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 (match_fd(fd))
 216                         continue;
 217
 218                 if (closeall) {
 219                         close(fd);
 220                         closedir(dir);
 221                         INFO("closed inherited fd %d", fd);
 222                         goto restart;
 223                 }
 224                 WARN("inherited fd %d", fd);
 225         }
 226
 227         closedir(dir); /* cannot fail */
 228         return 0;
 229 }
 230
 231 static int setup_signal_fd(sigset_t *oldmask)
 232 {
 233         sigset_t mask;
 234         int fd;
 235
 236         /* Block everything except serious error signals */
 237         if (sigfillset(&mask) ||
 238             sigdelset(&mask, SIGILL) ||
 239             sigdelset(&mask, SIGSEGV) ||
 240             sigdelset(&mask, SIGBUS) ||
 241             sigdelset(&mask, SIGWINCH) ||
 242             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 243                 SYSERROR("failed to set signal mask");
 244                 return -1;
 245         }
 246
 247         fd = signalfd(-1, &mask, 0);
 248         if (fd < 0) {
 249                 SYSERROR("failed to create the signal fd");
 250                 return -1;
 251         }
 252
 253         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 254                 SYSERROR("failed to set sigfd to close-on-exec");
 255                 close(fd);
 256                 return -1;
 257         }
 258
 259         DEBUG("sigchild handler set");
 260
 261         return fd;
 262 }
 263
 264 static int signal_handler(int fd, uint32_t events, void *data,
 265                            struct lxc_epoll_descr *descr)
 266 {
 267         struct signalfd_siginfo siginfo;
 268         siginfo_t info;
 269         int ret;
 270         pid_t *pid = data;
 271         bool init_died = false;
 272
 273         ret = read(fd, &siginfo, sizeof(siginfo));
 274         if (ret < 0) {
 275                 ERROR("failed to read signal info");
 276                 return -1;
 277         }
 278
 279         if (ret != sizeof(siginfo)) {
 280                 ERROR("unexpected siginfo size");
 281                 return -1;
 282         }
 283
 284         // check whether init is running
 285         info.si_pid = 0;
 286         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 287         if (ret == 0 && info.si_pid == *pid) {
 288                 init_died = true;
 289         }
 290
 291         if (siginfo.ssi_signo != SIGCHLD) {
 292                 kill(*pid, siginfo.ssi_signo);
 293                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 294                 return init_died ? 1 : 0;
 295         }
 296
 297         if (siginfo.ssi_code == CLD_STOPPED ||
 298             siginfo.ssi_code == CLD_CONTINUED) {
 299                 INFO("container init process was stopped/continued");
 300                 return init_died ? 1 : 0;
 301         }
 302
 303         /* more robustness, protect ourself from a SIGCHLD sent
 304          * by a process different from the container init
 305          */
 306         if (siginfo.ssi_pid != *pid) {
 307                 WARN("invalid pid for SIGCHLD");
 308                 return init_died ? 1 : 0;
 309         }
 310
 311         DEBUG("container init process exited");
 312         return 1;
 313 }
 314
 315 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 316 {
 317         handler->state = state;
 318         lxc_monitor_send_state(name, state, handler->lxcpath);
 319         return 0;
 320 }
 321
 322 int lxc_poll(const char *name, struct lxc_handler *handler)
 323 {
 324         int sigfd = handler->sigfd;
 325         int pid = handler->pid;
 326         struct lxc_epoll_descr descr;
 327
 328         if (lxc_mainloop_open(&descr)) {
 329                 ERROR("failed to create mainloop");
 330                 goto out_sigfd;
 331         }
 332
 333         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 334                 ERROR("failed to add handler for the signal");
 335                 goto out_mainloop_open;
 336         }
 337
 338         if (lxc_console_mainloop_add(&descr, handler)) {
 339                 ERROR("failed to add console handler to mainloop");
 340                 goto out_mainloop_open;
 341         }
 342
 343         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 344                 ERROR("failed to add command handler to mainloop");
 345                 goto out_mainloop_open;
 346         }
 347
 348         if (handler->conf->need_utmp_watch) {
 349                 #if HAVE_SYS_CAPABILITY_H
 350                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 351                         ERROR("failed to add utmp handler to mainloop");
 352                         goto out_mainloop_open;
 353                 }
 354                 #else
 355                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
 356                 #endif
 357         }
 358
 359         return lxc_mainloop(&descr, -1);
 360
 361 out_mainloop_open:
 362         lxc_mainloop_close(&descr);
 363 out_sigfd:
 364         close(sigfd);
 365         return -1;
 366 }
 367
 368 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 369 {
 370         struct lxc_handler *handler;
 371
 372         handler = malloc(sizeof(*handler));
 373         if (!handler)
 374                 return NULL;
