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 "lsm/lsm.h"
  73
  74 lxc_log_define(lxc_start, lxc);
  75
  76 const struct ns_info ns_info[LXC_NS_MAX] = {
  77         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  78         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  79         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  80         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  81         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  82         [LXC_NS_NET] = {"net", CLONE_NEWNET}
  83 };
  84
  85 static void print_top_failing_dir(const char *path)
  86 {
  87         size_t len = strlen(path);
  88         char *copy = alloca(len+1), *p, *e, saved;
  89         strcpy(copy, path);
  90
  91         p = copy;
  92         e = copy + len;
  93         while (p < e) {
  94                 while (p < e && *p == '/') p++;
  95                 while (p < e && *p != '/') p++;
  96                 saved = *p;
  97                 *p = '\0';
  98                 if (access(copy, X_OK)) {
  99                         SYSERROR("could not access %s.  Please grant it 'x' " \
 100                               "access, or add an ACL for the container root.",
 101                               copy);
 102                         return;
 103                 }
 104                 *p = saved;
 105         }
 106 }
 107
 108 static void close_ns(int ns_fd[LXC_NS_MAX]) {
 109         int i;
 110
 111         for (i = 0; i < LXC_NS_MAX; i++) {
 112                 if (ns_fd[i] > -1) {
 113                         close(ns_fd[i]);
 114                         ns_fd[i] = -1;
 115                 }
 116         }
 117 }
 118
 119 static int preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags) {
 120         int i, saved_errno;
 121         char path[MAXPATHLEN];
 122
 123         for (i = 0; i < LXC_NS_MAX; i++)
 124                 ns_fd[i] = -1;
 125
 126         if (access("/proc/self/ns", X_OK)) {
 127                 WARN("Kernel does not support attach; preserve_ns ignored");
 128                 return 0;
 129         }
 130
 131         for (i = 0; i < LXC_NS_MAX; i++) {
 132                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 133                         continue;
 134                 snprintf(path, MAXPATHLEN, "/proc/self/ns/%s", ns_info[i].proc_name);
 135                 ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC);
 136                 if (ns_fd[i] < 0)
 137                         goto error;
 138         }
 139
 140         return 0;
 141
 142 error:
 143         saved_errno = errno;
 144         close_ns(ns_fd);
 145         errno = saved_errno;
 146         SYSERROR("failed to open '%s'", path);
 147         return -1;
 148 }
 149
 150 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 151         int i;
 152
 153         for (i = 0; i < LXC_NS_MAX; i++) {
 154                 if (ns_fd[i] < 0)
 155                         continue;
 156
 157                 if (setns(ns_fd[i], 0) != 0)
 158                         goto error;
 159         }
 160         return 0;
 161
 162 error:
 163         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 164         return -1;
 165 }
 166
 167 static int match_fd(int fd)
 168 {
 169         return (fd == 0 || fd == 1 || fd == 2);
 170 }
 171
 172 int lxc_check_inherited(struct lxc_conf *conf, int fd_to_ignore)
 173 {
 174         struct dirent dirent, *direntp;
 175         int fd, fddir;
 176         DIR *dir;
 177
 178 restart:
 179         dir = opendir("/proc/self/fd");
 180         if (!dir) {
 181                 WARN("failed to open directory: %m");
 182                 return -1;
 183         }
 184
 185         fddir = dirfd(dir);
 186
 187         while (!readdir_r(dir, &dirent, &direntp)) {
 188                 if (!direntp)
 189                         break;
 190
 191                 if (!strcmp(direntp->d_name, "."))
 192                         continue;
 193
 194                 if (!strcmp(direntp->d_name, ".."))
