src/lxc/start.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #include "config.h"
  25
  26 #include <stdio.h>
  27 #include <string.h>
  28 #include <stdlib.h>
  29 #include <dirent.h>
  30 #include <errno.h>
  31 #include <unistd.h>
  32 #include <signal.h>
  33 #include <fcntl.h>
  34 #include <grp.h>
  35 #include <poll.h>
  36 #include <sys/param.h>
  37 #include <sys/file.h>
  38 #include <sys/mount.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/socket.h>
  42 #include <sys/prctl.h>
  43 #include <sys/types.h>
  44 #include <sys/wait.h>
  45 #include <sys/un.h>
  46 #include <sys/syscall.h>
  47
  48 #if HAVE_SYS_CAPABILITY_H
  49 #include <sys/capability.h>
  50 #endif
  51
  52 #if !HAVE_DECL_PR_CAPBSET_DROP
  53 #define PR_CAPBSET_DROP 24
  54 #endif
  55
  56 #include "start.h"
  57 #include "conf.h"
  58 #include "log.h"
  59 #include "cgroup.h"
  60 #include "error.h"
  61 #include "af_unix.h"
  62 #include "mainloop.h"
  63 #include "utils.h"
  64 #include "lxcutmp.h"
  65 #include "monitor.h"
  66 #include "commands.h"
  67 #include "console.h"
  68 #include "sync.h"
  69 #include "namespace.h"
  70 #include "lxcseccomp.h"
  71 #include "caps.h"
  72 #include "bdev.h"
  73 #include "lsm/lsm.h"
  74
  75 lxc_log_define(lxc_start, lxc);
  76
  77 const struct ns_info ns_info[LXC_NS_MAX] = {
  78         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  79         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  80         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  81         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  82         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  83         [LXC_NS_NET] = {"net", CLONE_NEWNET}
  84 };
  85
  86 static bool do_destroy_container(struct lxc_conf *conf);
  87 static int lxc_rmdir_onedev_wrapper(void *data);
  88 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
  89                                             const char *name);
  90
  91 static void print_top_failing_dir(const char *path)
  92 {
  93         size_t len = strlen(path);
  94         char *copy = alloca(len+1), *p, *e, saved;
  95         strcpy(copy, path);
  96
  97         p = copy;
  98         e = copy + len;
  99         while (p < e) {
 100                 while (p < e && *p == '/') p++;
 101                 while (p < e && *p != '/') p++;
 102                 saved = *p;
 103                 *p = '\0';
 104                 if (access(copy, X_OK)) {
 105                         SYSERROR("could not access %s.  Please grant it 'x' " \
 106                               "access, or add an ACL for the container root.",
 107                               copy);
 108                         return;
 109                 }
 110                 *p = saved;
 111         }
 112 }
 113
 114 static void close_ns(int ns_fd[LXC_NS_MAX]) {
 115         int i;
 116
 117         for (i = 0; i < LXC_NS_MAX; i++) {
 118                 if (ns_fd[i] > -1) {
 119                         close(ns_fd[i]);
 120                         ns_fd[i] = -1;
 121                 }
 122         }
 123 }
 124
 125 static int preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags) {
 126         int i, saved_errno;
 127         char path[MAXPATHLEN];
 128
 129         for (i = 0; i < LXC_NS_MAX; i++)
 130                 ns_fd[i] = -1;
 131
 132         if (access("/proc/self/ns", X_OK)) {
 133                 WARN("Kernel does not support attach; preserve_ns ignored");
 134                 return 0;
 135         }
 136
 137         for (i = 0; i < LXC_NS_MAX; i++) {
 138                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 139                         continue;
 140                 snprintf(path, MAXPATHLEN, "/proc/self/ns/%s", ns_info[i].proc_name);
 141                 ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC);
 142                 if (ns_fd[i] < 0)
 143                         goto error;
 144         }
 145
 146         return 0;
 147
 148 error:
 149         saved_errno = errno;
 150         close_ns(ns_fd);
 151         errno = saved_errno;
 152         SYSERROR("failed to open '%s'", path);
 153         return -1;
 154 }
 155
 156 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 157         int i;
 158
 159         for (i = 0; i < LXC_NS_MAX; i++) {
 160                 if (ns_fd[i] < 0)
 161                         continue;
 162
 163                 if (setns(ns_fd[i], 0) != 0)
 164                         goto error;
 165         }
 166         return 0;
 167
 168 error:
 169         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 170         return -1;
 171 }
 172
 173 static int match_fd(int fd)
 174 {
 175         return (fd == 0 || fd == 1 || fd == 2);
 176 }
 177
 178 /*
 179  * Check for any fds we need to close
 180  * * if fd_to_ignore != -1, then if we find that fd open we will ignore it.
 181  * * By default we warn about open fds we find.
 182  * * If closeall is true, we will close open fds.
 183  * * If lxc-start was passed "-C", then conf->close_all_fds will be true,
 184  *     in which case we also close all open fds.
 185  * * A daemonized container will always pass closeall=true.
