                 }
 693         }
 694
 695         if (access(handler->lxcpath, X_OK)) {
 696                 print_top_failing_dir(handler->lxcpath);
 697                 goto out_warn_father;
 698         }
 699
 700         #if HAVE_SYS_CAPABILITY_H
 701         if (handler->conf->need_utmp_watch) {
 702                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 703                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 704                         goto out_warn_father;
 705                 }
 706                 DEBUG("Dropped cap_sys_boot");
 707         }
 708         #endif
 709
 710         /* Setup the container, ip, names, utsname, ... */
 711         if (lxc_setup(handler)) {
 712                 ERROR("failed to setup the container");
 713                 goto out_warn_father;
 714         }
 715
 716         /* ask father to setup cgroups and wait for him to finish */
 717         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 718                 return -1;
 719
 720         /* Set the label to change to when we exec(2) the container's init */
 721         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 722                 goto out_warn_father;
 723
 724         /* Some init's such as busybox will set sane tty settings on stdin,
 725          * stdout, stderr which it thinks is the console. We already set them
 726          * the way we wanted on the real terminal, and we want init to do its
 727          * setup on its console ie. the pty allocated in lxc_console_create()
 728          * so make sure that that pty is stdin,stdout,stderr.
 729          */
 730         if (lxc_console_set_stdfds(handler) < 0)
 731                 goto out_warn_father;
 732
 733         /* If we mounted a temporary proc, then unmount it now */
 734         tmp_proc_unmount(handler->conf);
 735
 736         if (lxc_seccomp_load(handler->conf) != 0)
 737                 goto out_warn_father;
 738
 739         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 740                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 741                 goto out_warn_father;
 742         }
 743
 744         /* The clearenv() and putenv() calls have been moved here
 745          * to allow us to use environment variables passed to the various
 746          * hooks, such as the start hook above.  Not all of the
 747          * variables like CONFIG_PATH or ROOTFS are valid in this
 748          * context but others are. */
 749         if (clearenv()) {
 750                 SYSERROR("failed to clear environment");
 751                 /* don't error out though */
 752         }
 753
 754         lxc_list_for_each(iterator, &handler->conf->environment) {
 755                 if (putenv((char *)iterator->elem)) {
 756                         SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem);
 757                         goto out_warn_father;
 758                 }
 759         }
 760
 761         if (putenv("container=lxc")) {
 762                 SYSERROR("failed to set environment variable 'container=lxc'");
 763                 goto out_warn_father;
 764         }
 765
 766         if (handler->conf->pty_names) {
 767                 if (putenv(handler->conf->pty_names)) {
 768                         SYSERROR("failed to set environment variable for container ptys");
 769                         goto out_warn_father;
 770                 }
 771         }
 772
 773         close(handler->sigfd);
 774
 775         if (handler->backgrounded && null_stdfds() < 0)
 776                 goto out_warn_father;
 777
 778         /* after this call, we are in error because this
 779          * ops should not return as it execs */
 780         handler->ops->start(handler, handler->data);
 781
 782 out_warn_father:
 783         /* we want the parent to know something went wrong, so any
 784          * value other than what it expects is ok. */
 785         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 786         return -1;
 787 }
 788
 789 static int save_phys_nics(struct lxc_conf *conf)
 790 {
 791         struct lxc_list *iterator;
 792
 793         lxc_list_for_each(iterator, &conf->network) {
 794                 struct lxc_netdev *netdev = iterator->elem;
 795
 796                 if (netdev->type != LXC_NET_PHYS)
 797                         continue;
 798                 conf->saved_nics = realloc(conf->saved_nics,
 799                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 800                 if (!conf->saved_nics) {
 801                         SYSERROR("failed to allocate memory");
 802                         return -1;
 803                 }
 804                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 805                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 806                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 807                         SYSERROR("failed to allocate memory");
 808                         return -1;
 809                 }
 810                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 811                         conf->saved_nics[conf->num_savednics].ifindex,
 812                         conf->saved_nics[conf->num_savednics].orig_name);
 813                 conf->num_savednics++;
 814         }
 815
