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 #define _GNU_SOURCE
  25 #include "config.h"
  26
  27 #include <alloca.h>
  28 #include <dirent.h>
  29 #include <errno.h>
  30 #include <fcntl.h>
  31 #include <grp.h>
  32 #include <poll.h>
  33 #include <signal.h>
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <string.h>
  37 #include <unistd.h>
  38 #include <sys/file.h>
  39 #include <sys/mount.h>
  40 #include <sys/param.h>
  41 #include <sys/prctl.h>
  42 #include <sys/socket.h>
  43 #include <sys/stat.h>
  44 #include <sys/syscall.h>
  45 #include <sys/types.h>
  46 #include <sys/un.h>
  47 #include <sys/wait.h>
  48
  49 #if HAVE_LIBCAP
  50 #include <sys/capability.h>
  51 #endif
  52
  53 #ifndef HAVE_DECL_PR_CAPBSET_DROP
  54 #define PR_CAPBSET_DROP 24
  55 #endif
  56
  57 #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS
  58 #define PR_SET_NO_NEW_PRIVS 38
  59 #endif
  60
  61 #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS
  62 #define PR_GET_NO_NEW_PRIVS 39
  63 #endif
  64
  65 #include "af_unix.h"
  66 #include "bdev.h"
  67 #include "caps.h"
  68 #include "cgroup.h"
  69 #include "commands.h"
  70 #include "conf.h"
  71 #include "console.h"
  72 #include "error.h"
  73 #include "log.h"
  74 #include "lxclock.h"
  75 #include "lxcseccomp.h"
  76 #include "lxcutmp.h"
  77 #include "mainloop.h"
  78 #include "monitor.h"
  79 #include "namespace.h"
  80 #include "start.h"
  81 #include "sync.h"
  82 #include "utils.h"
  83 #include "lsm/lsm.h"
  84
  85 lxc_log_define(lxc_start, lxc);
  86
  87 extern void mod_all_rdeps(struct lxc_container *c, bool inc);
  88 static bool do_destroy_container(struct lxc_conf *conf);
  89 static int lxc_rmdir_onedev_wrapper(void *data);
  90 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
  91                                             const char *name);
  92
  93 static void print_top_failing_dir(const char *path)
  94 {
  95         size_t len = strlen(path);
  96         char *copy = alloca(len + 1), *p, *e, saved;
  97         strcpy(copy, path);
  98
  99         p = copy;
 100         e = copy + len;
 101         while (p < e) {
 102                 while (p < e && *p == '/')
 103                         p++;
 104                 while (p < e && *p != '/')
 105                         p++;
 106                 saved = *p;
 107                 *p = '\0';
 108                 if (access(copy, X_OK)) {
 109                         SYSERROR("Could not access %s. Please grant it x "
 110                                  "access, or add an ACL for the container "
 111                                  "root.", copy);
 112                         return;
 113                 }
 114                 *p = saved;
 115         }
 116 }
 117
 118 static void close_ns(int ns_fd[LXC_NS_MAX])
 119 {
 120         int i;
 121
 122         for (i = 0; i < LXC_NS_MAX; i++) {
 123                 if (ns_fd[i] > -1) {
 124                         close(ns_fd[i]);
 125                         ns_fd[i] = -1;
 126                 }
 127         }
 128 }
 129
 130 /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
 131  * in clone_flags.
 132  * Return true on success, false on failure.
 133  */
 134 static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid)
 135 {
 136         int i, ret;
 137
 138         for (i = 0; i < LXC_NS_MAX; i++)
 139                 ns_fd[i] = -1;
 140
 141         ret = lxc_preserve_ns(pid, "");
 142         if (ret < 0) {
 143                 SYSERROR("Kernel does not support attaching to namespaces.");
 144                 return false;
 145         } else {
 146                 close(ret);
 147         }
 148
 149         for (i = 0; i < LXC_NS_MAX; i++) {
 150                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 151                         continue;
 152                 ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name);
 153                 if (ns_fd[i] < 0)
 154                         goto error;
 155         }
 156
 157         return true;
 158
 159 error:
 160         if (errno == ENOENT)
 161                 SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name);
 162         else
 163                 SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno));
 164         close_ns(ns_fd);
 165         return false;
 166 }
 167
 168 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 169         int i;
 170
 171         for (i = 0; i < LXC_NS_MAX; i++) {
 172                 if (ns_fd[i] < 0)
 173                         continue;
 174
 175                 if (setns(ns_fd[i], 0) != 0)
 176                         goto error;
 177         }
 178         return 0;
 179
 180 error:
 181         SYSERROR("Failed to attach %s namespace.", ns_info[i].proc_name);
 182         return -1;
 183 }
 184
 185 static int match_fd(int fd)
 186 {
 187         return (fd == 0 || fd == 1 || fd == 2);
 188 }
 189
 190 /* Check for any fds we need to close.