 186  */
 187 int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore)
 188 {
 189         struct dirent dirent, *direntp;
 190         int fd, fddir;
 191         DIR *dir;
 192
 193         if (conf && conf->close_all_fds)
 194                 closeall = true;
 195
 196 restart:
 197         dir = opendir("/proc/self/fd");
 198         if (!dir) {
 199                 WARN("failed to open directory: %m");
 200                 return -1;
 201         }
 202
 203         fddir = dirfd(dir);
 204
 205         while (!readdir_r(dir, &dirent, &direntp)) {
 206                 if (!direntp)
 207                         break;
 208
 209                 if (!strcmp(direntp->d_name, "."))
 210                         continue;
 211
 212                 if (!strcmp(direntp->d_name, ".."))
 213                         continue;
 214
 215                 fd = atoi(direntp->d_name);
 216
 217                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 218                         continue;
 219
 220                 if (current_config && fd == current_config->logfd)
 221                         continue;
 222
 223                 if (match_fd(fd))
 224                         continue;
 225
 226                 if (closeall) {
 227                         close(fd);
 228                         closedir(dir);
 229                         INFO("closed inherited fd %d", fd);
 230                         goto restart;
 231                 }
 232                 WARN("inherited fd %d", fd);
 233         }
 234
 235         closedir(dir); /* cannot fail */
 236         return 0;
 237 }
 238
 239 static int setup_signal_fd(sigset_t *oldmask)
 240 {
 241         sigset_t mask;
 242         int fd;
 243
 244         /* Block everything except serious error signals */
 245         if (sigfillset(&mask) ||
 246             sigdelset(&mask, SIGILL) ||
 247             sigdelset(&mask, SIGSEGV) ||
 248             sigdelset(&mask, SIGBUS) ||
 249             sigdelset(&mask, SIGWINCH) ||
 250             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 251                 SYSERROR("failed to set signal mask");
 252                 return -1;
 253         }
 254
 255         fd = signalfd(-1, &mask, 0);
 256         if (fd < 0) {
 257                 SYSERROR("failed to create the signal fd");
 258                 return -1;
 259         }
 260
 261         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 262                 SYSERROR("failed to set sigfd to close-on-exec");
 263                 close(fd);
 264                 return -1;
 265         }
 266
 267         DEBUG("sigchild handler set");
 268
 269         return fd;
 270 }
 271
 272 static int signal_handler(int fd, uint32_t events, void *data,
 273                            struct lxc_epoll_descr *descr)
 274 {
 275         struct signalfd_siginfo siginfo;
 276         siginfo_t info;
 277         int ret;
 278         pid_t *pid = data;
 279         bool init_died = false;
 280
 281         ret = read(fd, &siginfo, sizeof(siginfo));
 282         if (ret < 0) {
 283                 ERROR("failed to read signal info");
 284                 return -1;
 285         }
 286
 287         if (ret != sizeof(siginfo)) {
 288                 ERROR("unexpected siginfo size");
 289                 return -1;
 290         }
 291
 292         // check whether init is running
 293         info.si_pid = 0;
 294         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 295         if (ret == 0 && info.si_pid == *pid) {
 296                 init_died = true;
 297         }
 298
 299         if (siginfo.ssi_signo != SIGCHLD) {
 300                 kill(*pid, siginfo.ssi_signo);
 301                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 302                 return init_died ? 1 : 0;
 303         }
 304
 305         if (siginfo.ssi_code == CLD_STOPPED ||
 306             siginfo.ssi_code == CLD_CONTINUED) {
 307                 INFO("container init process was stopped/continued");
 308                 return init_died ? 1 : 0;
 309         }
 310
 311         /* more robustness, protect ourself from a SIGCHLD sent
 312          * by a process different from the container init
 313          */
 314         if (siginfo.ssi_pid != *pid) {
 315                 WARN("invalid pid for SIGCHLD");
 316                 return init_died ? 1 : 0;
 317         }
 318
 319         DEBUG("container init process exited");
 320         return 1;
 321 }
 322
 323 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 324 {
 325         handler->state = state;
 326         lxc_monitor_send_state(name, state, handler->lxcpath);
 327         return 0;
 328 }
 329
 330 int lxc_poll(const char *name, struct lxc_handler *handler)
 331 {
 332         int sigfd = handler->sigfd;
 333         int pid = handler->pid;
 334         struct lxc_epoll_descr descr;
 335
 336         if (lxc_mainloop_open(&descr)) {
 337                 ERROR("failed to create mainloop");
 338                 goto out_sigfd;
 339         }
 340
 341         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 342                 ERROR("failed to add handler for the signal");
 343                 goto out_mainloop_open;
 344         }
 345
 346         if (lxc_console_mainloop_add(&descr, handler)) {
 347                 ERROR("failed to add console handler to mainloop");
 348                 goto out_mainloop_open;
 349         }
 350
 351         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 352                 ERROR("failed to add command handler to mainloop");
 353                 goto out_mainloop_open;
 354         }
 355
 356         if (handler->conf->need_utmp_watch) {
 357                 #if HAVE_SYS_CAPABILITY_H
 358                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 359                         ERROR("failed to add utmp handler to mainloop");
 360                         goto out_mainloop_open;
 361                 }
 362                 #else
 363                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