 816         return 0;
 817 }
 818
 819 static int recv_fd(int sock, int *fd)
 820 {
 821         if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) {
 822                 SYSERROR("Error receiving tty fd from child");
 823                 return -1;
 824         }
 825         if (*fd == -1)
 826                 return -1;
 827         return 0;
 828 }
 829
 830 static int recv_ttys_from_child(struct lxc_handler *handler)
 831 {
 832         struct lxc_conf *conf = handler->conf;
 833         int i, sock = handler->ttysock[1];
 834         struct lxc_tty_info *tty_info = &conf->tty_info;
 835
 836         if (!conf->tty)
 837                 return 0;
 838
 839         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty);
 840         if (!tty_info->pty_info) {
 841                 SYSERROR("failed to allocate pty_info");
 842                 return -1;
 843         }
 844
 845         for (i = 0; i < conf->tty; i++) {
 846                 struct lxc_pty_info *pty_info = &tty_info->pty_info[i];
 847                 pty_info->busy = 0;
 848                 if (recv_fd(sock, &pty_info->slave) < 0 ||
 849                                 recv_fd(sock, &pty_info->master) < 0) {
 850                         ERROR("Error receiving tty info from child");
 851                         return -1;
 852                 }
 853         }
 854         tty_info->nbtty = conf->tty;
 855
 856         return 0;
 857 }
 858
 859 void resolve_clone_flags(struct lxc_handler *handler)
 860 {
 861         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
 862
 863         if (!lxc_list_empty(&handler->conf->id_map)) {
 864                 INFO("Cloning a new user namespace");
 865                 handler->clone_flags |= CLONE_NEWUSER;
 866         }
 867
 868         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 869                 if (!lxc_requests_empty_network(handler))
 870                         handler->clone_flags |= CLONE_NEWNET;
 871         } else {
 872                 INFO("Inheriting a net namespace");
 873         }
 874
 875         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 876                 handler->clone_flags |= CLONE_NEWIPC;
 877         } else {
 878                 INFO("Inheriting an IPC namespace");
 879         }
 880
 881         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 882                 handler->clone_flags |= CLONE_NEWUTS;
 883         } else {
 884                 INFO("Inheriting a UTS namespace");
 885         }
 886 }
 887
 888 static int lxc_spawn(struct lxc_handler *handler)
 889 {
 890         int failed_before_rename = 0;
 891         const char *name = handler->name;
 892         bool cgroups_connected = false;
 893         int saved_ns_fd[LXC_NS_MAX];
 894         int preserve_mask = 0, i;
 895         int netpipepair[2], nveths;
 896
 897         netpipe = -1;
 898
 899         for (i = 0; i < LXC_NS_MAX; i++)
 900                 if (handler->conf->inherit_ns_fd[i] != -1)
 901                         preserve_mask |= ns_info[i].clone_flag;
 902
 903         if (lxc_sync_init(handler))
 904                 return -1;
 905
 906         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
 907                 lxc_sync_fini(handler);
 908                 return -1;
 909         }
 910
 911         resolve_clone_flags(handler);
 912
 913         if (handler->clone_flags & CLONE_NEWNET) {
 914                 if (!lxc_list_empty(&handler->conf->network)) {
 915
 916                         /* Find gateway addresses from the link device, which is
 917                          * no longer accessible inside the container. Do this
 918                          * before creating network interfaces, since goto
 919                          * out_delete_net does not work before lxc_clone. */
 920                         if (lxc_find_gateway_addresses(handler)) {
 921                                 ERROR("failed to find gateway addresses");
 922                                 lxc_sync_fini(handler);
 923                                 return -1;
 924                         }
 925
 926                         /* that should be done before the clone because we will
 927                          * fill the netdev index and use them in the child
 928                          */
 929                         if (lxc_create_network(handler)) {
 930                                 ERROR("failed to create the network");
 931                                 lxc_sync_fini(handler);
 932                                 return -1;
 933                         }
 934                 }
 935
 936                 if (save_phys_nics(handler->conf)) {
 937                         ERROR("failed to save physical nic info");
 938                         goto out_abort;
 939                 }
 940         }
 941
 942         if (!cgroup_init(handler)) {
 943                 ERROR("failed initializing cgroup support");
 944                 goto out_delete_net;
 945         }
 946
 947         cgroups_connected = true;
 948
 949         if (!cgroup_create(handler)) {
 950                 ERROR("failed creating cgroups");
 951                 goto out_delete_net;
 952         }
 953
 954         /*
 955          * if the rootfs is not a blockdev, prevent the container from
 956          * marking it readonly.
