src/lxc/start.c

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