 191  * - If fd_to_ignore != -1, then if we find that fd open we will ignore it.
 192  * - By default we warn about open fds we find.
 193  * - If closeall is true, we will close open fds.
 194  * - If lxc-start was passed "-C", then conf->close_all_fds will be true, in
 195  *   which case we also close all open fds.
 196  * - A daemonized container will always pass closeall=true.
 197  */
 198 int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore)
 199 {
 200         struct dirent *direntp;
 201         int fd, fddir;
 202         DIR *dir;
 203
 204         if (conf && conf->close_all_fds)
 205                 closeall = true;
 206
 207 restart:
 208         dir = opendir("/proc/self/fd");
 209         if (!dir) {
 210                 WARN("Failed to open directory: %s.", strerror(errno));
 211                 return -1;
 212         }
 213
 214         fddir = dirfd(dir);
 215
 216         while ((direntp = readdir(dir))) {
 217                 if (!direntp)
 218                         break;
 219
 220                 if (!strcmp(direntp->d_name, "."))
 221                         continue;
 222
 223                 if (!strcmp(direntp->d_name, ".."))
 224                         continue;
 225
 226                 if (lxc_safe_int(direntp->d_name, &fd) < 0) {
 227                         INFO("Could not parse file descriptor for: %s", direntp->d_name);
 228                         continue;
 229                 }
 230
 231                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 232                         continue;
 233
 234                 if (current_config && fd == current_config->logfd)
 235                         continue;
 236
 237                 if (match_fd(fd))
 238                         continue;
 239
 240                 if (closeall) {
 241                         close(fd);
 242                         closedir(dir);
 243                         INFO("Closed inherited fd: %d.", fd);
 244                         goto restart;
 245                 }
 246                 WARN("Inherited fd: %d.", fd);
 247         }
 248
 249         /* Only enable syslog at this point to avoid the above logging function
 250          * to open a new fd and make the check_inherited function enter an
 251          * infinite loop.
 252          */
 253         lxc_log_enable_syslog();
 254
 255         closedir(dir); /* cannot fail */
 256         return 0;
 257 }
 258
 259 static int setup_signal_fd(sigset_t *oldmask)
 260 {
 261         sigset_t mask;
 262         int fd;
 263
 264         /* Block everything except serious error signals. */
 265         if (sigfillset(&mask) ||
 266             sigdelset(&mask, SIGILL) ||
 267             sigdelset(&mask, SIGSEGV) ||
 268             sigdelset(&mask, SIGBUS) ||
 269             sigdelset(&mask, SIGWINCH) ||
 270             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 271                 SYSERROR("Failed to set signal mask.");
 272                 return -1;
 273         }
 274
 275         fd = signalfd(-1, &mask, 0);
 276         if (fd < 0) {
 277                 SYSERROR("Failed to create signal file descriptor.");
 278                 return -1;
 279         }
 280
 281         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 282                 SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd);
 283                 close(fd);
 284                 return -1;
 285         }
 286
 287         DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd);
 288
 289         return fd;
 290 }
 291
 292 static int signal_handler(int fd, uint32_t events, void *data,
 293                           struct lxc_epoll_descr *descr)
 294 {
 295         struct signalfd_siginfo siginfo;
 296         siginfo_t info;
 297         int ret;
 298         pid_t *pid = data;
 299         bool init_died = false;
 300
 301         ret = read(fd, &siginfo, sizeof(siginfo));
 302         if (ret < 0) {
 303                 ERROR("Failed to read signal info from signal file descriptor: %d.", fd);
 304                 return -1;
 305         }
 306
 307         if (ret != sizeof(siginfo)) {
 308                 ERROR("Unexpected size for siginfo struct.");
 309                 return -1;
 310         }
 311
 312         /* Check whether init is running. */
 313         info.si_pid = 0;
 314         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 315         if (ret == 0 && info.si_pid == *pid)
 316                 init_died = true;
 317
 318         if (siginfo.ssi_signo != SIGCHLD) {
 319                 kill(*pid, siginfo.ssi_signo);
 320                 INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid);
 321                 return init_died ? 1 : 0;
 322         }
 323
 324         if (siginfo.ssi_code == CLD_STOPPED) {
 325                 INFO("Container init process was stopped.");
 326                 return init_died ? 1 : 0;
 327         } else if (siginfo.ssi_code == CLD_CONTINUED) {
 328                 INFO("Container init process was continued.");
 329                 return init_died ? 1 : 0;
 330         }
 331
 332         /* More robustness, protect ourself from a SIGCHLD sent
 333          * by a process different from the container init.