 195                         continue;
 196
 197                 fd = atoi(direntp->d_name);
 198
 199                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 200                         continue;
 201
 202                 if (match_fd(fd))
 203                         continue;
 204
 205                 if (conf->close_all_fds) {
 206                         close(fd);
 207                         closedir(dir);
 208                         INFO("closed inherited fd %d", fd);
 209                         goto restart;
 210                 }
 211                 WARN("inherited fd %d", fd);
 212         }
 213
 214         closedir(dir); /* cannot fail */
 215         return 0;
 216 }
 217
 218 static int setup_signal_fd(sigset_t *oldmask)
 219 {
 220         sigset_t mask;
 221         int fd;
 222
 223         /* Block everything except serious error signals */
 224         if (sigfillset(&mask) ||
 225             sigdelset(&mask, SIGILL) ||
 226             sigdelset(&mask, SIGSEGV) ||
 227             sigdelset(&mask, SIGBUS) ||
 228             sigdelset(&mask, SIGWINCH) ||
 229             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 230                 SYSERROR("failed to set signal mask");
 231                 return -1;
 232         }
 233
 234         fd = signalfd(-1, &mask, 0);
 235         if (fd < 0) {
 236                 SYSERROR("failed to create the signal fd");
 237                 return -1;
 238         }
 239
 240         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 241                 SYSERROR("failed to set sigfd to close-on-exec");
 242                 close(fd);
 243                 return -1;
 244         }
 245
 246         DEBUG("sigchild handler set");
 247
 248         return fd;
 249 }
 250
 251 static int signal_handler(int fd, uint32_t events, void *data,
 252                            struct lxc_epoll_descr *descr)
 253 {
 254         struct signalfd_siginfo siginfo;
 255         siginfo_t info;
 256         int ret;
 257         pid_t *pid = data;
 258         bool init_died = false;
 259
 260         ret = read(fd, &siginfo, sizeof(siginfo));
 261         if (ret < 0) {
 262                 ERROR("failed to read signal info");
 263                 return -1;
 264         }
 265
 266         if (ret != sizeof(siginfo)) {
 267                 ERROR("unexpected siginfo size");
 268                 return -1;
 269         }
 270
 271         // check whether init is running
 272         info.si_pid = 0;
 273         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 274         if (ret == 0 && info.si_pid == *pid) {
 275                 init_died = true;
 276         }
 277
 278         if (siginfo.ssi_signo != SIGCHLD) {
 279                 kill(*pid, siginfo.ssi_signo);
 280                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 281                 return init_died ? 1 : 0;
 282         }
 283
 284         if (siginfo.ssi_code == CLD_STOPPED ||
 285             siginfo.ssi_code == CLD_CONTINUED) {
 286                 INFO("container init process was stopped/continued");
 287                 return init_died ? 1 : 0;
 288         }
 289
 290         /* more robustness, protect ourself from a SIGCHLD sent
 291          * by a process different from the container init
 292          */
 293         if (siginfo.ssi_pid != *pid) {
 294                 WARN("invalid pid for SIGCHLD");
 295                 return init_died ? 1 : 0;
 296         }
 297
 298         DEBUG("container init process exited");
 299         return 1;
 300 }
 301
 302 static int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 303 {
 304         handler->state = state;
 305         lxc_monitor_send_state(name, state, handler->lxcpath);
 306         return 0;
 307 }
 308
 309 static int lxc_poll(const char *name, struct lxc_handler *handler)
 310 {
 311         int sigfd = handler->sigfd;
 312         int pid = handler->pid;
 313         struct lxc_epoll_descr descr;
 314
 315         if (lxc_mainloop_open(&descr)) {
 316                 ERROR("failed to create mainloop");
 317                 goto out_sigfd;
 318         }
 319
 320         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 321                 ERROR("failed to add handler for the signal");
 322                 goto out_mainloop_open;
 323         }
 324
 325         if (lxc_console_mainloop_add(&descr, handler)) {
 326                 ERROR("failed to add console handler to mainloop");
 327                 goto out_mainloop_open;
 328         }
 329
 330         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 331                 ERROR("failed to add command handler to mainloop");
 332                 goto out_mainloop_open;
 333         }
 334
 335         if (handler->conf->need_utmp_watch) {
 336                 #if HAVE_SYS_CAPABILITY_H
 337                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 338                         ERROR("failed to add utmp handler to mainloop");
 339                         goto out_mainloop_open;
 340                 }
 341                 #else
 342                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
 343                 #endif
 344         }
 345
 346         return lxc_mainloop(&descr, -1);
 347
 348 out_mainloop_open:
 349         lxc_mainloop_close(&descr);
 350 out_sigfd:
 351         close(sigfd);
 352         return -1;
 353 }
 354
 355 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 356 {
