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 #include "lxclock.h"
  75
  76 lxc_log_define(lxc_start, lxc);
  77
  78 const struct ns_info ns_info[LXC_NS_MAX] = {
  79         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  80         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  81         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  82         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  83         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  84         [LXC_NS_NET] = {"net", CLONE_NEWNET}
  85 };
  86
  87 extern void mod_all_rdeps(struct lxc_container *c, bool inc);
  88 static bool do_destroy_container(struct lxc_conf *conf);
  89 static int lxc_rmdir_onedev_wrapper(void *data);
  90 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
  91                                             const char *name);
  92
  93 static void print_top_failing_dir(const char *path)
  94 {
  95         size_t len = strlen(path);
  96         char *copy = alloca(len+1), *p, *e, saved;
  97         strcpy(copy, path);
  98
  99         p = copy;
 100         e = copy + len;
 101         while (p < e) {
 102                 while (p < e && *p == '/') p++;
 103                 while (p < e && *p != '/') p++;
 104                 saved = *p;
 105                 *p = '\0';
 106                 if (access(copy, X_OK)) {
 107                         SYSERROR("could not access %s.  Please grant it 'x' " \
 108                               "access, or add an ACL for the container root.",
 109                               copy);
 110                         return;
 111                 }
 112                 *p = saved;
 113         }
 114 }
 115
 116 static void close_ns(int ns_fd[LXC_NS_MAX]) {
 117         int i;
 118
 119         for (i = 0; i < LXC_NS_MAX; i++) {
 120                 if (ns_fd[i] > -1) {
 121                         close(ns_fd[i]);
 122                         ns_fd[i] = -1;
 123                 }
 124         }
 125 }
 126
 127 static int preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid) {
 128         int i, saved_errno;
 129         char path[MAXPATHLEN];
 130
 131         for (i = 0; i < LXC_NS_MAX; i++)
 132                 ns_fd[i] = -1;
 133
 134         snprintf(path, MAXPATHLEN, "/proc/%d/ns", pid);
 135         if (access(path, X_OK)) {
 136                 WARN("Kernel does not support attach; preserve_ns ignored");
 137                 return 0;
 138         }
 139
 140         for (i = 0; i < LXC_NS_MAX; i++) {
 141                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 142                         continue;
 143                 snprintf(path, MAXPATHLEN, "/proc/%d/ns/%s", pid,
 144                          ns_info[i].proc_name);
 145                 ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC);
 146                 if (ns_fd[i] < 0)
 147                         goto error;
 148         }
 149
 150         return 0;
 151
 152 error:
 153         saved_errno = errno;
 154         close_ns(ns_fd);
 155         errno = saved_errno;
 156         SYSERROR("failed to open '%s'", path);
 157         return -1;
 158 }
 159
 160 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 161         int i;
 162
 163         for (i = 0; i < LXC_NS_MAX; i++) {
 164                 if (ns_fd[i] < 0)
 165                         continue;
 166
 167                 if (setns(ns_fd[i], 0) != 0)
 168                         goto error;
 169         }
 170         return 0;
 171
 172 error:
 173         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 174         return -1;
 175 }
 176
 177 static int match_fd(int fd)
 178 {
 179         return (fd == 0 || fd == 1 || fd == 2);
 180 }
 181
 182 /*
 183  * Check for any fds we need to close
 184  * * if fd_to_ignore != -1, then if we find that fd open we will ignore it.
 185  * * By default we warn about open fds we find.
 186  * * If closeall is true, we will close open fds.
 187  * * If lxc-start was passed "-C", then conf->close_all_fds will be true,
 188  *     in which case we also close all open fds.
 189  * * A daemonized container will always pass closeall=true.
 190  */
 191 int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore)
 192 {
 193         struct dirent dirent, *direntp;
 194         int fd, fddir;
 195         DIR *dir;
 196
 197         if (conf && conf->close_all_fds)
 198                 closeall = true;
 199
 200 restart:
 201         dir = opendir("/proc/self/fd");
 202         if (!dir) {
 203                 WARN("failed to open directory: %m");
 204                 return -1;
 205         }
 206
 207         fddir = dirfd(dir);
 208
 209         while (!readdir_r(dir, &dirent, &direntp)) {
 210                 if (!direntp)
 211                         break;
 212
 213                 if (!strcmp(direntp->d_name, "."))
 214                         continue;
 215
 216                 if (!strcmp(direntp->d_name, ".."))