 375
 376         memset(handler, 0, sizeof(*handler));
 377
 378         handler->ttysock[0] = handler->ttysock[1] = -1;
 379         handler->conf = conf;
 380         handler->lxcpath = lxcpath;
 381         handler->pinfd = -1;
 382
 383         lsm_init();
 384
 385         handler->name = strdup(name);
 386         if (!handler->name) {
 387                 ERROR("failed to allocate memory");
 388                 goto out_free;
 389         }
 390
 391         if (lxc_cmd_init(name, handler, lxcpath))
 392                 goto out_free_name;
 393
 394         if (lxc_read_seccomp_config(conf) != 0) {
 395                 ERROR("failed loading seccomp policy");
 396                 goto out_close_maincmd_fd;
 397         }
 398
 399         /* Begin by setting the state to STARTING */
 400         if (lxc_set_state(name, handler, STARTING)) {
 401                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 402                 goto out_close_maincmd_fd;
 403         }
 404
 405         /* Start of environment variable setup for hooks */
 406         if (setenv("LXC_NAME", name, 1)) {
 407                 SYSERROR("failed to set environment variable for container name");
 408         }
 409         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 410                 SYSERROR("failed to set environment variable for config path");
 411         }
 412         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 413                 SYSERROR("failed to set environment variable for rootfs mount");
 414         }
 415         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 416                 SYSERROR("failed to set environment variable for rootfs mount");
 417         }
 418         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 419                 SYSERROR("failed to set environment variable for console path");
 420         }
 421         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 422                 SYSERROR("failed to set environment variable for console log");
 423         }
 424         /* End of environment variable setup for hooks */
 425
 426         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 427                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 428                 goto out_aborting;
 429         }
 430
 431         /* the signal fd has to be created before forking otherwise
 432          * if the child process exits before we setup the signal fd,
 433          * the event will be lost and the command will be stuck */
 434         handler->sigfd = setup_signal_fd(&handler->oldmask);
 435         if (handler->sigfd < 0) {
 436                 ERROR("failed to set sigchild fd handler");
 437                 goto out_delete_tty;
 438         }
 439
 440         /* do this after setting up signals since it might unblock SIGWINCH */
 441         if (lxc_console_create(conf)) {
 442                 ERROR("failed to create console");
 443                 goto out_restore_sigmask;
 444         }
 445
 446         if (ttys_shift_ids(conf) < 0) {
 447                 ERROR("Failed to shift tty into container");
 448                 goto out_restore_sigmask;
 449         }
 450
 451         INFO("'%s' is initialized", name);
 452         return handler;
 453
 454 out_restore_sigmask:
 455         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 456 out_delete_tty:
 457         lxc_delete_tty(&conf->tty_info);
 458 out_aborting:
 459         lxc_set_state(name, handler, ABORTING);
 460 out_close_maincmd_fd:
 461         close(conf->maincmd_fd);
 462         conf->maincmd_fd = -1;
 463 out_free_name:
 464         free(handler->name);
 465         handler->name = NULL;
 466 out_free:
 467         free(handler);
 468         return NULL;
 469 }
 470
 471 void lxc_fini(const char *name, struct lxc_handler *handler)
 472 {
 473         /* The STOPPING state is there for future cleanup code
 474          * which can take awhile
 475          */
 476         lxc_set_state(name, handler, STOPPING);
 477         lxc_set_state(name, handler, STOPPED);
 478
 479         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 480                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 481
 482         /* reset mask set by setup_signal_fd */
 483         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 484                 WARN("failed to restore sigprocmask");
 485
 486         lxc_console_delete(&handler->conf->console);
 487         lxc_delete_tty(&handler->conf->tty_info);
 488         close(handler->conf->maincmd_fd);
 489         handler->conf->maincmd_fd = -1;
 490         free(handler->name);
 491         if (handler->ttysock[0] != -1) {
 492                 close(handler->ttysock[0]);
 493                 close(handler->ttysock[1]);
 494         }
 495         cgroup_destroy(handler);
 496         free(handler);
 497 }
 498
 499 void lxc_abort(const char *name, struct lxc_handler *handler)
 500 {
 501         int ret, status;
 502
 503         lxc_set_state(name, handler, ABORTING);
 504         if (handler->pid > 0)
 505                 kill(handler->pid, SIGKILL);
 506         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 507 }
 508
 509 #include <sys/reboot.h>
 510 #include <linux/reboot.h>
 511
 512 /*
 513  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 514  * in a child pid namespace if container reboot support exists.