 357         struct lxc_handler *handler;
 358
 359         handler = malloc(sizeof(*handler));
 360         if (!handler)
 361                 return NULL;
 362
 363         memset(handler, 0, sizeof(*handler));
 364
 365         handler->conf = conf;
 366         handler->lxcpath = lxcpath;
 367         handler->pinfd = -1;
 368
 369         lsm_init();
 370
 371         handler->name = strdup(name);
 372         if (!handler->name) {
 373                 ERROR("failed to allocate memory");
 374                 goto out_free;
 375         }
 376
 377         if (lxc_cmd_init(name, handler, lxcpath))
 378                 goto out_free_name;
 379
 380         if (lxc_read_seccomp_config(conf) != 0) {
 381                 ERROR("failed loading seccomp policy");
 382                 goto out_close_maincmd_fd;
 383         }
 384
 385         /* Begin by setting the state to STARTING */
 386         if (lxc_set_state(name, handler, STARTING)) {
 387                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 388                 goto out_close_maincmd_fd;
 389         }
 390
 391         /* Start of environment variable setup for hooks */
 392         if (setenv("LXC_NAME", name, 1)) {
 393                 SYSERROR("failed to set environment variable for container name");
 394         }
 395         if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 396                 SYSERROR("failed to set environment variable for config path");
 397         }
 398         if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 399                 SYSERROR("failed to set environment variable for rootfs mount");
 400         }
 401         if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 402                 SYSERROR("failed to set environment variable for rootfs mount");
 403         }
 404         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 405                 SYSERROR("failed to set environment variable for console path");
 406         }
 407         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 408                 SYSERROR("failed to set environment variable for console log");
 409         }
 410         /* End of environment variable setup for hooks */
 411
 412         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 413                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 414                 goto out_aborting;
 415         }
 416
 417         if (lxc_create_tty(name, conf)) {
 418                 ERROR("failed to create the ttys");
 419                 goto out_aborting;
 420         }
 421
 422         /* the signal fd has to be created before forking otherwise
 423          * if the child process exits before we setup the signal fd,
 424          * the event will be lost and the command will be stuck */
 425         handler->sigfd = setup_signal_fd(&handler->oldmask);
 426         if (handler->sigfd < 0) {
 427                 ERROR("failed to set sigchild fd handler");
 428                 goto out_delete_tty;
 429         }
 430
 431         /* do this after setting up signals since it might unblock SIGWINCH */
 432         if (lxc_console_create(conf)) {
 433                 ERROR("failed to create console");
 434                 goto out_restore_sigmask;
 435         }
 436
 437         if (ttys_shift_ids(conf) < 0) {
 438                 ERROR("Failed to shift tty into container");
 439                 goto out_restore_sigmask;
 440         }
 441
 442         INFO("'%s' is initialized", name);
 443         return handler;
 444
 445 out_restore_sigmask:
 446         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 447 out_delete_tty:
 448         lxc_delete_tty(&conf->tty_info);
 449 out_aborting:
 450         lxc_set_state(name, handler, ABORTING);
 451 out_close_maincmd_fd:
 452         close(conf->maincmd_fd);
 453         conf->maincmd_fd = -1;
 454 out_free_name:
 455         free(handler->name);
 456         handler->name = NULL;
 457 out_free:
 458         free(handler);
 459         return NULL;
 460 }
 461
 462 static void lxc_fini(const char *name, struct lxc_handler *handler)
 463 {
 464         /* The STOPPING state is there for future cleanup code
 465          * which can take awhile
 466          */
 467         lxc_set_state(name, handler, STOPPING);
 468         lxc_set_state(name, handler, STOPPED);
 469
 470         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 471                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 472
 473         /* reset mask set by setup_signal_fd */
 474         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 475                 WARN("failed to restore sigprocmask");
 476
 477         lxc_console_delete(&handler->conf->console);
 478         lxc_delete_tty(&handler->conf->tty_info);
 479         close(handler->conf->maincmd_fd);
 480         handler->conf->maincmd_fd = -1;
 481         free(handler->name);
 482         cgroup_destroy(handler);
 483         free(handler);
 484 }
 485
 486 static void lxc_abort(const char *name, struct lxc_handler *handler)
 487 {
 488         int ret, status;
 489
 490         lxc_set_state(name, handler, ABORTING);
 491         if (handler->pid > 0)
 492                 kill(handler->pid, SIGKILL);
 493         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 494 }
 495
 496 #include <sys/reboot.h>
 497 #include <linux/reboot.h>
 498
 499 /*
 500  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 501  * in a child pid namespace if container reboot support exists.