 217                         continue;
 218
 219                 fd = atoi(direntp->d_name);
 220
 221                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 222                         continue;
 223
 224                 if (current_config && fd == current_config->logfd)
 225                         continue;
 226
 227                 if (match_fd(fd))
 228                         continue;
 229
 230                 if (closeall) {
 231                         close(fd);
 232                         closedir(dir);
 233                         INFO("closed inherited fd %d", fd);
 234                         goto restart;
 235                 }
 236                 WARN("inherited fd %d", fd);
 237         }
 238
 239         closedir(dir); /* cannot fail */
 240         return 0;
 241 }
 242
 243 static int setup_signal_fd(sigset_t *oldmask)
 244 {
 245         sigset_t mask;
 246         int fd;
 247
 248         /* Block everything except serious error signals */
 249         if (sigfillset(&mask) ||
 250             sigdelset(&mask, SIGILL) ||
 251             sigdelset(&mask, SIGSEGV) ||
 252             sigdelset(&mask, SIGBUS) ||
 253             sigdelset(&mask, SIGWINCH) ||
 254             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 255                 SYSERROR("failed to set signal mask");
 256                 return -1;
 257         }
 258
 259         fd = signalfd(-1, &mask, 0);
 260         if (fd < 0) {
 261                 SYSERROR("failed to create the signal fd");
 262                 return -1;
 263         }
 264
 265         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 266                 SYSERROR("failed to set sigfd to close-on-exec");
 267                 close(fd);
 268                 return -1;
 269         }
 270
 271         DEBUG("sigchild handler set");
 272
 273         return fd;
 274 }
 275
 276 static int signal_handler(int fd, uint32_t events, void *data,
 277                            struct lxc_epoll_descr *descr)
 278 {
 279         struct signalfd_siginfo siginfo;
 280         siginfo_t info;
 281         int ret;
 282         pid_t *pid = data;
 283         bool init_died = false;
 284
 285         ret = read(fd, &siginfo, sizeof(siginfo));
 286         if (ret < 0) {
 287                 ERROR("failed to read signal info");
 288                 return -1;
 289         }
 290
 291         if (ret != sizeof(siginfo)) {
 292                 ERROR("unexpected siginfo size");
 293                 return -1;
 294         }
 295
 296         // check whether init is running
 297         info.si_pid = 0;
 298         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 299         if (ret == 0 && info.si_pid == *pid) {
 300                 init_died = true;
 301         }
 302
 303         if (siginfo.ssi_signo != SIGCHLD) {
 304                 kill(*pid, siginfo.ssi_signo);
 305                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 306                 return init_died ? 1 : 0;
 307         }
 308
 309         if (siginfo.ssi_code == CLD_STOPPED ||
 310             siginfo.ssi_code == CLD_CONTINUED) {
 311                 INFO("container init process was stopped/continued");
 312                 return init_died ? 1 : 0;
 313         }
 314
 315         /* more robustness, protect ourself from a SIGCHLD sent
 316          * by a process different from the container init
 317          */
 318         if (siginfo.ssi_pid != *pid) {
 319                 WARN("invalid pid for SIGCHLD");
 320                 return init_died ? 1 : 0;
 321         }
 322
 323         DEBUG("container init process exited");
 324         return 1;
 325 }
 326
 327 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 328 {
 329         handler->state = state;
 330         lxc_monitor_send_state(name, state, handler->lxcpath);
 331         return 0;
 332 }
 333
 334 int lxc_poll(const char *name, struct lxc_handler *handler)
 335 {
 336         int sigfd = handler->sigfd;
 337         int pid = handler->pid;
 338         struct lxc_epoll_descr descr;
 339
 340         if (lxc_mainloop_open(&descr)) {
 341                 ERROR("failed to create mainloop");
 342                 goto out_sigfd;
 343         }
 344
 345         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 346                 ERROR("failed to add handler for the signal");
 347                 goto out_mainloop_open;
 348         }
 349
 350         if (lxc_console_mainloop_add(&descr, handler)) {
 351                 ERROR("failed to add console handler to mainloop");
 352                 goto out_mainloop_open;
 353         }
 354
 355         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 356                 ERROR("failed to add command handler to mainloop");
 357                 goto out_mainloop_open;
 358         }
 359
 360         if (handler->conf->need_utmp_watch) {
 361                 #if HAVE_SYS_CAPABILITY_H
 362                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 363                         ERROR("failed to add utmp handler to mainloop");
 364                         goto out_mainloop_open;
 365                 }
 366                 #else
 367                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
 368                 #endif
 369         }
 370
 371         return lxc_mainloop(&descr, -1);
 372
 373 out_mainloop_open:
 374         lxc_mainloop_close(&descr);
 375 out_sigfd:
 376         close(sigfd);
 377         return -1;
 378 }
 379
 380 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 381 {
 382         struct lxc_handler *handler;
 383
 384         handler = malloc(sizeof(*handler));