 364                 #endif
 365         }
 366
 367         return lxc_mainloop(&descr, -1);
 368
 369 out_mainloop_open:
 370         lxc_mainloop_close(&descr);
 371 out_sigfd:
 372         close(sigfd);
 373         return -1;
 374 }
 375
 376 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 377 {
 378         struct lxc_handler *handler;
 379
 380         handler = malloc(sizeof(*handler));
 381         if (!handler)
 382                 return NULL;
 383
 384         memset(handler, 0, sizeof(*handler));
 385
 386         handler->ttysock[0] = handler->ttysock[1] = -1;
 387         handler->conf = conf;
 388         handler->lxcpath = lxcpath;
 389         handler->pinfd = -1;
 390
 391         lsm_init();
 392
 393         handler->name = strdup(name);
 394         if (!handler->name) {
 395                 ERROR("failed to allocate memory");
 396                 goto out_free;
 397         }
 398
 399         if (lxc_cmd_init(name, handler, lxcpath))
 400                 goto out_free_name;
 401
 402         if (lxc_read_seccomp_config(conf) != 0) {
 403                 ERROR("failed loading seccomp policy");
 404                 goto out_close_maincmd_fd;
 405         }
 406
 407         /* Begin by setting the state to STARTING */
 408         if (lxc_set_state(name, handler, STARTING)) {
 409                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 410                 goto out_close_maincmd_fd;
 411         }
 412
 413         /* Start of environment variable setup for hooks */
 414         if (name && setenv("LXC_NAME", name, 1)) {
 415                 SYSERROR("failed to set environment variable for container name");
 416         }
 417         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 418                 SYSERROR("failed to set environment variable for config path");
 419         }
 420         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 421                 SYSERROR("failed to set environment variable for rootfs mount");
 422         }
 423         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 424                 SYSERROR("failed to set environment variable for rootfs mount");
 425         }
 426         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 427                 SYSERROR("failed to set environment variable for console path");
 428         }
 429         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 430                 SYSERROR("failed to set environment variable for console log");
 431         }
 432         /* End of environment variable setup for hooks */
 433
 434         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 435                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 436                 goto out_aborting;
 437         }
 438
 439         /* the signal fd has to be created before forking otherwise
 440          * if the child process exits before we setup the signal fd,
 441          * the event will be lost and the command will be stuck */
 442         handler->sigfd = setup_signal_fd(&handler->oldmask);
 443         if (handler->sigfd < 0) {
 444                 ERROR("failed to set sigchild fd handler");
 445                 goto out_delete_tty;
 446         }
 447
 448         /* do this after setting up signals since it might unblock SIGWINCH */
 449         if (lxc_console_create(conf)) {
 450                 ERROR("failed to create console");
 451                 goto out_restore_sigmask;
 452         }
 453
 454         if (ttys_shift_ids(conf) < 0) {
 455                 ERROR("Failed to shift tty into container");
 456                 goto out_restore_sigmask;
 457         }
 458
 459         INFO("'%s' is initialized", name);
 460         return handler;
 461
 462 out_restore_sigmask:
 463         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 464 out_delete_tty:
 465         lxc_delete_tty(&conf->tty_info);
 466 out_aborting:
 467         lxc_set_state(name, handler, ABORTING);
 468 out_close_maincmd_fd:
 469         close(conf->maincmd_fd);
 470         conf->maincmd_fd = -1;
 471 out_free_name:
 472         free(handler->name);
 473         handler->name = NULL;
 474 out_free:
 475         free(handler);
 476         return NULL;
 477 }
 478
 479 void lxc_fini(const char *name, struct lxc_handler *handler)
 480 {
 481         /* The STOPPING state is there for future cleanup code
 482          * which can take awhile
 483          */
 484         lxc_set_state(name, handler, STOPPING);
 485         lxc_set_state(name, handler, STOPPED);
 486
 487         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL))
 488                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 489
 490         /* reset mask set by setup_signal_fd */
 491         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 492                 WARN("failed to restore sigprocmask");
 493
 494         lxc_console_delete(&handler->conf->console);
 495         lxc_delete_tty(&handler->conf->tty_info);
 496         close(handler->conf->maincmd_fd);
 497         handler->conf->maincmd_fd = -1;
 498         free(handler->name);
 499         if (handler->ttysock[0] != -1) {
 500                 close(handler->ttysock[0]);
 501                 close(handler->ttysock[1]);
 502         }
 503         if (handler->conf->ephemeral == 1) {
 504                 lxc_destroy_container_on_signal(handler, name);
 505         }
 506         cgroup_destroy(handler);
 507         free(handler);
 508 }
 509
 510 void lxc_abort(const char *name, struct lxc_handler *handler)
 511 {
 512         int ret, status;
 513
 514         lxc_set_state(name, handler, ABORTING);
 515         if (handler->pid > 0)
 516                 kill(handler->pid, SIGKILL);
 517         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 518 }
 519
 520 #include <sys/reboot.h>
 521 #include <linux/reboot.h>
 522
 523 /*
 524  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 525  * in a child pid namespace if container reboot support exists.
