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