 385         if (!handler)
 386                 return NULL;
 387
 388         memset(handler, 0, sizeof(*handler));
 389
 390         handler->ttysock[0] = handler->ttysock[1] = -1;
 391         handler->conf = conf;
 392         handler->lxcpath = lxcpath;
 393         handler->pinfd = -1;
 394
 395         lsm_init();
 396
 397         handler->name = strdup(name);
 398         if (!handler->name) {
 399                 ERROR("failed to allocate memory");
 400                 goto out_free;
 401         }
 402
 403         if (lxc_cmd_init(name, handler, lxcpath))
 404                 goto out_free_name;
 405
 406         if (lxc_read_seccomp_config(conf) != 0) {
 407                 ERROR("failed loading seccomp policy");
 408                 goto out_close_maincmd_fd;
 409         }
 410
 411         /* Begin by setting the state to STARTING */
 412         if (lxc_set_state(name, handler, STARTING)) {
 413                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 414                 goto out_close_maincmd_fd;
 415         }
 416
 417         /* Start of environment variable setup for hooks */
 418         if (name && setenv("LXC_NAME", name, 1)) {
 419                 SYSERROR("failed to set environment variable for container name");
 420         }
 421         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 422                 SYSERROR("failed to set environment variable for config path");
 423         }
 424         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 425                 SYSERROR("failed to set environment variable for rootfs mount");
 426         }
 427         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 428                 SYSERROR("failed to set environment variable for rootfs mount");
 429         }
 430         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 431                 SYSERROR("failed to set environment variable for console path");
 432         }
 433         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 434                 SYSERROR("failed to set environment variable for console log");
 435         }
 436         /* End of environment variable setup for hooks */
 437
 438         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 439                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 440                 goto out_aborting;
 441         }
 442
 443         /* the signal fd has to be created before forking otherwise
 444          * if the child process exits before we setup the signal fd,
 445          * the event will be lost and the command will be stuck */
 446         handler->sigfd = setup_signal_fd(&handler->oldmask);
 447         if (handler->sigfd < 0) {
 448                 ERROR("failed to set sigchild fd handler");
 449                 goto out_delete_tty;
 450         }
 451
 452         /* do this after setting up signals since it might unblock SIGWINCH */
 453         if (lxc_console_create(conf)) {
 454                 ERROR("failed to create console");
 455                 goto out_restore_sigmask;
 456         }
 457
 458         if (ttys_shift_ids(conf) < 0) {
 459                 ERROR("Failed to shift tty into container");
 460                 goto out_restore_sigmask;
 461         }
 462
 463         INFO("'%s' is initialized", name);
 464         return handler;
 465
 466 out_restore_sigmask:
 467         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 468 out_delete_tty:
 469         lxc_delete_tty(&conf->tty_info);
 470 out_aborting:
 471         lxc_set_state(name, handler, ABORTING);
 472 out_close_maincmd_fd:
 473         close(conf->maincmd_fd);
 474         conf->maincmd_fd = -1;
 475 out_free_name:
 476         free(handler->name);
 477         handler->name = NULL;
 478 out_free:
 479         free(handler);
 480         return NULL;
 481 }
 482
 483 void lxc_fini(const char *name, struct lxc_handler *handler)
 484 {
 485         /* The STOPPING state is there for future cleanup code
 486          * which can take awhile
 487          */
 488         lxc_set_state(name, handler, STOPPING);
 489         lxc_set_state(name, handler, STOPPED);
 490
 491         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 492                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 493
 494         /* reset mask set by setup_signal_fd */
 495         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 496                 WARN("failed to restore sigprocmask");
 497
 498         lxc_console_delete(&handler->conf->console);
 499         lxc_delete_tty(&handler->conf->tty_info);
 500         close(handler->conf->maincmd_fd);
 501         handler->conf->maincmd_fd = -1;
 502         free(handler->name);
 503         if (handler->ttysock[0] != -1) {
 504                 close(handler->ttysock[0]);
 505                 close(handler->ttysock[1]);
 506         }
 507         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) {
 508                 lxc_destroy_container_on_signal(handler, name);
 509         }
 510         cgroup_destroy(handler);
 511         free(handler);
 512 }
 513
 514 void lxc_abort(const char *name, struct lxc_handler *handler)
 515 {
 516         int ret, status;
 517
 518         lxc_set_state(name, handler, ABORTING);
 519         if (handler->pid > 0)
 520                 kill(handler->pid, SIGKILL);
 521         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 522 }
 523
 524 #include <sys/reboot.h>
 525 #include <linux/reboot.h>
 526
 527 /*
 528  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 529  * in a child pid namespace if container reboot support exists.
 530  * Otherwise, it will either succeed or return -EPERM.