 526  * Otherwise, it will either succeed or return -EPERM.
 527  */
 528 static int container_reboot_supported(void *arg)
 529 {
 530         int *cmd = arg;
 531         int ret;
 532
 533         ret = reboot(*cmd);
 534         if (ret == -1 && errno == EINVAL)
 535                 return 1;
 536         return 0;
 537 }
 538
 539 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 540 {
 541         FILE *f;
 542         int ret, cmd, v, flags;
 543         long stack_size = 4096;
 544         void *stack = alloca(stack_size);
 545         int status;
 546         pid_t pid;
 547
 548         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 549         if (!f) {
 550                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 551                 return 1;
 552         }
 553
 554         ret = fscanf(f, "%d", &v);
 555         fclose(f);
 556         if (ret != 1) {
 557                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 558                 return 1;
 559         }
 560         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 561
 562         flags = CLONE_NEWPID | SIGCHLD;
 563         if (!lxc_list_empty(&conf->id_map))
 564                 flags |= CLONE_NEWUSER;
 565
 566 #ifdef __ia64__
 567         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 568 #else
 569         stack += stack_size;
 570         pid = clone(container_reboot_supported, stack, flags, &cmd);
 571 #endif
 572         if (pid < 0) {
 573                 if (flags & CLONE_NEWUSER)
 574                         ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno));
 575                 else
 576                         ERROR("failed to clone (%#x): %s", flags, strerror(errno));
 577                 return -1;
 578         }
 579         if (wait(&status) < 0) {
 580                 SYSERROR("unexpected wait error: %m");
 581                 return -1;
 582         }
 583
 584         if (WEXITSTATUS(status) != 1)
 585                 return 1;
 586
 587         return 0;
 588 }
 589
 590 /*
 591  * netpipe is used in the unprivileged case to transfer the ifindexes
 592  * from parent to child
 593  */
 594 static int netpipe = -1;
 595
 596 static inline int count_veths(struct lxc_list *network)
 597 {
 598         struct lxc_list *iterator;
 599         struct lxc_netdev *netdev;
 600         int count = 0;
 601
 602         lxc_list_for_each(iterator, network) {
 603                 netdev = iterator->elem;
 604                 if (netdev->type != LXC_NET_VETH)
 605                         continue;
 606                 count++;
 607         }
 608         return count;
 609 }
 610
 611 static int read_unpriv_netifindex(struct lxc_list *network)
 612 {
 613         struct lxc_list *iterator;
 614         struct lxc_netdev *netdev;
 615
 616         if (netpipe == -1)
 617                 return 0;
 618         lxc_list_for_each(iterator, network) {
 619                 netdev = iterator->elem;
 620                 if (netdev->type != LXC_NET_VETH)
 621                         continue;
 622                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 623                         ERROR("Out of memory");
 624                         close(netpipe);
 625                         return -1;
 626                 }
 627                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 628                         close(netpipe);
 629                         return -1;
 630                 }
 631         }
 632         close(netpipe);
 633         return 0;
 634 }
 635
 636 static int do_start(void *data)
 637 {
 638         struct lxc_list *iterator;
 639         struct lxc_handler *handler = data;
 640
 641         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 642                 SYSERROR("failed to set sigprocmask");
 643                 return -1;
 644         }
 645
 646         /* This prctl must be before the synchro, so if the parent
 647          * dies before we set the parent death signal, we will detect
 648          * its death with the synchro right after, otherwise we have
 649          * a window where the parent can exit before we set the pdeath
 650          * signal leading to a unsupervized container.
 651          */
 652         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 653                 SYSERROR("failed to set pdeath signal");
 654                 return -1;
 655         }
 656
 657         lxc_sync_fini_parent(handler);
 658
 659         /* don't leak the pinfd to the container */
 660         if (handler->pinfd >= 0) {
 661                 close(handler->pinfd);
 662         }
 663
 664         /* Tell the parent task it can begin to configure the
 665          * container and wait for it to finish
 666          */
 667         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 668                 return -1;
 669
 670         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 671                 goto out_warn_father;
 672
 673         /*
 674          * if we are in a new user namespace, become root there to have
 675          * privilege over our namespace. When using lxc-execute we default to root,
 676          * but this can be overriden using the lxc.init_uid and lxc.init_gid
 677          * configuration options.