 515  * Otherwise, it will either succeed or return -EPERM.
 516  */
 517 static int container_reboot_supported(void *arg)
 518 {
 519         int *cmd = arg;
 520         int ret;
 521
 522         ret = reboot(*cmd);
 523         if (ret == -1 && errno == EINVAL)
 524                 return 1;
 525         return 0;
 526 }
 527
 528 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 529 {
 530         FILE *f;
 531         int ret, cmd, v, flags;
 532         long stack_size = 4096;
 533         void *stack = alloca(stack_size);
 534         int status;
 535         pid_t pid;
 536
 537         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 538         if (!f) {
 539                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 540                 return 1;
 541         }
 542
 543         ret = fscanf(f, "%d", &v);
 544         fclose(f);
 545         if (ret != 1) {
 546                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 547                 return 1;
 548         }
 549         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 550
 551         flags = CLONE_NEWPID | SIGCHLD;
 552         if (!lxc_list_empty(&conf->id_map))
 553                 flags |= CLONE_NEWUSER;
 554
 555 #ifdef __ia64__
 556         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 557 #else
 558         stack += stack_size;
 559         pid = clone(container_reboot_supported, stack, flags, &cmd);
 560 #endif
 561         if (pid < 0) {
 562                 if (flags & CLONE_NEWUSER)
 563                         ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno));
 564                 else
 565                         ERROR("failed to clone (%#x): %s", flags, strerror(errno));
 566                 return -1;
 567         }
 568         if (wait(&status) < 0) {
 569                 SYSERROR("unexpected wait error: %m");
 570                 return -1;
 571         }
 572
 573         if (WEXITSTATUS(status) != 1)
 574                 return 1;
 575
 576         return 0;
 577 }
 578
 579 /*
 580  * netpipe is used in the unprivileged case to transfer the ifindexes
 581  * from parent to child
 582  */
 583 static int netpipe = -1;
 584
 585 static inline int count_veths(struct lxc_list *network)
 586 {
 587         struct lxc_list *iterator;
 588         struct lxc_netdev *netdev;
 589         int count = 0;
 590
 591         lxc_list_for_each(iterator, network) {
 592                 netdev = iterator->elem;
 593                 if (netdev->type != LXC_NET_VETH)
 594                         continue;
 595                 count++;
 596         }
 597         return count;
 598 }
 599
 600 static int read_unpriv_netifindex(struct lxc_list *network)
 601 {
 602         struct lxc_list *iterator;
 603         struct lxc_netdev *netdev;
 604
 605         if (netpipe == -1)
 606                 return 0;
 607         lxc_list_for_each(iterator, network) {
 608                 netdev = iterator->elem;
 609                 if (netdev->type != LXC_NET_VETH)
 610                         continue;
 611                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 612                         ERROR("Out of memory");
 613                         close(netpipe);
 614                         return -1;
 615                 }
 616                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 617                         close(netpipe);
 618                         return -1;
 619                 }
 620         }
 621         close(netpipe);
 622         return 0;
 623 }
 624
 625 static int do_start(void *data)
 626 {
 627         struct lxc_list *iterator;
 628         struct lxc_handler *handler = data;
 629
 630         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 631                 SYSERROR("failed to set sigprocmask");
 632                 return -1;
 633         }
 634
 635         /* This prctl must be before the synchro, so if the parent
 636          * dies before we set the parent death signal, we will detect
 637          * its death with the synchro right after, otherwise we have
 638          * a window where the parent can exit before we set the pdeath
 639          * signal leading to a unsupervized container.