 531  */
 532 static int container_reboot_supported(void *arg)
 533 {
 534         int *cmd = arg;
 535         int ret;
 536
 537         ret = reboot(*cmd);
 538         if (ret == -1 && errno == EINVAL)
 539                 return 1;
 540         return 0;
 541 }
 542
 543 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 544 {
 545         FILE *f;
 546         int ret, cmd, v, flags;
 547         long stack_size = 4096;
 548         void *stack = alloca(stack_size);
 549         int status;
 550         pid_t pid;
 551
 552         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 553         if (!f) {
 554                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 555                 return 1;
 556         }
 557
 558         ret = fscanf(f, "%d", &v);
 559         fclose(f);
 560         if (ret != 1) {
 561                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 562                 return 1;
 563         }
 564         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 565
 566         flags = CLONE_NEWPID | SIGCHLD;
 567         if (!lxc_list_empty(&conf->id_map))
 568                 flags |= CLONE_NEWUSER;
 569
 570 #ifdef __ia64__
 571         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 572 #else
 573         stack += stack_size;
 574         pid = clone(container_reboot_supported, stack, flags, &cmd);
 575 #endif
 576         if (pid < 0) {
 577                 if (flags & CLONE_NEWUSER)
 578                         ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno));
 579                 else
 580                         ERROR("failed to clone (%#x): %s", flags, strerror(errno));
 581                 return -1;
 582         }
 583         if (wait(&status) < 0) {
 584                 SYSERROR("unexpected wait error: %m");
 585                 return -1;
 586         }
 587
 588         if (WEXITSTATUS(status) != 1)
 589                 return 1;
 590
 591         return 0;
 592 }
 593
 594 /*
 595  * netpipe is used in the unprivileged case to transfer the ifindexes
 596  * from parent to child
 597  */
 598 static int netpipe = -1;
 599
 600 static inline int count_veths(struct lxc_list *network)
 601 {
 602         struct lxc_list *iterator;
 603         struct lxc_netdev *netdev;
 604         int count = 0;
 605
 606         lxc_list_for_each(iterator, network) {
 607                 netdev = iterator->elem;
 608                 if (netdev->type != LXC_NET_VETH)
 609                         continue;
 610                 count++;
 611         }
 612         return count;
 613 }
 614
 615 static int read_unpriv_netifindex(struct lxc_list *network)
 616 {
 617         struct lxc_list *iterator;
 618         struct lxc_netdev *netdev;
 619
 620         if (netpipe == -1)
 621                 return 0;
 622         lxc_list_for_each(iterator, network) {
 623                 netdev = iterator->elem;
 624                 if (netdev->type != LXC_NET_VETH)
 625                         continue;
 626                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 627                         ERROR("Out of memory");
 628                         close(netpipe);
 629                         return -1;
 630                 }
 631                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 632                         close(netpipe);
 633                         return -1;
 634                 }
 635         }
 636         close(netpipe);
 637         return 0;
 638 }
 639
 640 static int do_start(void *data)
 641 {
 642         struct lxc_list *iterator;
 643         struct lxc_handler *handler = data;
 644
 645         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 646                 SYSERROR("failed to set sigprocmask");
 647                 return -1;
 648         }
 649
 650         /* This prctl must be before the synchro, so if the parent
 651          * dies before we set the parent death signal, we will detect
 652          * its death with the synchro right after, otherwise we have
 653          * a window where the parent can exit before we set the pdeath
 654          * signal leading to a unsupervized container.
 655          */
 656         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 657                 SYSERROR("failed to set pdeath signal");
 658                 return -1;
 659         }
 660
 661         lxc_sync_fini_parent(handler);
 662
 663         /* don't leak the pinfd to the container */
 664         if (handler->pinfd >= 0) {
 665                 close(handler->pinfd);
 666         }
 667
 668         /* Tell the parent task it can begin to configure the
 669          * container and wait for it to finish
 670          */
 671         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 672                 return -1;
 673
 674         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 675                 goto out_warn_father;
 676
 677         /*
 678          * if we are in a new user namespace, become root there to have
 679          * privilege over our namespace. When using lxc-execute we default to root,
 680          * but this can be overriden using the lxc.init_uid and lxc.init_gid
 681          * configuration options.
 682          */
 683         if (!lxc_list_empty(&handler->conf->id_map)) {
 684                 gid_t new_gid = 0;
 685                 if (handler->conf->is_execute && handler->conf->init_gid)
 686                         new_gid = handler->conf->init_gid;
 687
 688                 uid_t new_uid = 0;
 689                 if (handler->conf->is_execute && handler->conf->init_uid)
 690                         new_uid = handler->conf->init_uid;
 691
 692                 NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid);
 693                 if (setgid(new_gid)) {
 694                         SYSERROR("setgid");
 695                         goto out_warn_father;
 696                 }
 697                 if (setuid(new_uid)) {
 698                         SYSERROR("setuid");
 699                         goto out_warn_father;
 700                 }
 701                 if (setgroups(0, NULL)) {
 702                         SYSERROR("setgroups");
 703                         goto out_warn_father;
 704                 }
 705         }
 706
 707         if (access(handler->lxcpath, X_OK)) {
 708                 print_top_failing_dir(handler->lxcpath);
 709                 goto out_warn_father;
 710         }
 711
 712         #if HAVE_SYS_CAPABILITY_H
 713         if (handler->conf->need_utmp_watch) {
 714                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 715                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 716                         goto out_warn_father;
 717                 }
 718                 DEBUG("Dropped cap_sys_boot");
 719         }
 720         #endif
 721
 722         /* Setup the container, ip, names, utsname, ... */
 723         if (lxc_setup(handler)) {
 724                 ERROR("failed to setup the container");
 725                 goto out_warn_father;
 726         }
 727
 728         /* ask father to setup cgroups and wait for him to finish */
 729         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 730                 return -1;
 731
 732         /* Set the label to change to when we exec(2) the container's init */
 733         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 734                 goto out_warn_father;
 735
 736         /* Some init's such as busybox will set sane tty settings on stdin,
 737          * stdout, stderr which it thinks is the console. We already set them
 738          * the way we wanted on the real terminal, and we want init to do its
 739          * setup on its console ie. the pty allocated in lxc_console_create()
 740          * so make sure that that pty is stdin,stdout,stderr.