 678          */
 679         if (!lxc_list_empty(&handler->conf->id_map)) {
 680                 gid_t new_gid = 0;
 681                 if (handler->conf->is_execute && handler->conf->init_gid)
 682                         new_gid = handler->conf->init_gid;
 683
 684                 uid_t new_uid = 0;
 685                 if (handler->conf->is_execute && handler->conf->init_uid)
 686                         new_uid = handler->conf->init_uid;
 687
 688                 NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid);
 689                 if (setgid(new_gid)) {
 690                         SYSERROR("setgid");
 691                         goto out_warn_father;
 692                 }
 693                 if (setuid(new_uid)) {
 694                         SYSERROR("setuid");
 695                         goto out_warn_father;
 696                 }
 697                 if (setgroups(0, NULL)) {
 698                         SYSERROR("setgroups");
 699                         goto out_warn_father;
 700                 }
 701         }
 702
 703         if (access(handler->lxcpath, X_OK)) {
 704                 print_top_failing_dir(handler->lxcpath);
 705                 goto out_warn_father;
 706         }
 707
 708         #if HAVE_SYS_CAPABILITY_H
 709         if (handler->conf->need_utmp_watch) {
 710                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 711                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 712                         goto out_warn_father;
 713                 }
 714                 DEBUG("Dropped cap_sys_boot");
 715         }
 716         #endif
 717
 718         /* Setup the container, ip, names, utsname, ... */
 719         if (lxc_setup(handler)) {
 720                 ERROR("failed to setup the container");
 721                 goto out_warn_father;
 722         }
 723
 724         /* ask father to setup cgroups and wait for him to finish */
 725         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 726                 return -1;
 727
 728         /* Set the label to change to when we exec(2) the container's init */
 729         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 730                 goto out_warn_father;
 731
 732         /* Some init's such as busybox will set sane tty settings on stdin,
 733          * stdout, stderr which it thinks is the console. We already set them
 734          * the way we wanted on the real terminal, and we want init to do its
 735          * setup on its console ie. the pty allocated in lxc_console_create()
 736          * so make sure that that pty is stdin,stdout,stderr.
 737          */
 738         if (lxc_console_set_stdfds(handler) < 0)
 739                 goto out_warn_father;
 740
 741         /* If we mounted a temporary proc, then unmount it now */
 742         tmp_proc_unmount(handler->conf);
 743
 744         if (lxc_seccomp_load(handler->conf) != 0)
 745                 goto out_warn_father;
 746
 747         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 748                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 749                 goto out_warn_father;
 750         }
 751
 752         /* The clearenv() and putenv() calls have been moved here
 753          * to allow us to use environment variables passed to the various
 754          * hooks, such as the start hook above.  Not all of the
 755          * variables like CONFIG_PATH or ROOTFS are valid in this
 756          * context but others are. */
 757         if (clearenv()) {
 758                 SYSERROR("failed to clear environment");
 759                 /* don't error out though */
 760         }
 761
 762         lxc_list_for_each(iterator, &handler->conf->environment) {
 763                 if (putenv((char *)iterator->elem)) {
 764                         SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem);
 765                         goto out_warn_father;
 766                 }
 767         }
 768
 769         if (putenv("container=lxc")) {
 770                 SYSERROR("failed to set environment variable 'container=lxc'");
 771                 goto out_warn_father;
 772         }
 773
 774         if (handler->conf->pty_names) {
 775                 if (putenv(handler->conf->pty_names)) {
 776                         SYSERROR("failed to set environment variable for container ptys");
 777                         goto out_warn_father;
 778                 }
 779         }
 780
 781         close(handler->sigfd);
 782
 783         if (handler->backgrounded && null_stdfds() < 0)
 784                 goto out_warn_father;
 785
 786         /* after this call, we are in error because this
 787          * ops should not return as it execs */
 788         handler->ops->start(handler, handler->data);
 789
 790 out_warn_father:
 791         /* we want the parent to know something went wrong, so any
 792          * value other than what it expects is ok. */
 793         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 794         return -1;
 795 }
 796
 797 static int save_phys_nics(struct lxc_conf *conf)
 798 {
 799         struct lxc_list *iterator;
 800
 801         lxc_list_for_each(iterator, &conf->network) {
 802                 struct lxc_netdev *netdev = iterator->elem;
 803
 804                 if (netdev->type != LXC_NET_PHYS)
 805                         continue;
 806                 conf->saved_nics = realloc(conf->saved_nics,
 807                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 808                 if (!conf->saved_nics) {
 809                         SYSERROR("failed to allocate memory");
 810                         return -1;
 811                 }
 812                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 813                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 814                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 815                         SYSERROR("failed to allocate memory");
 816                         return -1;
 817                 }
 818                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 819                         conf->saved_nics[conf->num_savednics].ifindex,
 820                         conf->saved_nics[conf->num_savednics].orig_name);
 821                 conf->num_savednics++;
 822         }
 823
 824         return 0;
 825 }
 826
 827 static int recv_fd(int sock, int *fd)
 828 {
 829         if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) {