 502  * Otherwise, it will either succeed or return -EPERM.
 503  */
 504 static int container_reboot_supported(void *arg)
 505 {
 506         int *cmd = arg;
 507         int ret;
 508
 509         ret = reboot(*cmd);
 510         if (ret == -1 && errno == EINVAL)
 511                 return 1;
 512         return 0;
 513 }
 514
 515 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 516 {
 517         FILE *f;
 518         int ret, cmd, v, flags;
 519         long stack_size = 4096;
 520         void *stack = alloca(stack_size);
 521         int status;
 522         pid_t pid;
 523
 524         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 525         if (!f) {
 526                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 527                 return 1;
 528         }
 529
 530         ret = fscanf(f, "%d", &v);
 531         fclose(f);
 532         if (ret != 1) {
 533                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 534                 return 1;
 535         }
 536         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 537
 538         flags = CLONE_NEWPID | SIGCHLD;
 539         if (!lxc_list_empty(&conf->id_map))
 540                 flags |= CLONE_NEWUSER;
 541
 542 #ifdef __ia64__
 543         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 544 #else
 545         stack += stack_size;
 546         pid = clone(container_reboot_supported, stack, flags, &cmd);
 547 #endif
 548         if (pid < 0) {
 549                 SYSERROR("failed to clone");
 550                 return -1;
 551         }
 552         if (wait(&status) < 0) {
 553                 SYSERROR("unexpected wait error: %m");
 554                 return -1;
 555         }
 556
 557         if (WEXITSTATUS(status) != 1)
 558                 return 1;
 559
 560         return 0;
 561 }
 562
 563 /*
 564  * netpipe is used in the unprivileged case to transfer the ifindexes
 565  * from parent to child
 566  */
 567 static int netpipe = -1;
 568
 569 static inline int count_veths(struct lxc_list *network)
 570 {
 571         struct lxc_list *iterator;
 572         struct lxc_netdev *netdev;
 573         int count = 0;
 574
 575         lxc_list_for_each(iterator, network) {
 576                 netdev = iterator->elem;
 577                 if (netdev->type != LXC_NET_VETH)
 578                         continue;
 579                 count++;
 580         }
 581         return count;
 582 }
 583
 584 static int read_unpriv_netifindex(struct lxc_list *network)
 585 {
 586         struct lxc_list *iterator;
 587         struct lxc_netdev *netdev;
 588
 589         if (netpipe == -1)
 590                 return 0;
 591         lxc_list_for_each(iterator, network) {
 592                 netdev = iterator->elem;
 593                 if (netdev->type != LXC_NET_VETH)
 594                         continue;
 595                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 596                         ERROR("Out of memory");
 597                         close(netpipe);
 598                         return -1;
 599                 }
 600                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 601                         close(netpipe);
 602                         return -1;
 603                 }
 604         }
 605         close(netpipe);
 606         return 0;
 607 }
 608
 609 static int do_start(void *data)
 610 {
 611         struct lxc_handler *handler = data;
 612         const char *lsm_label = NULL;
 613
 614         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 615                 SYSERROR("failed to set sigprocmask");
 616                 return -1;
 617         }
 618
 619         /* This prctl must be before the synchro, so if the parent
 620          * dies before we set the parent death signal, we will detect
 621          * its death with the synchro right after, otherwise we have
 622          * a window where the parent can exit before we set the pdeath
 623          * signal leading to a unsupervized container.