 741          */
 742         if (lxc_console_set_stdfds(handler) < 0)
 743                 goto out_warn_father;
 744
 745         /* If we mounted a temporary proc, then unmount it now */
 746         tmp_proc_unmount(handler->conf);
 747
 748         if (lxc_seccomp_load(handler->conf) != 0)
 749                 goto out_warn_father;
 750
 751         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 752                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 753                 goto out_warn_father;
 754         }
 755
 756         /* The clearenv() and putenv() calls have been moved here
 757          * to allow us to use environment variables passed to the various
 758          * hooks, such as the start hook above.  Not all of the
 759          * variables like CONFIG_PATH or ROOTFS are valid in this
 760          * context but others are. */
 761         if (clearenv()) {
 762                 SYSERROR("failed to clear environment");
 763                 /* don't error out though */
 764         }
 765
 766         lxc_list_for_each(iterator, &handler->conf->environment) {
 767                 if (putenv((char *)iterator->elem)) {
 768                         SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem);
 769                         goto out_warn_father;
 770                 }
 771         }
 772
 773         if (putenv("container=lxc")) {
 774                 SYSERROR("failed to set environment variable 'container=lxc'");
 775                 goto out_warn_father;
 776         }
 777
 778         if (handler->conf->pty_names) {
 779                 if (putenv(handler->conf->pty_names)) {
 780                         SYSERROR("failed to set environment variable for container ptys");
 781                         goto out_warn_father;
 782                 }
 783         }
 784
 785         close(handler->sigfd);
 786
 787         if (handler->backgrounded && null_stdfds() < 0)
 788                 goto out_warn_father;
 789
 790         /* after this call, we are in error because this
 791          * ops should not return as it execs */
 792         handler->ops->start(handler, handler->data);
 793
 794 out_warn_father:
 795         /* we want the parent to know something went wrong, so any
 796          * value other than what it expects is ok. */
 797         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 798         return -1;
 799 }
 800
 801 static int save_phys_nics(struct lxc_conf *conf)
 802 {
 803         struct lxc_list *iterator;
 804
 805         lxc_list_for_each(iterator, &conf->network) {
 806                 struct lxc_netdev *netdev = iterator->elem;
 807
 808                 if (netdev->type != LXC_NET_PHYS)
 809                         continue;
 810                 conf->saved_nics = realloc(conf->saved_nics,
 811                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 812                 if (!conf->saved_nics) {
 813                         SYSERROR("failed to allocate memory");
 814                         return -1;
 815                 }
 816                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 817                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 818                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 819                         SYSERROR("failed to allocate memory");
 820                         return -1;
 821                 }
 822                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 823                         conf->saved_nics[conf->num_savednics].ifindex,
 824                         conf->saved_nics[conf->num_savednics].orig_name);
 825                 conf->num_savednics++;
 826         }
 827
 828         return 0;
 829 }
 830
 831 static int recv_fd(int sock, int *fd)
 832 {
 833         if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) {
 834                 SYSERROR("Error receiving tty fd from child");
 835                 return -1;
 836         }
 837         if (*fd == -1)
 838                 return -1;
 839         return 0;
 840 }
 841
 842 static int recv_ttys_from_child(struct lxc_handler *handler)
 843 {
 844         struct lxc_conf *conf = handler->conf;
 845         int i, sock = handler->ttysock[1];
 846         struct lxc_tty_info *tty_info = &conf->tty_info;
 847
 848         if (!conf->tty)
 849                 return 0;
 850
 851         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty);
 852         if (!tty_info->pty_info) {
 853                 SYSERROR("failed to allocate pty_info");
 854                 return -1;
 855         }
 856
 857         for (i = 0; i < conf->tty; i++) {
 858                 struct lxc_pty_info *pty_info = &tty_info->pty_info[i];
 859                 pty_info->busy = 0;
 860                 if (recv_fd(sock, &pty_info->slave) < 0 ||
 861                                 recv_fd(sock, &pty_info->master) < 0) {
 862                         ERROR("Error receiving tty info from child");
 863                         return -1;
 864                 }
 865         }
 866         tty_info->nbtty = conf->tty;
 867
 868         return 0;
 869 }
 870
 871 void resolve_clone_flags(struct lxc_handler *handler)
 872 {
 873         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
 874
 875         if (!lxc_list_empty(&handler->conf->id_map)) {
 876                 INFO("Cloning a new user namespace");
 877                 handler->clone_flags |= CLONE_NEWUSER;
 878         }
 879
 880         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 881                 if (!lxc_requests_empty_network(handler))
 882                         handler->clone_flags |= CLONE_NEWNET;
 883         } else {
 884                 INFO("Inheriting a net namespace");
 885         }
 886
 887         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 888                 handler->clone_flags |= CLONE_NEWIPC;
 889         } else {
 890                 INFO("Inheriting an IPC namespace");
 891         }
 892