 830                 SYSERROR("Error receiving tty fd from child");
 831                 return -1;
 832         }
 833         if (*fd == -1)
 834                 return -1;
 835         return 0;
 836 }
 837
 838 static int recv_ttys_from_child(struct lxc_handler *handler)
 839 {
 840         struct lxc_conf *conf = handler->conf;
 841         int i, sock = handler->ttysock[1];
 842         struct lxc_tty_info *tty_info = &conf->tty_info;
 843
 844         if (!conf->tty)
 845                 return 0;
 846
 847         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty);
 848         if (!tty_info->pty_info) {
 849                 SYSERROR("failed to allocate pty_info");
 850                 return -1;
 851         }
 852
 853         for (i = 0; i < conf->tty; i++) {
 854                 struct lxc_pty_info *pty_info = &tty_info->pty_info[i];
 855                 pty_info->busy = 0;
 856                 if (recv_fd(sock, &pty_info->slave) < 0 ||
 857                                 recv_fd(sock, &pty_info->master) < 0) {
 858                         ERROR("Error receiving tty info from child");
 859                         return -1;
 860                 }
 861         }
 862         tty_info->nbtty = conf->tty;
 863
 864         return 0;
 865 }
 866
 867 void resolve_clone_flags(struct lxc_handler *handler)
 868 {
 869         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
 870
 871         if (!lxc_list_empty(&handler->conf->id_map)) {
 872                 INFO("Cloning a new user namespace");
 873                 handler->clone_flags |= CLONE_NEWUSER;
 874         }
 875
 876         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
 877                 if (!lxc_requests_empty_network(handler))
 878                         handler->clone_flags |= CLONE_NEWNET;
 879         } else {
 880                 INFO("Inheriting a net namespace");
 881         }
 882
 883         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
 884                 handler->clone_flags |= CLONE_NEWIPC;
 885         } else {
 886                 INFO("Inheriting an IPC namespace");
 887         }
 888
 889         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 890                 handler->clone_flags |= CLONE_NEWUTS;
 891         } else {
 892                 INFO("Inheriting a UTS namespace");
 893         }
 894 }
 895
 896 static int lxc_spawn(struct lxc_handler *handler)
 897 {
 898         int failed_before_rename = 0;
 899         const char *name = handler->name;
 900         bool cgroups_connected = false;
 901         int saved_ns_fd[LXC_NS_MAX];
 902         int preserve_mask = 0, i;
 903         int netpipepair[2], nveths;
 904
 905         netpipe = -1;
 906
 907         for (i = 0; i < LXC_NS_MAX; i++)
 908                 if (handler->conf->inherit_ns_fd[i] != -1)
 909                         preserve_mask |= ns_info[i].clone_flag;
 910
 911         if (lxc_sync_init(handler))
 912                 return -1;
 913
 914         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
 915                 lxc_sync_fini(handler);
 916                 return -1;
 917         }
 918
 919         resolve_clone_flags(handler);
 920
 921         if (handler->clone_flags & CLONE_NEWNET) {
 922                 if (!lxc_list_empty(&handler->conf->network)) {
 923
 924                         /* Find gateway addresses from the link device, which is
 925                          * no longer accessible inside the container. Do this
 926                          * before creating network interfaces, since goto
 927                          * out_delete_net does not work before lxc_clone. */
 928                         if (lxc_find_gateway_addresses(handler)) {
 929                                 ERROR("failed to find gateway addresses");
 930                                 lxc_sync_fini(handler);
 931                                 return -1;
 932                         }
 933
 934                         /* that should be done before the clone because we will
 935                          * fill the netdev index and use them in the child
 936                          */
 937                         if (lxc_create_network(handler)) {
 938                                 ERROR("failed to create the network");
 939                                 lxc_sync_fini(handler);
 940                                 return -1;
 941                         }
 942                 }
 943
 944                 if (save_phys_nics(handler->conf)) {
 945                         ERROR("failed to save physical nic info");
 946                         goto out_abort;
 947                 }
 948         }
 949
 950         if (!cgroup_init(handler)) {
 951                 ERROR("failed initializing cgroup support");
 952                 goto out_delete_net;
 953         }
 954
 955         cgroups_connected = true;
 956
 957         if (!cgroup_create(handler)) {
 958                 ERROR("failed creating cgroups");
 959                 goto out_delete_net;
 960         }
 961
 962         /*
 963          * if the rootfs is not a blockdev, prevent the container from
 964          * marking it readonly.
 965          *
 966          * if the container is unprivileged then skip rootfs pinning
 967          */
 968         if (lxc_list_empty(&handler->conf->id_map)) {
 969                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
 970                 if (handler->pinfd == -1)
 971                         INFO("failed to pin the container's rootfs");
 972         }
 973
 974         if (preserve_ns(saved_ns_fd, preserve_mask) < 0)
 975                 goto out_delete_net;
 976         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
 977                 goto out_delete_net;
 978
 979         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
 980                 if (pipe(netpipepair) < 0) {
 981                         SYSERROR("Error creating pipe");
 982                         goto out_delete_net;
 983                 }
 984                 /* store netpipe in the global var for do_start's use */
 985                 netpipe = netpipepair[0];
 986         }
 987
 988         /* Create a process in a new set of namespaces */
 989         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
 990         if (handler->pid < 0) {
 991                 SYSERROR("failed to fork into a new namespace");
 992                 goto out_delete_net;
 993         }
 994
 995         if (attach_ns(saved_ns_fd))
 996                 WARN("failed to restore saved namespaces");