 334          */
 335         if (siginfo.ssi_pid != *pid) {
 336                 NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid);
 337                 return init_died ? 1 : 0;
 338         }
 339
 340         DEBUG("Container init process %d exited.", *pid);
 341         return 1;
 342 }
 343
 344 int lxc_set_state(const char *name, struct lxc_handler *handler,
 345                   lxc_state_t state)
 346 {
 347         ssize_t ret;
 348         struct lxc_list *cur, *next;
 349         struct state_client *client;
 350         struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 351
 352         process_lock();
 353         /* Only set state under process lock held so that we don't cause
 354          * lxc_cmd_state_server() to miss a state.
 355          */
 356         handler->state = state;
 357         TRACE("set container state to %s", lxc_state2str(state));
 358
 359         if (lxc_list_empty(&handler->state_clients)) {
 360                 TRACE("no state clients registered");
 361                 process_unlock();
 362                 lxc_monitor_send_state(name, state, handler->lxcpath);
 363                 return 0;
 364         }
 365
 366         strncpy(msg.name, name, sizeof(msg.name));
 367         msg.name[sizeof(msg.name) - 1] = 0;
 368
 369         lxc_list_for_each_safe(cur, &handler->state_clients, next) {
 370                 client = cur->elem;
 371
 372                 if (!client->states[state]) {
 373                         TRACE("state %s not registered for state client %d",
 374                               lxc_state2str(state), client->clientfd);
 375                         continue;
 376                 }
 377
 378                 TRACE("sending state %s to state client %d",
 379                       lxc_state2str(state), client->clientfd);
 380
 381         again:
 382                 ret = send(client->clientfd, &msg, sizeof(msg), 0);
 383                 if (ret < 0) {
 384                         if (errno == EINTR)
 385                                 goto again;
 386
 387                         ERROR("failed to send message to client");
 388                 }
 389
 390                 /* kick client from list */
 391                 close(client->clientfd);
 392                 lxc_list_del(cur);
 393                 free(cur->elem);
 394                 free(cur);
 395         }
 396         process_unlock();
 397
 398         /* This function will try to connect to the legacy lxc-monitord state
 399          * server and only exists for backwards compatibility.
 400          */
 401         lxc_monitor_send_state(name, state, handler->lxcpath);
 402
 403         return 0;
 404 }
 405
 406 int lxc_poll(const char *name, struct lxc_handler *handler)
 407 {
 408         int sigfd = handler->sigfd;
 409         int pid = handler->pid;
 410         struct lxc_epoll_descr descr;
 411
 412         if (lxc_mainloop_open(&descr)) {
 413                 ERROR("Failed to create LXC mainloop.");
 414                 goto out_sigfd;
 415         }
 416
 417         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 418                 ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd);
 419                 goto out_mainloop_open;
 420         }
 421
 422         if (lxc_console_mainloop_add(&descr, handler->conf)) {
 423                 ERROR("Failed to add console handler to LXC mainloop.");
 424                 goto out_mainloop_open;
 425         }
 426
 427         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 428                 ERROR("Failed to add command handler to LXC mainloop.");
 429                 goto out_mainloop_open;
 430         }
 431
 432         if (handler->conf->need_utmp_watch) {
 433                 #if HAVE_LIBCAP
 434                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 435                         ERROR("Failed to add utmp handler to LXC mainloop.");
 436                         goto out_mainloop_open;
 437                 }
 438                 #else
 439                         DEBUG("Not starting utmp handler as CAP_SYS_BOOT cannot be dropped without capabilities support.");
 440                 #endif
 441         }
 442         TRACE("lxc mainloop is ready");
 443
 444         return lxc_mainloop(&descr, -1);
 445
 446 out_mainloop_open:
 447         lxc_mainloop_close(&descr);
 448
 449 out_sigfd:
 450         close(sigfd);
 451
 452         return -1;
 453 }
 454
 455 void lxc_free_handler(struct lxc_handler *handler)
 456 {
 457         if (handler->conf && handler->conf->maincmd_fd)
 458                 close(handler->conf->maincmd_fd);
 459
 460         if (handler->name)
 461                 free(handler->name);
 462
 463         handler->conf = NULL;
 464         free(handler);
 465 }
 466
 467 struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf,
 468                                      const char *lxcpath)
 469 {
 470         int i;
 471         struct lxc_handler *handler;
 472
 473         handler = malloc(sizeof(*handler));
 474         if (!handler) {
 475                 ERROR("failed to allocate memory");
 476                 return NULL;