 640          */
 641         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 642                 SYSERROR("failed to set pdeath signal");
 643                 return -1;
 644         }
 645
 646         lxc_sync_fini_parent(handler);
 647
 648         /* don't leak the pinfd to the container */
 649         if (handler->pinfd >= 0) {
 650                 close(handler->pinfd);
 651         }
 652
 653         /* Tell the parent task it can begin to configure the
 654          * container and wait for it to finish
 655          */
 656         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 657                 return -1;
 658
 659         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 660                 goto out_warn_father;
 661
 662         /*
 663          * if we are in a new user namespace, become root there to have
 664          * privilege over our namespace
 665          */
 666         if (!lxc_list_empty(&handler->conf->id_map)) {
 667                 NOTICE("switching to gid/uid 0 in new user namespace");
 668                 if (setgid(0)) {
 669                         SYSERROR("setgid");
 670                         goto out_warn_father;
 671                 }
 672                 if (setuid(0)) {
 673                         SYSERROR("setuid");
 674                         goto out_warn_father;
 675                 }
 676                 if (setgroups(0, NULL)) {
 677                         SYSERROR("setgroups");
 678                         goto out_warn_father;
 679                 }
 680         }
 681
 682         if (access(handler->lxcpath, X_OK)) {
 683                 print_top_failing_dir(handler->lxcpath);
 684                 goto out_warn_father;
 685         }
 686
 687         #if HAVE_SYS_CAPABILITY_H
 688         if (handler->conf->need_utmp_watch) {
 689                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 690                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 691                         goto out_warn_father;
 692                 }
 693                 DEBUG("Dropped cap_sys_boot");
 694         }
 695         #endif
 696
 697         /* Setup the container, ip, names, utsname, ... */
 698         if (lxc_setup(handler)) {
 699                 ERROR("failed to setup the container");
 700                 goto out_warn_father;
 701         }
 702
 703         /* ask father to setup cgroups and wait for him to finish */
 704         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 705                 return -1;
 706
 707         /* Set the label to change to when we exec(2) the container's init */
 708         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 709                 goto out_warn_father;
 710
 711         /* Some init's such as busybox will set sane tty settings on stdin,
 712          * stdout, stderr which it thinks is the console. We already set them
 713          * the way we wanted on the real terminal, and we want init to do its
 714          * setup on its console ie. the pty allocated in lxc_console_create()
 715          * so make sure that that pty is stdin,stdout,stderr.
 716          */
 717         if (lxc_console_set_stdfds(handler) < 0)
 718                 goto out_warn_father;
 719
 720         /* If we mounted a temporary proc, then unmount it now */
 721         tmp_proc_unmount(handler->conf);
 722
 723         if (lxc_seccomp_load(handler->conf) != 0)
 724                 goto out_warn_father;
 725
 726         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 727                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 728                 goto out_warn_father;
 729         }
 730
 731         /* The clearenv() and putenv() calls have been moved here
 732          * to allow us to use environment variables passed to the various
 733          * hooks, such as the start hook above.  Not all of the
 734          * variables like CONFIG_PATH or ROOTFS are valid in this
 735          * context but others are. */
 736         if (clearenv()) {
 737                 SYSERROR("failed to clear environment");
 738                 /* don't error out though */
 739         }
 740
 741         lxc_list_for_each(iterator, &handler->conf->environment) {
 742                 if (putenv((char *)iterator->elem)) {
 743                         SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem);
 744                         goto out_warn_father;
 745                 }
 746         }
 747
 748         if (putenv("container=lxc")) {
 749                 SYSERROR("failed to set environment variable 'container=lxc'");
 750                 goto out_warn_father;
 751         }
 752
 753         if (handler->conf->pty_names) {
 754                 if (putenv(handler->conf->pty_names)) {
 755                         SYSERROR("failed to set environment variable for container ptys");
 756                         goto out_warn_father;
 757                 }