 624          */
 625         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 626                 SYSERROR("failed to set pdeath signal");
 627                 return -1;
 628         }
 629
 630         lxc_sync_fini_parent(handler);
 631
 632         /* don't leak the pinfd to the container */
 633         if (handler->pinfd >= 0) {
 634                 close(handler->pinfd);
 635         }
 636
 637         /* Tell the parent task it can begin to configure the
 638          * container and wait for it to finish
 639          */
 640         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 641                 return -1;
 642
 643         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 644                 goto out_warn_father;
 645
 646         /*
 647          * if we are in a new user namespace, become root there to have
 648          * privilege over our namespace
 649          */
 650         if (!lxc_list_empty(&handler->conf->id_map)) {
 651                 NOTICE("switching to gid/uid 0 in new user namespace");
 652                 if (setgid(0)) {
 653                         SYSERROR("setgid");
 654                         goto out_warn_father;
 655                 }
 656                 if (setuid(0)) {
 657                         SYSERROR("setuid");
 658                         goto out_warn_father;
 659                 }
 660                 if (setgroups(0, NULL)) {
 661                         SYSERROR("setgroups");
 662                         goto out_warn_father;
 663                 }
 664         }
 665
 666         if (access(handler->lxcpath, X_OK)) {
 667                 print_top_failing_dir(handler->lxcpath);
 668                 goto out_warn_father;
 669         }
 670
 671         #if HAVE_SYS_CAPABILITY_H
 672         if (handler->conf->need_utmp_watch) {
 673                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 674                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 675                         goto out_warn_father;
 676                 }
 677                 DEBUG("Dropped cap_sys_boot");
 678         }
 679         #endif
 680
 681         /* Setup the container, ip, names, utsname, ... */
 682         if (lxc_setup(handler)) {
 683                 ERROR("failed to setup the container");
 684                 goto out_warn_father;
 685         }
 686
 687         /* ask father to setup cgroups and wait for him to finish */
 688         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 689                 return -1;
 690
 691         /* Set the label to change to when we exec(2) the container's init */
 692         if (!strcmp(lsm_name(), "AppArmor"))
 693                 lsm_label = handler->conf->lsm_aa_profile;
 694         else if (!strcmp(lsm_name(), "SELinux"))
 695                 lsm_label = handler->conf->lsm_se_context;
 696         if (lsm_process_label_set(lsm_label, 1, 1) < 0)
 697                 goto out_warn_father;
 698
 699         /* Some init's such as busybox will set sane tty settings on stdin,
 700          * stdout, stderr which it thinks is the console. We already set them
 701          * the way we wanted on the real terminal, and we want init to do its
 702          * setup on its console ie. the pty allocated in lxc_console_create()
 703          * so make sure that that pty is stdin,stdout,stderr.
 704          */
 705         if (lxc_console_set_stdfds(handler) < 0)
 706                 goto out_warn_father;
 707
 708         /* If we mounted a temporary proc, then unmount it now */
 709         tmp_proc_unmount(handler->conf);
 710
 711         if (lxc_seccomp_load(handler->conf) != 0)
 712                 goto out_warn_father;
 713
 714         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 715                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 716                 goto out_warn_father;
 717         }
 718
 719         /* The clearenv() and putenv() calls have been moved here
 720          * to allow us to use enviroment variables passed to the various
 721          * hooks, such as the start hook above.  Not all of the
 722          * variables like CONFIG_PATH or ROOTFS are valid in this
 723          * context but others are. */
 724         if (clearenv()) {
 725                 SYSERROR("failed to clear environment");
 726                 /* don't error out though */
 727         }
 728
 729         if (putenv("container=lxc")) {
 730                 SYSERROR("failed to set environment variable");
 731                 goto out_warn_father;
 732         }
 733
 734         close(handler->sigfd);
 735
 736         /* after this call, we are in error because this
 737          * ops should not return as it execs */
 738         handler->ops->start(handler, handler->data);
 739
 740 out_warn_father:
 741         /* we want the parent to know something went wrong, so any
 742          * value other than what it expects is ok. */