 957          *
 958          * if the container is unprivileged then skip rootfs pinning
 959          */
 960         if (lxc_list_empty(&handler->conf->id_map)) {
 961                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 962                 if (handler->pinfd == -1)
 963                         INFO("failed to pin the container's rootfs");
 964         }
 965
 966         if (preserve_ns(saved_ns_fd, preserve_mask) < 0)
 967                 goto out_delete_net;
 968         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
 969                 goto out_delete_net;
 970
 971         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
 972                 if (pipe(netpipepair) < 0) {
 973                         SYSERROR("Error creating pipe");
 974                         goto out_delete_net;
 975                 }
 976                 /* store netpipe in the global var for do_start's use */
 977                 netpipe = netpipepair[0];
 978         }
 979
 980         /* Create a process in a new set of namespaces */
 981         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 982         if (handler->pid < 0) {
 983                 SYSERROR("failed to fork into a new namespace");
 984                 goto out_delete_net;
 985         }
 986
 987         if (attach_ns(saved_ns_fd))
 988                 WARN("failed to restore saved namespaces");
 989
 990         lxc_sync_fini_child(handler);
 991
 992         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 993                 failed_before_rename = 1;
 994
 995         if (!cgroup_create_legacy(handler)) {
 996                 ERROR("failed to setup the legacy cgroups for %s", name);
 997                 goto out_delete_net;
 998         }
 999         if (!cgroup_setup_limits(handler, false)) {
1000                 ERROR("failed to setup the cgroup limits for '%s'", name);
1001                 goto out_delete_net;
1002         }
1003
1004         if (!cgroup_enter(handler))
1005                 goto out_delete_net;
1006
1007         if (!cgroup_chown(handler))
1008                 goto out_delete_net;
1009
1010         if (failed_before_rename)
1011                 goto out_delete_net;
1012
1013         /* Create the network configuration */
1014         if (handler->clone_flags & CLONE_NEWNET) {
1015                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
1016                         ERROR("failed to create the configured network");
1017                         goto out_delete_net;
1018                 }
1019         }
1020
1021         if (netpipe != -1) {
1022                 struct lxc_list *iterator;
1023                 struct lxc_netdev *netdev;
1024
1025                 close(netpipe);
1026                 lxc_list_for_each(iterator, &handler->conf->network) {
1027                         netdev = iterator->elem;
1028                         if (netdev->type != LXC_NET_VETH)
1029                                 continue;
1030                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1031                                 ERROR("Error writing veth name to container");
1032                                 goto out_delete_net;
1033                         }
1034                 }
1035                 close(netpipepair[1]);
1036         }
1037
1038         /* map the container uids - the container became an invalid
1039          * userid the moment it was cloned with CLONE_NEWUSER - this
1040          * call doesn't change anything immediately, but allows the
1041          * container to setuid(0) (0 being mapped to something else on
1042          * the host) later to become a valid uid again */
1043         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
1044                 ERROR("failed to set up id mapping");
1045                 goto out_delete_net;
1046         }
1047
1048         /* Tell the child to continue its initialization.  we'll get
1049          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1050          */
1051         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1052                 goto out_delete_net;
1053
1054         if (!cgroup_setup_limits(handler, true)) {
1055                 ERROR("failed to setup the devices cgroup for '%s'", name);
1056                 goto out_delete_net;
1057         }
1058
1059         cgroup_disconnect();
1060         cgroups_connected = false;
1061
1062         /* read tty fds allocated by child */
1063         if (recv_ttys_from_child(handler) < 0) {
1064                 ERROR("failed to receive tty info from child");
1065                 goto out_delete_net;
1066         }
1067
1068         /* Tell the child to complete its initialization and wait for
1069          * it to exec or return an error.  (the child will never
1070          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
1071          * sync pipe, causing lxc_sync_barrier_child to return
1072          * success, or return a different value, causing us to error
1073          * out).
1074          */
1075         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1076                 return -1;
1077
1078         if (detect_shared_rootfs())
1079                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
1080
1081         if (handler->ops->post_start(handler, handler->data))
1082                 goto out_abort;
1083
1084         if (lxc_set_state(name, handler, RUNNING)) {
1085                 ERROR("failed to set state to %s",
1086                               lxc_state2str(RUNNING));
1087                 goto out_abort;
1088         }
1089
1090         lxc_sync_fini(handler);
1091
1092         return 0;
1093
1094 out_delete_net:
1095         if (cgroups_connected)
1096                 cgroup_disconnect();
1097         if (handler->clone_flags & CLONE_NEWNET)
1098                 lxc_delete_network(handler);
1099 out_abort:
1100         lxc_abort(name, handler);
1101         lxc_sync_fini(handler);
1102         if (handler->pinfd >= 0) {
1103                 close(handler->pinfd);
1104                 handler->pinfd = -1;
1105         }
1106
1107         return -1;
1108 }
1109
1110 int get_netns_fd(int pid)
1111 {
1112         char path[MAXPATHLEN];
1113         int ret, fd;
1114
1115         ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid);