 758         }
 759
 760         close(handler->sigfd);
 761
 762         if (handler->backgrounded) {
 763                 close(0);
 764                 close(1);
 765                 close(2);
 766                 open("/dev/zero", O_RDONLY);
 767                 open("/dev/null", O_RDWR);
 768                 open("/dev/null", O_RDWR);
 769         }
 770
 771         /* after this call, we are in error because this
 772          * ops should not return as it execs */
 773         handler->ops->start(handler, handler->data);
 774
 775 out_warn_father:
 776         /* we want the parent to know something went wrong, so any
 777          * value other than what it expects is ok. */
 778         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 779         return -1;
 780 }
 781
 782 static int save_phys_nics(struct lxc_conf *conf)
 783 {
 784         struct lxc_list *iterator;
 785
 786         lxc_list_for_each(iterator, &conf->network) {
 787                 struct lxc_netdev *netdev = iterator->elem;
 788
 789                 if (netdev->type != LXC_NET_PHYS)
 790                         continue;
 791                 conf->saved_nics = realloc(conf->saved_nics,
 792                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 793                 if (!conf->saved_nics) {
 794                         SYSERROR("failed to allocate memory");
 795                         return -1;
 796                 }
 797                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 798                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 799                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 800                         SYSERROR("failed to allocate memory");
 801                         return -1;
 802                 }
 803                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 804                         conf->saved_nics[conf->num_savednics].ifindex,
 805                         conf->saved_nics[conf->num_savednics].orig_name);
 806                 conf->num_savednics++;
 807         }
 808
 809         return 0;
 810 }
 811
 812 static int recv_fd(int sock, int *fd)
 813 {
 814         if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) {
 815                 SYSERROR("Error receiving tty fd from child");
 816                 return -1;
 817         }
 818         if (*fd == -1)
 819                 return -1;
 820         return 0;
 821 }
 822
 823 static int recv_ttys_from_child(struct lxc_handler *handler)
 824 {
 825         struct lxc_conf *conf = handler->conf;
 826         int i, sock = handler->ttysock[1];
 827         struct lxc_tty_info *tty_info = &conf->tty_info;
 828
 829         if (!conf->tty)
 830                 return 0;
 831
 832         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty);
 833         if (!tty_info->pty_info) {
 834                 SYSERROR("failed to allocate pty_info");
 835                 return -1;
 836         }
 837
 838         for (i = 0; i < conf->tty; i++) {
 839                 struct lxc_pty_info *pty_info = &tty_info->pty_info[i];
 840                 pty_info->busy = 0;
 841                 if (recv_fd(sock, &pty_info->slave) < 0 ||
 842                                 recv_fd(sock, &pty_info->master) < 0) {
 843                         ERROR("Error receiving tty info from child");
 844                         return -1;
 845                 }
 846         }
 847         tty_info->nbtty = conf->tty;
 848
 849         return 0;
 850 }
 851
 852 void resolve_clone_flags(struct lxc_handler *handler)
 853 {
 854         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
 855
 856         if (!lxc_list_empty(&handler->conf->id_map)) {
 857                 INFO("Cloning a new user namespace");
 858                 handler->clone_flags |= CLONE_NEWUSER;
 859         }
 860
 861         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 862                 if (!lxc_requests_empty_network(handler))
 863                         handler->clone_flags |= CLONE_NEWNET;
 864         } else {
 865                 INFO("Inheriting a net namespace");
 866         }
 867
 868         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 869                 handler->clone_flags |= CLONE_NEWIPC;
 870         } else {
 871                 INFO("Inheriting an IPC namespace");
 872         }
 873
 874         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 875                 handler->clone_flags |= CLONE_NEWUTS;
 876         } else {
 877                 INFO("Inheriting a UTS namespace");
 878         }
 879 }
 880
 881 static int lxc_spawn(struct lxc_handler *handler)
 882 {
 883         int failed_before_rename = 0;
 884         const char *name = handler->name;
 885         bool cgroups_connected = false;
 886         int saved_ns_fd[LXC_NS_MAX];
 887         int preserve_mask = 0, i;
 888         int netpipepair[2], nveths;
 889
 890         netpipe = -1;