 743         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 744         return -1;
 745 }
 746
 747 static int save_phys_nics(struct lxc_conf *conf)
 748 {
 749         struct lxc_list *iterator;
 750
 751         lxc_list_for_each(iterator, &conf->network) {
 752                 struct lxc_netdev *netdev = iterator->elem;
 753
 754                 if (netdev->type != LXC_NET_PHYS)
 755                         continue;
 756                 conf->saved_nics = realloc(conf->saved_nics,
 757                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 758                 if (!conf->saved_nics) {
 759                         SYSERROR("failed to allocate memory");
 760                         return -1;
 761                 }
 762                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 763                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 764                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 765                         SYSERROR("failed to allocate memory");
 766                         return -1;
 767                 }
 768                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 769                         conf->saved_nics[conf->num_savednics].ifindex,
 770                         conf->saved_nics[conf->num_savednics].orig_name);
 771                 conf->num_savednics++;
 772         }
 773
 774         return 0;
 775 }
 776
 777 static int lxc_spawn(struct lxc_handler *handler)
 778 {
 779         int failed_before_rename = 0;
 780         const char *name = handler->name;
 781         bool cgroups_connected = false;
 782         int saved_ns_fd[LXC_NS_MAX];
 783         int preserve_mask = 0, i;
 784         int netpipepair[2], nveths;
 785
 786         netpipe = -1;
 787
 788         for (i = 0; i < LXC_NS_MAX; i++)
 789                 if (handler->conf->inherit_ns_fd[i] != -1)
 790                         preserve_mask |= ns_info[i].clone_flag;
 791
 792         if (lxc_sync_init(handler))
 793                 return -1;
 794
 795         handler->clone_flags = CLONE_NEWPID|CLONE_NEWNS;
 796         if (!lxc_list_empty(&handler->conf->id_map)) {
 797                 INFO("Cloning a new user namespace");
 798                 handler->clone_flags |= CLONE_NEWUSER;
 799         }
 800
 801         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 802                 if (!lxc_requests_empty_network(handler))
 803                         handler->clone_flags |= CLONE_NEWNET;
 804
 805                 if (!lxc_list_empty(&handler->conf->network)) {
 806
 807                         /* Find gateway addresses from the link device, which is
 808                          * no longer accessible inside the container. Do this
 809                          * before creating network interfaces, since goto
 810                          * out_delete_net does not work before lxc_clone. */
 811                         if (lxc_find_gateway_addresses(handler)) {
 812                                 ERROR("failed to find gateway addresses");
 813                                 lxc_sync_fini(handler);
 814                                 return -1;
 815                         }
 816
 817                         /* that should be done before the clone because we will
 818                          * fill the netdev index and use them in the child
 819                          */
 820                         if (lxc_create_network(handler)) {
 821                                 ERROR("failed to create the network");
 822                                 lxc_sync_fini(handler);
 823                                 return -1;
 824                         }
 825                 }
 826
 827                 if (save_phys_nics(handler->conf)) {
 828                         ERROR("failed to save physical nic info");
 829                         goto out_abort;
 830                 }
 831         } else {
 832                 INFO("Inheriting a net namespace");
 833         }
 834
 835         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 836                 handler->clone_flags |= CLONE_NEWIPC;
 837         } else {
 838                 INFO("Inheriting an IPC namespace");
 839         }
 840
 841         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 842                 handler->clone_flags |= CLONE_NEWUTS;
 843         } else {
 844                 INFO("Inheriting a UTS namespace");
 845         }
 846
 847
 848         if (!cgroup_init(handler)) {
 849                 ERROR("failed initializing cgroup support");
 850                 goto out_delete_net;
 851         }
 852
 853         cgroups_connected = true;
 854
 855         if (!cgroup_create(handler)) {
 856                 ERROR("failed creating cgroups");
 857                 goto out_delete_net;
 858         }
 859
 860         /*
 861          * if the rootfs is not a blockdev, prevent the container from
 862          * marking it readonly.