 477         }
 478
 479         memset(handler, 0, sizeof(*handler));
 480
 481         handler->ttysock[0] = handler->ttysock[1] = -1;
 482         handler->conf = conf;
 483         handler->lxcpath = lxcpath;
 484         handler->pinfd = -1;
 485         lxc_list_init(&handler->state_clients);
 486
 487         for (i = 0; i < LXC_NS_MAX; i++)
 488                 handler->nsfd[i] = -1;
 489
 490         handler->name = strdup(name);
 491         if (!handler->name) {
 492                 ERROR("failed to allocate memory");
 493                 goto on_error;
 494         }
 495
 496         if (lxc_cmd_init(name, handler, lxcpath)) {
 497                 ERROR("failed to set up command socket");
 498                 goto on_error;
 499         }
 500
 501         TRACE("unix domain socket %d for command server is ready",
 502               handler->conf->maincmd_fd);
 503
 504         return handler;
 505
 506 on_error:
 507         lxc_free_handler(handler);
 508
 509         return NULL;
 510 }
 511
 512 int lxc_init(const char *name, struct lxc_handler *handler)
 513 {
 514         struct lxc_conf *conf = handler->conf;
 515
 516         lsm_init();
 517         TRACE("initialized LSM");
 518
 519         if (lxc_read_seccomp_config(conf) != 0) {
 520                 ERROR("Failed loading seccomp policy.");
 521                 goto out_close_maincmd_fd;
 522         }
 523         TRACE("read seccomp policy");
 524
 525         /* Begin by setting the state to STARTING. */
 526         if (lxc_set_state(name, handler, STARTING)) {
 527                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING));
 528                 goto out_close_maincmd_fd;
 529         }
 530         TRACE("set container state to \"STARTING\"");
 531
 532         /* Start of environment variable setup for hooks. */
 533         if (name && setenv("LXC_NAME", name, 1))
 534                 SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name);
 535
 536         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1))
 537                 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile);
 538
 539         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1))
 540                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount);
 541
 542         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1))
 543                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path);
 544
 545         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1))
 546                 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path);
 547
 548         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1))
 549                 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path);
 550
 551         if (setenv("LXC_CGNS_AWARE", "1", 1))
 552                 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1.");
 553         /* End of environment variable setup for hooks. */
 554
 555         TRACE("set environment variables");
 556
 557         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 558                 ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name);
 559                 goto out_aborting;
 560         }
 561         TRACE("ran pre-start hooks");
 562
 563         /* The signal fd has to be created before forking otherwise if the child
 564          * process exits before we setup the signal fd, the event will be lost
 565          * and the command will be stuck.
 566          */
 567         handler->sigfd = setup_signal_fd(&handler->oldmask);
 568         if (handler->sigfd < 0) {
 569                 ERROR("Failed to setup SIGCHLD fd handler.");
 570                 goto out_delete_tty;
 571         }
 572         TRACE("set up signal fd");
 573
 574         /* Do this after setting up signals since it might unblock SIGWINCH. */
 575         if (lxc_console_create(conf)) {
 576                 ERROR("Failed to create console for container \"%s\".", name);
 577                 goto out_restore_sigmask;
 578         }
 579         TRACE("created console");
 580
 581         if (lxc_ttys_shift_ids(conf) < 0) {
 582                 ERROR("Failed to shift tty into container.");
 583                 goto out_restore_sigmask;
 584         }
 585         TRACE("shifted tty ids");
 586
 587         INFO("container \"%s\" is initialized", name);
 588         return 0;
 589
 590 out_restore_sigmask:
 591         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 592 out_delete_tty:
 593         lxc_delete_tty(&conf->tty_info);
 594 out_aborting:
 595         lxc_set_state(name, handler, ABORTING);
 596 out_close_maincmd_fd:
 597         close(conf->maincmd_fd);
 598         conf->maincmd_fd = -1;
 599         return -1;
 600 }
 601
 602 void lxc_fini(const char *name, struct lxc_handler *handler)
 603 {
 604         int i, rc;
 605         struct lxc_list *cur, *next;
 606         pid_t self = getpid();
 607         char *namespaces[LXC_NS_MAX + 1];
 608         size_t namespace_count = 0;
 609
 610         /* The STOPPING state is there for future cleanup code which can take
 611          * awhile.