 891
 892         for (i = 0; i < LXC_NS_MAX; i++)
 893                 if (handler->conf->inherit_ns_fd[i] != -1)
 894                         preserve_mask |= ns_info[i].clone_flag;
 895
 896         if (lxc_sync_init(handler))
 897                 return -1;
 898
 899         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
 900                 lxc_sync_fini(handler);
 901                 return -1;
 902         }
 903
 904         resolve_clone_flags(handler);
 905
 906         if (handler->clone_flags & CLONE_NEWNET) {
 907                 if (!lxc_list_empty(&handler->conf->network)) {
 908
 909                         /* Find gateway addresses from the link device, which is
 910                          * no longer accessible inside the container. Do this
 911                          * before creating network interfaces, since goto
 912                          * out_delete_net does not work before lxc_clone. */
 913                         if (lxc_find_gateway_addresses(handler)) {
 914                                 ERROR("failed to find gateway addresses");
 915                                 lxc_sync_fini(handler);
 916                                 return -1;
 917                         }
 918
 919                         /* that should be done before the clone because we will
 920                          * fill the netdev index and use them in the child
 921                          */
 922                         if (lxc_create_network(handler)) {
 923                                 ERROR("failed to create the network");
 924                                 lxc_sync_fini(handler);
 925                                 return -1;
 926                         }
 927                 }
 928
 929                 if (save_phys_nics(handler->conf)) {
 930                         ERROR("failed to save physical nic info");
 931                         goto out_abort;
 932                 }
 933         }
 934
 935         if (!cgroup_init(handler)) {
 936                 ERROR("failed initializing cgroup support");
 937                 goto out_delete_net;
 938         }
 939
 940         cgroups_connected = true;
 941
 942         if (!cgroup_create(handler)) {
 943                 ERROR("failed creating cgroups");
 944                 goto out_delete_net;
 945         }
 946
 947         /*
 948          * if the rootfs is not a blockdev, prevent the container from
 949          * marking it readonly.
 950          *
 951          * if the container is unprivileged then skip rootfs pinning
 952          */
 953         if (lxc_list_empty(&handler->conf->id_map)) {
 954                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 955                 if (handler->pinfd == -1)
 956                         INFO("failed to pin the container's rootfs");
 957         }
 958
 959         if (preserve_ns(saved_ns_fd, preserve_mask) < 0)
 960                 goto out_delete_net;
 961         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
 962                 goto out_delete_net;
 963
 964         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
 965                 if (pipe(netpipepair) < 0) {
 966                         SYSERROR("Error creating pipe");
 967                         goto out_delete_net;
 968                 }
 969                 /* store netpipe in the global var for do_start's use */
 970                 netpipe = netpipepair[0];
 971         }
 972
 973         /* Create a process in a new set of namespaces */
 974         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 975         if (handler->pid < 0) {
 976                 SYSERROR("failed to fork into a new namespace");
 977                 goto out_delete_net;
 978         }
 979
 980         if (attach_ns(saved_ns_fd))
 981                 WARN("failed to restore saved namespaces");
 982
 983         lxc_sync_fini_child(handler);
 984
 985         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 986                 failed_before_rename = 1;
 987
 988         if (!cgroup_create_legacy(handler)) {
 989                 ERROR("failed to setup the legacy cgroups for %s", name);
 990                 goto out_delete_net;
 991         }
 992         if (!cgroup_setup_limits(handler, false)) {
 993                 ERROR("failed to setup the cgroup limits for '%s'", name);
 994                 goto out_delete_net;
 995         }
 996
 997         if (!cgroup_enter(handler))
 998                 goto out_delete_net;
 999
1000         if (!cgroup_chown(handler))
1001                 goto out_delete_net;
1002
1003         if (failed_before_rename)
1004                 goto out_delete_net;
1005
1006         /* Create the network configuration */
1007         if (handler->clone_flags & CLONE_NEWNET) {
1008                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
1009                         ERROR("failed to create the configured network");