 863          *
 864          * if the container is unprivileged then skip rootfs pinning
 865          */
 866         if (lxc_list_empty(&handler->conf->id_map)) {
 867                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 868                 if (handler->pinfd == -1)
 869                         INFO("failed to pin the container's rootfs");
 870         }
 871
 872         if (preserve_ns(saved_ns_fd, preserve_mask) < 0)
 873                 goto out_delete_net;
 874         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
 875                 goto out_delete_net;
 876
 877         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
 878                 if (pipe(netpipepair) < 0) {
 879                         SYSERROR("Error creating pipe");
 880                         goto out_delete_net;
 881                 }
 882                 /* store netpipe in the global var for do_start's use */
 883                 netpipe = netpipepair[0];
 884         }
 885
 886         /* Create a process in a new set of namespaces */
 887         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 888         if (handler->pid < 0) {
 889                 SYSERROR("failed to fork into a new namespace");
 890                 goto out_delete_net;
 891         }
 892
 893         if (attach_ns(saved_ns_fd))
 894                 WARN("failed to restore saved namespaces");
 895
 896         lxc_sync_fini_child(handler);
 897
 898         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 899                 failed_before_rename = 1;
 900
 901         if (!cgroup_create_legacy(handler)) {
 902                 ERROR("failed to setup the legacy cgroups for %s", name);
 903                 goto out_delete_net;
 904         }
 905         if (!cgroup_setup_limits(handler, false)) {
 906                 ERROR("failed to setup the cgroup limits for '%s'", name);
 907                 goto out_delete_net;
 908         }
 909
 910         if (!cgroup_enter(handler))
 911                 goto out_delete_net;
 912
 913         if (!cgroup_chown(handler))
 914                 goto out_delete_net;
 915
 916         if (failed_before_rename)
 917                 goto out_delete_net;
 918
 919         /* Create the network configuration */
 920         if (handler->clone_flags & CLONE_NEWNET) {
 921                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 922                         ERROR("failed to create the configured network");
 923                         goto out_delete_net;
 924                 }
 925         }
 926
 927         if (netpipe != -1) {
 928                 struct lxc_list *iterator;
 929                 struct lxc_netdev *netdev;
 930
 931                 close(netpipe);
 932                 lxc_list_for_each(iterator, &handler->conf->network) {
 933                         netdev = iterator->elem;
 934                         if (netdev->type != LXC_NET_VETH)
 935                                 continue;
 936                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 937                                 ERROR("Error writing veth name to container");
 938                                 goto out_delete_net;
 939                         }
 940                 }
 941                 close(netpipepair[1]);
 942         }
 943
 944         /* map the container uids - the container became an invalid
 945          * userid the moment it was cloned with CLONE_NEWUSER - this
 946          * call doesn't change anything immediately, but allows the
 947          * container to setuid(0) (0 being mapped to something else on
 948          * the host) later to become a valid uid again */
 949         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
 950                 ERROR("failed to set up id mapping");
 951                 goto out_delete_net;
 952         }
 953
 954         /* Tell the child to continue its initialization.  we'll get
 955          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
 956          */
 957         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 958                 goto out_delete_net;
 959
 960         if (!cgroup_setup_limits(handler, true)) {
 961                 ERROR("failed to setup the devices cgroup for '%s'", name);
 962                 goto out_delete_net;
 963         }
 964
 965         cgroup_disconnect();
 966         cgroups_connected = false;
 967
 968         /* Tell the child to complete its initialization and wait for
 969          * it to exec or return an error.  (the child will never
 970          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
 971          * sync pipe, causing lxc_sync_barrier_child to return
 972          * success, or return a different value, causing us to error
 973          * out).