 893         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 894                 handler->clone_flags |= CLONE_NEWUTS;
 895         } else {
 896                 INFO("Inheriting a UTS namespace");
 897         }
 898 }
 899
 900 static int lxc_spawn(struct lxc_handler *handler)
 901 {
 902         int failed_before_rename = 0;
 903         const char *name = handler->name;
 904         bool cgroups_connected = false;
 905         int saved_ns_fd[LXC_NS_MAX];
 906         int preserve_mask = 0, i;
 907         int netpipepair[2], nveths;
 908
 909         netpipe = -1;
 910
 911         for (i = 0; i < LXC_NS_MAX; i++)
 912                 if (handler->conf->inherit_ns_fd[i] != -1)
 913                         preserve_mask |= ns_info[i].clone_flag;
 914
 915         if (lxc_sync_init(handler))
 916                 return -1;
 917
 918         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
 919                 lxc_sync_fini(handler);
 920                 return -1;
 921         }
 922
 923         resolve_clone_flags(handler);
 924
 925         if (handler->clone_flags & CLONE_NEWNET) {
 926                 if (!lxc_list_empty(&handler->conf->network)) {
 927
 928                         /* Find gateway addresses from the link device, which is
 929                          * no longer accessible inside the container. Do this
 930                          * before creating network interfaces, since goto
 931                          * out_delete_net does not work before lxc_clone. */
 932                         if (lxc_find_gateway_addresses(handler)) {
 933                                 ERROR("failed to find gateway addresses");
 934                                 lxc_sync_fini(handler);
 935                                 return -1;
 936                         }
 937
 938                         /* that should be done before the clone because we will
 939                          * fill the netdev index and use them in the child
 940                          */
 941                         if (lxc_create_network(handler)) {
 942                                 ERROR("failed to create the network");
 943                                 lxc_sync_fini(handler);
 944                                 return -1;
 945                         }
 946                 }
 947
 948                 if (save_phys_nics(handler->conf)) {
 949                         ERROR("failed to save physical nic info");
 950                         goto out_abort;
 951                 }
 952         }
 953
 954         if (!cgroup_init(handler)) {
 955                 ERROR("failed initializing cgroup support");
 956                 goto out_delete_net;
 957         }
 958
 959         cgroups_connected = true;
 960
 961         if (!cgroup_create(handler)) {
 962                 ERROR("failed creating cgroups");
 963                 goto out_delete_net;
 964         }
 965
 966         /*
 967          * if the rootfs is not a blockdev, prevent the container from
 968          * marking it readonly.
 969          *
 970          * if the container is unprivileged then skip rootfs pinning
 971          */
 972         if (lxc_list_empty(&handler->conf->id_map)) {
 973                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 974                 if (handler->pinfd == -1)
 975                         INFO("failed to pin the container's rootfs");
 976         }
 977
 978         if (preserve_ns(saved_ns_fd, preserve_mask, getpid()) < 0)
 979                 goto out_delete_net;
 980         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
 981                 goto out_delete_net;
 982
 983         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
 984                 if (pipe(netpipepair) < 0) {
 985                         SYSERROR("Error creating pipe");
 986                         goto out_delete_net;
 987                 }
 988                 /* store netpipe in the global var for do_start's use */
 989                 netpipe = netpipepair[0];
 990         }
 991
 992         /* Create a process in a new set of namespaces */
 993         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 994         if (handler->pid < 0) {
 995                 SYSERROR("failed to fork into a new namespace");
 996                 goto out_delete_net;
 997         }
 998
 999         if (attach_ns(saved_ns_fd))
1000                 WARN("failed to restore saved namespaces");
1001
1002         lxc_sync_fini_child(handler);
1003
1004         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
1005                 failed_before_rename = 1;
1006
1007         if (!cgroup_create_legacy(handler)) {
1008                 ERROR("failed to setup the legacy cgroups for %s", name);
1009                 goto out_delete_net;
1010         }
1011         if (!cgroup_setup_limits(handler, false)) {
1012                 ERROR("failed to setup the cgroup limits for '%s'", name);
1013                 goto out_delete_net;
1014         }
1015
1016         if (!cgroup_enter(handler))
1017                 goto out_delete_net;
1018
1019         if (!cgroup_chown(handler))
1020                 goto out_delete_net;
1021
1022         if (failed_before_rename)
1023                 goto out_delete_net;
1024
1025         /* Create the network configuration */
1026         if (handler->clone_flags & CLONE_NEWNET) {
1027                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
1028                         ERROR("failed to create the configured network");
1029                         goto out_delete_net;
1030                 }
1031         }
1032
1033         if (netpipe != -1) {
1034                 struct lxc_list *iterator;
1035                 struct lxc_netdev *netdev;
1036
1037                 close(netpipe);
1038                 lxc_list_for_each(iterator, &handler->conf->network) {
1039                         netdev = iterator->elem;
1040                         if (netdev->type != LXC_NET_VETH)