 997
 998         lxc_sync_fini_child(handler);
 999
1000         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
1001                 failed_before_rename = 1;
1002
1003         if (!cgroup_create_legacy(handler)) {
1004                 ERROR("failed to setup the legacy cgroups for %s", name);
1005                 goto out_delete_net;
1006         }
1007         if (!cgroup_setup_limits(handler, false)) {
1008                 ERROR("failed to setup the cgroup limits for '%s'", name);
1009                 goto out_delete_net;
1010         }
1011
1012         if (!cgroup_enter(handler))
1013                 goto out_delete_net;
1014
1015         if (!cgroup_chown(handler))
1016                 goto out_delete_net;
1017
1018         if (failed_before_rename)
1019                 goto out_delete_net;
1020
1021         /* Create the network configuration */
1022         if (handler->clone_flags & CLONE_NEWNET) {
1023                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
1024                         ERROR("failed to create the configured network");
1025                         goto out_delete_net;
1026                 }
1027         }
1028
1029         if (netpipe != -1) {
1030                 struct lxc_list *iterator;
1031                 struct lxc_netdev *netdev;
1032
1033                 close(netpipe);
1034                 lxc_list_for_each(iterator, &handler->conf->network) {
1035                         netdev = iterator->elem;
1036                         if (netdev->type != LXC_NET_VETH)
1037                                 continue;
1038                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1039                                 ERROR("Error writing veth name to container");
1040                                 goto out_delete_net;
1041                         }
1042                 }
1043                 close(netpipepair[1]);
1044         }
1045
1046         /* map the container uids - the container became an invalid
1047          * userid the moment it was cloned with CLONE_NEWUSER - this
1048          * call doesn't change anything immediately, but allows the
1049          * container to setuid(0) (0 being mapped to something else on
1050          * the host) later to become a valid uid again */
1051         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
1052                 ERROR("failed to set up id mapping");
1053                 goto out_delete_net;
1054         }
1055
1056         /* Tell the child to continue its initialization.  we'll get
1057          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1058          */
1059         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1060                 goto out_delete_net;
1061
1062         if (!cgroup_setup_limits(handler, true)) {
1063                 ERROR("failed to setup the devices cgroup for '%s'", name);
1064                 goto out_delete_net;
1065         }
1066
1067         cgroup_disconnect();
1068         cgroups_connected = false;
1069
1070         /* read tty fds allocated by child */
1071         if (recv_ttys_from_child(handler) < 0) {
1072                 ERROR("failed to receive tty info from child");
1073                 goto out_delete_net;
1074         }
1075
1076         /* Tell the child to complete its initialization and wait for
1077          * it to exec or return an error.  (the child will never
1078          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
1079          * sync pipe, causing lxc_sync_barrier_child to return
1080          * success, or return a different value, causing us to error
1081          * out).
1082          */
1083         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1084                 return -1;
1085
1086         if (detect_shared_rootfs())
1087                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
1088
1089         if (handler->ops->post_start(handler, handler->data))
1090                 goto out_abort;
1091
1092         if (lxc_set_state(name, handler, RUNNING)) {
1093                 ERROR("failed to set state to %s",
1094                               lxc_state2str(RUNNING));
1095                 goto out_abort;
1096         }
1097
1098         lxc_sync_fini(handler);
1099
1100         return 0;
1101
1102 out_delete_net:
1103         if (cgroups_connected)
1104                 cgroup_disconnect();
1105         if (handler->clone_flags & CLONE_NEWNET)
1106                 lxc_delete_network(handler);
1107 out_abort:
1108         lxc_abort(name, handler);
1109         lxc_sync_fini(handler);
1110         if (handler->pinfd >= 0) {
1111                 close(handler->pinfd);
1112                 handler->pinfd = -1;
1113         }
1114
1115         return -1;
1116 }
1117
1118 int get_netns_fd(int pid)
1119 {
1120         char path[MAXPATHLEN];
1121         int ret, fd;
1122
1123         ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid);
1124         if (ret < 0 || ret >= MAXPATHLEN) {
1125                 WARN("Failed to pin netns file for pid %d", pid);
1126                 return -1;
1127         }
1128
1129         fd = open(path, O_RDONLY);
1130         if (fd < 0) {
1131                 WARN("Failed to pin netns file %s for pid %d: %s",
1132                                 path, pid, strerror(errno));
1133                 return -1;
1134         }
1135         return fd;
1136 }
1137
1138 int __lxc_start(const char *name, struct lxc_conf *conf,
1139                 struct lxc_operations* ops, void *data, const char *lxcpath,
1140                 bool backgrounded)
1141 {
1142         struct lxc_handler *handler;
1143         int err = -1;
1144         int status;
1145         int netnsfd = -1;
1146
1147         handler = lxc_init(name, conf, lxcpath);
1148         if (!handler) {
1149                 ERROR("failed to initialize the container");
1150                 return -1;
1151         }
1152         handler->ops = ops;
1153         handler->data = data;
1154         handler->backgrounded = backgrounded;
1155
1156         if (must_drop_cap_sys_boot(handler->conf)) {
1157                 #if HAVE_SYS_CAPABILITY_H
1158                 DEBUG("Dropping cap_sys_boot");
1159                 #else
1160                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1161                 #endif
1162         } else {
1163                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1164                 handler->conf->need_utmp_watch = 0;
1165         }
1166
1167         if (!attach_block_device(handler->conf)) {