1010                         goto out_delete_net;
1011                 }
1012         }
1013
1014         if (netpipe != -1) {
1015                 struct lxc_list *iterator;
1016                 struct lxc_netdev *netdev;
1017
1018                 close(netpipe);
1019                 lxc_list_for_each(iterator, &handler->conf->network) {
1020                         netdev = iterator->elem;
1021                         if (netdev->type != LXC_NET_VETH)
1022                                 continue;
1023                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1024                                 ERROR("Error writing veth name to container");
1025                                 goto out_delete_net;
1026                         }
1027                 }
1028                 close(netpipepair[1]);
1029         }
1030
1031         /* map the container uids - the container became an invalid
1032          * userid the moment it was cloned with CLONE_NEWUSER - this
1033          * call doesn't change anything immediately, but allows the
1034          * container to setuid(0) (0 being mapped to something else on
1035          * the host) later to become a valid uid again */
1036         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
1037                 ERROR("failed to set up id mapping");
1038                 goto out_delete_net;
1039         }
1040
1041         /* Tell the child to continue its initialization.  we'll get
1042          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1043          */
1044         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1045                 goto out_delete_net;
1046
1047         if (!cgroup_setup_limits(handler, true)) {
1048                 ERROR("failed to setup the devices cgroup for '%s'", name);
1049                 goto out_delete_net;
1050         }
1051
1052         cgroup_disconnect();
1053         cgroups_connected = false;
1054
1055         /* read tty fds allocated by child */
1056         if (recv_ttys_from_child(handler) < 0) {
1057                 ERROR("failed to receive tty info from child");
1058                 goto out_delete_net;
1059         }
1060
1061         /* Tell the child to complete its initialization and wait for
1062          * it to exec or return an error.  (the child will never
1063          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
1064          * sync pipe, causing lxc_sync_barrier_child to return
1065          * success, or return a different value, causing us to error
1066          * out).
1067          */
1068         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1069                 return -1;
1070
1071         if (detect_shared_rootfs())
1072                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
1073
1074         if (handler->ops->post_start(handler, handler->data))
1075                 goto out_abort;
1076
1077         if (lxc_set_state(name, handler, RUNNING)) {
1078                 ERROR("failed to set state to %s",
1079                               lxc_state2str(RUNNING));
1080                 goto out_abort;
1081         }
1082
1083         lxc_sync_fini(handler);
1084
1085         return 0;
1086
1087 out_delete_net:
1088         if (cgroups_connected)
1089                 cgroup_disconnect();
1090         if (handler->clone_flags & CLONE_NEWNET)
1091                 lxc_delete_network(handler);
1092 out_abort:
1093         lxc_abort(name, handler);
1094         lxc_sync_fini(handler);
1095         if (handler->pinfd >= 0) {
1096                 close(handler->pinfd);
1097                 handler->pinfd = -1;
1098         }
1099
1100         return -1;
1101 }
1102
1103 int get_netns_fd(int pid)
1104 {
1105         char path[MAXPATHLEN];
1106         int ret, fd;
1107
1108         ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid);
1109         if (ret < 0 || ret >= MAXPATHLEN) {
1110                 WARN("Failed to pin netns file for pid %d", pid);
1111                 return -1;
1112         }
1113
1114         fd = open(path, O_RDONLY);
1115         if (fd < 0) {
1116                 WARN("Failed to pin netns file %s for pid %d: %s",
1117                                 path, pid, strerror(errno));
1118                 return -1;
1119         }
1120         return fd;
1121 }
1122
1123 int __lxc_start(const char *name, struct lxc_conf *conf,
1124                 struct lxc_operations* ops, void *data, const char *lxcpath,
1125                 bool backgrounded)
1126 {
1127         struct lxc_handler *handler;
1128         int err = -1;
1129         int status;
1130         int netnsfd = -1;
1131
1132         handler = lxc_init(name, conf, lxcpath);
1133         if (!handler) {
1134                 ERROR("failed to initialize the container");
1135                 return -1;
1136         }
1137         handler->ops = ops;
1138         handler->data = data;
1139         handler->backgrounded = backgrounded;
1140
