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