1041                                 continue;
1042                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1043                                 ERROR("Error writing veth name to container");
1044                                 goto out_delete_net;
1045                         }
1046                 }
1047                 close(netpipepair[1]);
1048         }
1049
1050         /* map the container uids - the container became an invalid
1051          * userid the moment it was cloned with CLONE_NEWUSER - this
1052          * call doesn't change anything immediately, but allows the
1053          * container to setuid(0) (0 being mapped to something else on
1054          * the host) later to become a valid uid again */
1055         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
1056                 ERROR("failed to set up id mapping");
1057                 goto out_delete_net;
1058         }
1059
1060         /* Tell the child to continue its initialization.  we'll get
1061          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1062          */
1063         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1064                 goto out_delete_net;
1065
1066         if (!cgroup_setup_limits(handler, true)) {
1067                 ERROR("failed to setup the devices cgroup for '%s'", name);
1068                 goto out_delete_net;
1069         }
1070
1071         cgroup_disconnect();
1072         cgroups_connected = false;
1073
1074         /* read tty fds allocated by child */
1075         if (recv_ttys_from_child(handler) < 0) {
1076                 ERROR("failed to receive tty info from child");
1077                 goto out_delete_net;
1078         }
1079
1080         /* Tell the child to complete its initialization and wait for
1081          * it to exec or return an error.  (the child will never
1082          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
1083          * sync pipe, causing lxc_sync_barrier_child to return
1084          * success, or return a different value, causing us to error
1085          * out).
1086          */
1087         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1088                 return -1;
1089
1090         if (detect_shared_rootfs())
1091                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
1092
1093         if (handler->ops->post_start(handler, handler->data))
1094                 goto out_abort;
1095
1096         if (lxc_set_state(name, handler, RUNNING)) {
1097                 ERROR("failed to set state to %s",
1098                               lxc_state2str(RUNNING));
1099                 goto out_abort;
1100         }
1101
1102         lxc_sync_fini(handler);
1103
1104         return 0;
1105
1106 out_delete_net:
1107         if (cgroups_connected)
1108                 cgroup_disconnect();
1109         if (handler->clone_flags & CLONE_NEWNET)
1110                 lxc_delete_network(handler);
1111 out_abort:
1112         lxc_abort(name, handler);
1113         lxc_sync_fini(handler);
1114         if (handler->pinfd >= 0) {
1115                 close(handler->pinfd);
1116                 handler->pinfd = -1;
1117         }
1118
1119         return -1;
1120 }
1121
1122 int get_netns_fd(int pid)
1123 {
1124         char path[MAXPATHLEN];
1125         int ret, fd;
1126
1127         ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid);
1128         if (ret < 0 || ret >= MAXPATHLEN) {
1129                 WARN("Failed to pin netns file for pid %d", pid);
1130                 return -1;
1131         }
1132
1133         fd = open(path, O_RDONLY);
1134         if (fd < 0) {
1135                 WARN("Failed to pin netns file %s for pid %d: %s",
1136                                 path, pid, strerror(errno));
1137                 return -1;
1138         }
1139         return fd;
1140 }
1141
1142 int __lxc_start(const char *name, struct lxc_conf *conf,
1143                 struct lxc_operations* ops, void *data, const char *lxcpath,
1144                 bool backgrounded)
1145 {
1146         struct lxc_handler *handler;
1147         int err = -1;
1148         int status;
1149         int netnsfd = -1;
1150
1151         handler = lxc_init(name, conf, lxcpath);
1152         if (!handler) {
1153                 ERROR("failed to initialize the container");
1154                 return -1;
1155         }
1156         handler->ops = ops;
1157         handler->data = data;
1158         handler->backgrounded = backgrounded;
1159
1160         if (must_drop_cap_sys_boot(handler->conf)) {
1161                 #if HAVE_SYS_CAPABILITY_H
1162                 DEBUG("Dropping cap_sys_boot");
1163                 #else
1164                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1165                 #endif
1166         } else {
1167                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1168                 handler->conf->need_utmp_watch = 0;
1169         }
1170
1171         if (!attach_block_device(handler->conf)) {
1172                 ERROR("Failure attaching block device");
1173                 goto out_fini_nonet;
1174         }
1175
1176         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1177                 /* if the backing store is a device, mount it here and now */
1178                 if (rootfs_is_blockdev(conf)) {
1179                         if (unshare(CLONE_NEWNS) < 0) {
1180                                 ERROR("Error unsharing mounts");
1181                                 goto out_fini_nonet;
1182                         }
1183                         remount_all_slave();
1184                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1185                                 ERROR("Error setting up rootfs mount as root before spawn");
1186                                 goto out_fini_nonet;
1187                         }
1188                         INFO("Set up container rootfs as host root");
1189                 }
1190         }
1191
1192         err = lxc_spawn(handler);