1141         if (must_drop_cap_sys_boot(handler->conf)) {
1142                 #if HAVE_SYS_CAPABILITY_H
1143                 DEBUG("Dropping cap_sys_boot");
1144                 #else
1145                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1146                 #endif
1147         } else {
1148                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1149                 handler->conf->need_utmp_watch = 0;
1150         }
1151
1152         if (!attach_block_device(handler->conf)) {
1153                 ERROR("Failure attaching block device");
1154                 goto out_fini_nonet;
1155         }
1156
1157         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1158                 /* if the backing store is a device, mount it here and now */
1159                 if (rootfs_is_blockdev(conf)) {
1160                         if (unshare(CLONE_NEWNS) < 0) {
1161                                 ERROR("Error unsharing mounts");
1162                                 goto out_fini_nonet;
1163                         }
1164                         remount_all_slave();
1165                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1166                                 ERROR("Error setting up rootfs mount as root before spawn");
1167                                 goto out_fini_nonet;
1168                         }
1169                         INFO("Set up container rootfs as host root");
1170                 }
1171         }
1172
1173         err = lxc_spawn(handler);
1174         if (err) {
1175                 ERROR("failed to spawn '%s'", name);
1176                 goto out_detach_blockdev;
1177         }
1178
1179         netnsfd = get_netns_fd(handler->pid);
1180
1181         err = lxc_poll(name, handler);
1182         if (err) {
1183                 ERROR("mainloop exited with an error");
1184                 if (netnsfd >= 0)
1185                         close(netnsfd);
1186                 goto out_abort;
1187         }
1188
1189         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1190                 continue;
1191
1192         /*
1193          * If the child process exited but was not signaled,
1194          * it didn't call reboot.  This should mean it was an
1195          * lxc-execute which simply exited.  In any case, treat
1196          * it as a 'halt'
1197          */
1198         if (WIFSIGNALED(status)) {
1199                 switch(WTERMSIG(status)) {
1200                 case SIGINT: /* halt */
1201                         DEBUG("Container halting");
1202                         break;
1203                 case SIGHUP: /* reboot */
1204                         DEBUG("Container rebooting");
1205                         handler->conf->reboot = 1;
1206                         break;
1207                 case SIGSYS: /* seccomp */
1208                         DEBUG("Container violated its seccomp policy");
1209                         break;
1210                 default:
1211                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1212                         break;
1213                 }
1214         }
1215
1216         lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf);
1217         if (netnsfd >= 0)
1218                 close(netnsfd);
1219
1220         if (handler->pinfd >= 0) {
1221                 close(handler->pinfd);
1222                 handler->pinfd = -1;
1223         }
1224
1225         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1226         err =  lxc_error_set_and_log(handler->pid, status);
1227 out_fini:
1228         lxc_delete_network(handler);
1229
1230 out_detach_blockdev:
1231         detach_block_device(handler->conf);
1232
1233 out_fini_nonet:
1234         lxc_fini(name, handler);
1235         return err;
1236
1237 out_abort:
1238         lxc_abort(name, handler);
1239         goto out_fini;
1240 }
1241
1242 struct start_args {
1243         char *const *argv;
1244 };
1245
1246 static int start(struct lxc_handler *handler, void* data)
1247 {
1248         struct start_args *arg = data;
1249
1250         NOTICE("exec'ing '%s'", arg->argv[0]);
1251
1252         execvp(arg->argv[0], arg->argv);
1253         SYSERROR("failed to exec %s", arg->argv[0]);
1254         return 0;
1255 }
1256
1257 static int post_start(struct lxc_handler *handler, void* data)
1258 {
1259         struct start_args *arg = data;
1260
1261         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1262         return 0;
1263 }
1264
1265 static struct lxc_operations start_ops = {
1266         .start = start,
1267         .post_start = post_start
1268 };
1269
1270 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1271               const char *lxcpath, bool backgrounded)
1272 {
1273         struct start_args start_arg = {
1274                 .argv = argv,
1275         };
1276
1277         conf->need_utmp_watch = 1;
1278         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded);
1279 }