1168                 ERROR("Failure attaching block device");
1169                 goto out_fini_nonet;
1170         }
1171
1172         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1173                 /* if the backing store is a device, mount it here and now */
1174                 if (rootfs_is_blockdev(conf)) {
1175                         if (unshare(CLONE_NEWNS) < 0) {
1176                                 ERROR("Error unsharing mounts");
1177                                 goto out_fini_nonet;
1178                         }
1179                         remount_all_slave();
1180                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1181                                 ERROR("Error setting up rootfs mount as root before spawn");
1182                                 goto out_fini_nonet;
1183                         }
1184                         INFO("Set up container rootfs as host root");
1185                 }
1186         }
1187
1188         err = lxc_spawn(handler);
1189         if (err) {
1190                 ERROR("failed to spawn '%s'", name);
1191                 goto out_detach_blockdev;
1192         }
1193
1194         handler->conf->reboot = 0;
1195
1196         netnsfd = get_netns_fd(handler->pid);
1197
1198         err = lxc_poll(name, handler);
1199         if (err) {
1200                 ERROR("mainloop exited with an error");
1201                 if (netnsfd >= 0)
1202                         close(netnsfd);
1203                 goto out_abort;
1204         }
1205
1206         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1207                 continue;
1208
1209         /*
1210          * If the child process exited but was not signaled,
1211          * it didn't call reboot.  This should mean it was an
1212          * lxc-execute which simply exited.  In any case, treat
1213          * it as a 'halt'
1214          */
1215         if (WIFSIGNALED(status)) {
1216                 switch(WTERMSIG(status)) {
1217                 case SIGINT: /* halt */
1218                         DEBUG("Container halting");
1219                         break;
1220                 case SIGHUP: /* reboot */
1221                         DEBUG("Container rebooting");
1222                         handler->conf->reboot = 1;
1223                         break;
1224                 case SIGSYS: /* seccomp */
1225                         DEBUG("Container violated its seccomp policy");
1226                         break;
1227                 default:
1228                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1229                         break;
1230                 }
1231         }
1232
1233         DEBUG("Pushing physical nics back to host namespace");
1234         lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf);
1235
1236         DEBUG("Tearing down virtual network devices used by container");
1237         lxc_delete_network(handler);
1238
1239         if (netnsfd >= 0)
1240                 close(netnsfd);
1241
1242         if (handler->pinfd >= 0) {
1243                 close(handler->pinfd);
1244                 handler->pinfd = -1;
1245         }
1246
1247         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1248         err =  lxc_error_set_and_log(handler->pid, status);
1249 out_fini:
1250         lxc_delete_network(handler);
1251
1252 out_detach_blockdev:
1253         detach_block_device(handler->conf);
1254
1255 out_fini_nonet:
1256         lxc_fini(name, handler);
1257         return err;
1258
1259 out_abort:
1260         lxc_abort(name, handler);
1261         goto out_fini;
1262 }
1263
1264 struct start_args {
1265         char *const *argv;
1266 };
1267
1268 static int start(struct lxc_handler *handler, void* data)
1269 {
1270         struct start_args *arg = data;
1271
1272         NOTICE("exec'ing '%s'", arg->argv[0]);
1273
1274         execvp(arg->argv[0], arg->argv);
1275         SYSERROR("failed to exec %s", arg->argv[0]);
1276         return 0;
1277 }
1278
1279 static int post_start(struct lxc_handler *handler, void* data)
1280 {
1281         struct start_args *arg = data;
1282
1283         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1284         return 0;
1285 }
1286
1287 static struct lxc_operations start_ops = {
1288         .start = start,
1289         .post_start = post_start
1290 };
1291
1292 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1293               const char *lxcpath, bool backgrounded)
1294 {
1295         struct start_args start_arg = {
1296                 .argv = argv,
1297         };
1298
1299         conf->need_utmp_watch = 1;
1300         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded);
1301 }
1302
1303 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1304                                             const char *name)
1305 {
1306         char destroy[MAXPATHLEN];
1307         bool bret = true;
1308         int ret = 0;
1309         if (handler->conf && handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1310                 bret = do_destroy_container(handler->conf);
1311                 if (!bret) {
1312                         ERROR("Error destroying rootfs for %s", name);
1313                         return;
1314                 }
1315         }
1316         INFO("Destroyed rootfs for %s", name);
1317
1318         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1319         if (ret < 0 || ret >= MAXPATHLEN) {
1320                 ERROR("Error printing path for %s", name);
1321                 ERROR("Error destroying directory for %s", name);
1322                 return;
1323         }
1324
1325         if (am_unpriv())
1326                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, destroy);
1327         else
1328                 ret = lxc_rmdir_onedev(destroy, NULL);
1329
1330         if (ret < 0) {
1331                 ERROR("Error destroying directory for %s", name);
1332                 return;
1333         }
1334         INFO("Destroyed directory for %s", name);
1335 }
1336
1337 static int lxc_rmdir_onedev_wrapper(void *data)
1338 {
1339         char *arg = (char *) data;
1340         return lxc_rmdir_onedev(arg, NULL);
1341 }
1342
1343 static bool do_destroy_container(struct lxc_conf *conf) {
1344         if (am_unpriv()) {
1345                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf) < 0)
1346                         return false;
1347                 return true;
1348         }
1349         return bdev_destroy(conf);
1350 }
1351