 974          */
 975         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
 976                 return -1;
 977
 978         if (detect_shared_rootfs())
 979                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
 980
 981         if (handler->ops->post_start(handler, handler->data))
 982                 goto out_abort;
 983
 984         if (lxc_set_state(name, handler, RUNNING)) {
 985                 ERROR("failed to set state to %s",
 986                               lxc_state2str(RUNNING));
 987                 goto out_abort;
 988         }
 989
 990         lxc_sync_fini(handler);
 991
 992         return 0;
 993
 994 out_delete_net:
 995         if (cgroups_connected)
 996                 cgroup_disconnect();
 997         if (handler->clone_flags & CLONE_NEWNET)
 998                 lxc_delete_network(handler);
 999 out_abort:
1000         lxc_abort(name, handler);
1001         lxc_sync_fini(handler);
1002         if (handler->pinfd >= 0) {
1003                 close(handler->pinfd);
1004                 handler->pinfd = -1;
1005         }
1006
1007         return -1;
1008 }
1009
1010 int get_netns_fd(int pid)
1011 {
1012         char path[MAXPATHLEN];
1013         int ret, fd;
1014
1015         ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid);
1016         if (ret < 0 || ret >= MAXPATHLEN) {
1017                 WARN("Failed to pin netns file for pid %d", pid);
1018                 return -1;
1019         }
1020
1021         fd = open(path, O_RDONLY);
1022         if (fd < 0) {
1023                 WARN("Failed to pin netns file %s for pid %d: %s",
1024                                 path, pid, strerror(errno));
1025                 return -1;
1026         }
1027         return fd;
1028 }
1029
1030 int __lxc_start(const char *name, struct lxc_conf *conf,
1031                 struct lxc_operations* ops, void *data, const char *lxcpath)
1032 {
1033         struct lxc_handler *handler;
1034         int err = -1;
1035         int status;
1036         int netnsfd = -1;
1037
1038         handler = lxc_init(name, conf, lxcpath);
1039         if (!handler) {
1040                 ERROR("failed to initialize the container");
1041                 return -1;
1042         }
1043         handler->ops = ops;
1044         handler->data = data;
1045
1046         if (must_drop_cap_sys_boot(handler->conf)) {
1047                 #if HAVE_SYS_CAPABILITY_H
1048                 DEBUG("Dropping cap_sys_boot");
1049                 #else
1050                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1051                 #endif
1052         } else {
1053                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1054                 handler->conf->need_utmp_watch = 0;
1055         }
1056
1057         err = lxc_spawn(handler);
1058         if (err) {
1059                 ERROR("failed to spawn '%s'", name);
1060                 goto out_fini_nonet;
1061         }
1062
1063         netnsfd = get_netns_fd(handler->pid);
1064
1065         err = lxc_poll(name, handler);
1066         if (err) {
1067                 ERROR("mainloop exited with an error");
1068                 if (netnsfd >= 0)
1069                         close(netnsfd);
1070                 goto out_abort;
1071         }
1072
1073         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1074                 continue;
1075
1076         /*
1077          * If the child process exited but was not signaled,
1078          * it didn't call reboot.  This should mean it was an
1079          * lxc-execute which simply exited.  In any case, treat
1080          * it as a 'halt'
1081          */
1082         if (WIFSIGNALED(status)) {
1083                 switch(WTERMSIG(status)) {
1084                 case SIGINT: /* halt */
1085                         DEBUG("Container halting");
1086                         break;
1087                 case SIGHUP: /* reboot */
1088                         DEBUG("Container rebooting");
1089                         handler->conf->reboot = 1;
1090                         break;
1091                 case SIGSYS: /* seccomp */
1092                         DEBUG("Container violated its seccomp policy");
1093                         break;
1094                 default:
1095                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1096                         break;
1097                 }
1098         }
1099
1100         lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf);
1101         if (netnsfd >= 0)
1102                 close(netnsfd);
1103
1104         if (handler->pinfd >= 0) {
1105                 close(handler->pinfd);
1106                 handler->pinfd = -1;
1107         }
1108
1109         err =  lxc_error_set_and_log(handler->pid, status);
1110 out_fini:
1111         lxc_delete_network(handler);
1112
1113 out_fini_nonet:
1114         lxc_fini(name, handler);
1115         return err;
1116
1117 out_abort:
1118         lxc_abort(name, handler);
1119         goto out_fini;
1120 }
1121
1122 struct start_args {
1123         char *const *argv;
1124 };
1125
1126 static int start(struct lxc_handler *handler, void* data)
1127 {
1128         struct start_args *arg = data;
1129
1130         NOTICE("exec'ing '%s'", arg->argv[0]);
1131
1132         execvp(arg->argv[0], arg->argv);
1133         SYSERROR("failed to exec %s", arg->argv[0]);
1134         return 0;
1135 }
1136
1137 static int post_start(struct lxc_handler *handler, void* data)
1138 {
1139         struct start_args *arg = data;
1140
1141         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1142         return 0;
1143 }
1144
1145 static struct lxc_operations start_ops = {
1146         .start = start,
1147         .post_start = post_start
1148 };
1149
1150 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1151               const char *lxcpath)
1152 {
1153         struct start_args start_arg = {
1154                 .argv = argv,
1155         };
1156
1157         if (lxc_check_inherited(conf, -1))
1158                 return -1;
1159
1160         conf->need_utmp_watch = 1;
1161         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath);
1162 }