1116         if (ret < 0 || ret >= MAXPATHLEN) {
1117                 WARN("Failed to pin netns file for pid %d", pid);
1118                 return -1;
1119         }
1120
1121         fd = open(path, O_RDONLY);
1122         if (fd < 0) {
1123                 WARN("Failed to pin netns file %s for pid %d: %s",
1124                                 path, pid, strerror(errno));
1125                 return -1;
1126         }
1127         return fd;
1128 }
1129
1130 int __lxc_start(const char *name, struct lxc_conf *conf,
1131                 struct lxc_operations* ops, void *data, const char *lxcpath,
1132                 bool backgrounded)
1133 {
1134         struct lxc_handler *handler;
1135         int err = -1;
1136         int status;
1137         int netnsfd = -1;
1138
1139         handler = lxc_init(name, conf, lxcpath);
1140         if (!handler) {
1141                 ERROR("failed to initialize the container");
1142                 return -1;
1143         }
1144         handler->ops = ops;
1145         handler->data = data;
1146         handler->backgrounded = backgrounded;
1147
1148         if (must_drop_cap_sys_boot(handler->conf)) {
1149                 #if HAVE_SYS_CAPABILITY_H
1150                 DEBUG("Dropping cap_sys_boot");
1151                 #else
1152                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1153                 #endif
1154         } else {
1155                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1156                 handler->conf->need_utmp_watch = 0;
1157         }
1158
1159         if (!attach_block_device(handler->conf)) {
1160                 ERROR("Failure attaching block device");
1161                 goto out_fini_nonet;
1162         }
1163
1164         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1165                 /* if the backing store is a device, mount it here and now */
1166                 if (rootfs_is_blockdev(conf)) {
1167                         if (unshare(CLONE_NEWNS) < 0) {
1168                                 ERROR("Error unsharing mounts");
1169                                 goto out_fini_nonet;
1170                         }
1171                         remount_all_slave();
1172                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1173                                 ERROR("Error setting up rootfs mount as root before spawn");
1174                                 goto out_fini_nonet;
1175                         }
1176                         INFO("Set up container rootfs as host root");
1177                 }
1178         }
1179
1180         err = lxc_spawn(handler);
1181         if (err) {
1182                 ERROR("failed to spawn '%s'", name);
1183                 goto out_detach_blockdev;
1184         }
1185
1186         handler->conf->reboot = 0;
1187
1188         netnsfd = get_netns_fd(handler->pid);
1189
1190         err = lxc_poll(name, handler);
1191         if (err) {
1192                 ERROR("mainloop exited with an error");
1193                 if (netnsfd >= 0)
1194                         close(netnsfd);
1195                 goto out_abort;
1196         }
1197
1198         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1199                 continue;
1200
1201         /*
1202          * If the child process exited but was not signaled,
1203          * it didn't call reboot.  This should mean it was an
1204          * lxc-execute which simply exited.  In any case, treat
1205          * it as a 'halt'
1206          */
1207         if (WIFSIGNALED(status)) {
1208                 switch(WTERMSIG(status)) {
1209                 case SIGINT: /* halt */
1210                         DEBUG("Container halting");
1211                         break;
1212                 case SIGHUP: /* reboot */
1213                         DEBUG("Container rebooting");
1214                         handler->conf->reboot = 1;
1215                         break;
1216                 case SIGSYS: /* seccomp */
1217                         DEBUG("Container violated its seccomp policy");
1218                         break;
1219                 default:
1220                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1221                         break;
1222                 }
1223         }
1224
1225         DEBUG("Pushing physical nics back to host namespace");
1226         lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf);
1227
1228         DEBUG("Tearing down virtual network devices used by container");
1229         lxc_delete_network(handler);
1230
1231         if (netnsfd >= 0)
1232                 close(netnsfd);
1233
1234         if (handler->pinfd >= 0) {
1235                 close(handler->pinfd);
1236                 handler->pinfd = -1;
1237         }
1238
1239         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1240         err =  lxc_error_set_and_log(handler->pid, status);
1241 out_fini:
1242         lxc_delete_network(handler);
1243
1244 out_detach_blockdev:
1245         detach_block_device(handler->conf);
1246
1247 out_fini_nonet:
1248         lxc_fini(name, handler);
1249         return err;
1250
1251 out_abort:
1252         lxc_abort(name, handler);
1253         goto out_fini;
1254 }
1255
1256 struct start_args {
1257         char *const *argv;
1258 };
1259
1260 static int start(struct lxc_handler *handler, void* data)
1261 {
1262         struct start_args *arg = data;
1263
1264         NOTICE("exec'ing '%s'", arg->argv[0]);
1265
1266         execvp(arg->argv[0], arg->argv);
1267         SYSERROR("failed to exec %s", arg->argv[0]);
1268         return 0;
1269 }
1270
1271 static int post_start(struct lxc_handler *handler, void* data)
1272 {
1273         struct start_args *arg = data;
1274
1275         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1276         return 0;
1277 }
1278
1279 static struct lxc_operations start_ops = {
1280         .start = start,
1281         .post_start = post_start
1282 };
1283
1284 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1285               const char *lxcpath, bool backgrounded)
1286 {
1287         struct start_args start_arg = {
1288                 .argv = argv,
1289         };
1290
1291         conf->need_utmp_watch = 1;
1292         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded);
1293 }