1193         if (err) {
1194                 ERROR("failed to spawn '%s'", name);
1195                 goto out_detach_blockdev;
1196         }
1197
1198         handler->conf->reboot = 0;
1199
1200         netnsfd = get_netns_fd(handler->pid);
1201
1202         err = lxc_poll(name, handler);
1203         if (err) {
1204                 ERROR("mainloop exited with an error");
1205                 if (netnsfd >= 0)
1206                         close(netnsfd);
1207                 goto out_abort;
1208         }
1209
1210         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1211                 continue;
1212
1213         /*
1214          * If the child process exited but was not signaled,
1215          * it didn't call reboot.  This should mean it was an
1216          * lxc-execute which simply exited.  In any case, treat
1217          * it as a 'halt'
1218          */
1219         if (WIFSIGNALED(status)) {
1220                 switch(WTERMSIG(status)) {
1221                 case SIGINT: /* halt */
1222                         DEBUG("Container halting");
1223                         break;
1224                 case SIGHUP: /* reboot */
1225                         DEBUG("Container rebooting");
1226                         handler->conf->reboot = 1;
1227                         break;
1228                 case SIGSYS: /* seccomp */
1229                         DEBUG("Container violated its seccomp policy");
1230                         break;
1231                 default:
1232                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1233                         break;
1234                 }
1235         }
1236
1237         DEBUG("Pushing physical nics back to host namespace");
1238         lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf);
1239
1240         DEBUG("Tearing down virtual network devices used by container");
1241         lxc_delete_network(handler);
1242
1243         if (netnsfd >= 0)
1244                 close(netnsfd);
1245
1246         if (handler->pinfd >= 0) {
1247                 close(handler->pinfd);
1248                 handler->pinfd = -1;
1249         }
1250
1251         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1252         err =  lxc_error_set_and_log(handler->pid, status);
1253 out_fini:
1254         lxc_delete_network(handler);
1255
1256 out_detach_blockdev:
1257         detach_block_device(handler->conf);
1258
1259 out_fini_nonet:
1260         lxc_fini(name, handler);
1261         return err;
1262
1263 out_abort:
1264         lxc_abort(name, handler);
1265         goto out_fini;
1266 }
1267
1268 struct start_args {
1269         char *const *argv;
1270 };
1271
1272 static int start(struct lxc_handler *handler, void* data)
1273 {
1274         struct start_args *arg = data;
1275
1276         NOTICE("exec'ing '%s'", arg->argv[0]);
1277
1278         execvp(arg->argv[0], arg->argv);
1279         SYSERROR("failed to exec %s", arg->argv[0]);
1280         return 0;
1281 }
1282
1283 static int post_start(struct lxc_handler *handler, void* data)
1284 {
1285         struct start_args *arg = data;
1286
1287         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1288         return 0;
1289 }
1290
1291 static struct lxc_operations start_ops = {
1292         .start = start,
1293         .post_start = post_start
1294 };
1295
1296 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1297               const char *lxcpath, bool backgrounded)
1298 {
1299         struct start_args start_arg = {
1300                 .argv = argv,
1301         };
1302
1303         conf->need_utmp_watch = 1;
1304         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded);
1305 }
1306
1307 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1308                                             const char *name)
1309 {
1310         char destroy[MAXPATHLEN];
1311         bool bret = true;
1312         int ret = 0;
1313         struct lxc_container *c;
1314         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1315                 bret = do_destroy_container(handler->conf);
1316                 if (!bret) {
1317                         ERROR("Error destroying rootfs for %s", name);
1318                         return;
1319                 }
1320         }
1321         INFO("Destroyed rootfs for %s", name);
1322
1323         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1324         if (ret < 0 || ret >= MAXPATHLEN) {
1325                 ERROR("Error printing path for %s", name);
1326                 ERROR("Error destroying directory for %s", name);
1327                 return;
1328         }
1329
1330         c = lxc_container_new(name, handler->lxcpath);
1331         if (c) {
1332                 if (container_disk_lock(c)) {
1333                         INFO("Could not update lxc_snapshots file");
1334                         lxc_container_put(c);
1335                 } else {
1336                         mod_all_rdeps(c, false);
1337                         container_disk_unlock(c);
1338                         lxc_container_put(c);
1339                 }
1340         }
1341
1342         if (am_unpriv())
1343                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, destroy);
1344         else
1345                 ret = lxc_rmdir_onedev(destroy, NULL);
1346
1347         if (ret < 0) {
1348                 ERROR("Error destroying directory for %s", name);
1349                 return;
1350         }
1351         INFO("Destroyed directory for %s", name);
1352 }
1353
1354 static int lxc_rmdir_onedev_wrapper(void *data)
1355 {
1356         char *arg = (char *) data;
1357         return lxc_rmdir_onedev(arg, NULL);
1358 }
1359
1360 static bool do_destroy_container(struct lxc_conf *conf) {
1361         if (am_unpriv()) {
1362                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf) < 0)
1363                         return false;
1364                 return true;
1365         }
1366         return bdev_destroy(conf);
1367 }
1368