 612          */
 613         lxc_set_state(name, handler, STOPPING);
 614
 615         for (i = 0; i < LXC_NS_MAX; i++) {
 616                 if (handler->nsfd[i] != -1) {
 617                         rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d",
 618                                       ns_info[i].proc_name, self, handler->nsfd[i]);
 619                         if (rc == -1) {
 620                                 SYSERROR("Failed to allocate memory.");
 621                                 break;
 622                         }
 623                         ++namespace_count;
 624                 }
 625         }
 626         namespaces[namespace_count] = NULL;
 627
 628         if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1))
 629                 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot.");
 630
 631         if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1))
 632                 SYSERROR("Failed to set environment variable: LXC_TARGET=stop.");
 633
 634         if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces))
 635                 ERROR("Failed to run lxc.hook.stop for container \"%s\".", name);
 636
 637         while (namespace_count--)
 638                 free(namespaces[namespace_count]);
 639         for (i = 0; i < LXC_NS_MAX; i++) {
 640                 if (handler->nsfd[i] != -1) {
 641                         close(handler->nsfd[i]);
 642                         handler->nsfd[i] = -1;
 643                 }
 644         }
 645
 646         if (handler->netnsfd >= 0) {
 647                 close(handler->netnsfd);
 648                 handler->netnsfd = -1;
 649         }
 650
 651         lxc_set_state(name, handler, STOPPED);
 652
 653         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) {
 654                 ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name);
 655                 if (handler->conf->reboot) {
 656                         WARN("Container will be stopped instead of rebooted.");
 657                         handler->conf->reboot = 0;
 658                         if (setenv("LXC_TARGET", "stop", 1))
 659                                 WARN("Failed to set environment variable: LXC_TARGET=stop.");
 660                 }
 661         }
 662
 663         /* Reset mask set by setup_signal_fd. */
 664         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 665                 WARN("Failed to restore signal mask.");
 666
 667         lxc_console_delete(&handler->conf->console);
 668         lxc_delete_tty(&handler->conf->tty_info);
 669
 670         /* close the command socket */
 671         close(handler->conf->maincmd_fd);
 672         handler->conf->maincmd_fd = -1;
 673
 674         /* The command socket is now closed, no more state clients can register
 675          * themselves from now on. So free the list of state clients.
 676          */
 677         lxc_list_for_each_safe(cur, &handler->state_clients, next) {
 678                 struct state_client *client = cur->elem;
 679                 /* close state client socket */
 680                 close(client->clientfd);
 681                 lxc_list_del(cur);
 682                 free(cur->elem);
 683                 free(cur);
 684         }
 685
 686         free(handler->name);
 687         if (handler->ttysock[0] != -1) {
 688                 close(handler->ttysock[0]);
 689                 close(handler->ttysock[1]);
 690         }
 691
 692         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1)
 693                 lxc_destroy_container_on_signal(handler, name);
 694
 695         cgroup_destroy(handler);
 696         free(handler);
 697 }
 698
 699 void lxc_abort(const char *name, struct lxc_handler *handler)
 700 {
 701         int ret, status;
 702
 703         lxc_set_state(name, handler, ABORTING);
 704         if (handler->pid > 0)
 705                 kill(handler->pid, SIGKILL);
 706         while ((ret = waitpid(-1, &status, 0)) > 0) {
 707                 ;
 708         }
 709 }
 710
 711 #include <sys/reboot.h>
 712 #include <linux/reboot.h>
 713
 714 /* reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL in a child pid namespace
 715  * if container reboot support exists.  Otherwise, it will either succeed or
 716  * return -EPERM.
 717  */
 718 static int container_reboot_supported(void *arg)
 719 {
 720         int *cmd = arg;
 721         int ret;
 722
 723         ret = reboot(*cmd);
 724         if (ret == -1 && errno == EINVAL)
 725                 return 1;
 726         return 0;
 727 }
 728
 729 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 730 {
 731         FILE *f;
 732         int ret, cmd, v, flags;
 733         long stack_size = 4096;
 734         void *stack = alloca(stack_size);
 735         int status;
 736         pid_t pid;
 737
 738         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 739         if (!f) {
 740                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 741                 return 1;
 742         }
 743
 744         ret = fscanf(f, "%d", &v);
 745         fclose(f);
 746         if (ret != 1) {
 747                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del.");
 748                 return 1;
 749         }
 750         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 751
 752         flags = CLONE_NEWPID | SIGCHLD;
 753         if (!lxc_list_empty(&conf->id_map))
 754                 flags |= CLONE_NEWUSER;
 755
 756 #ifdef __ia64__
 757         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 758 #else
 759         stack += stack_size;
 760         pid = clone(container_reboot_supported, stack, flags, &cmd);
 761 #endif
 762         if (pid < 0) {
 763                 if (flags & CLONE_NEWUSER)
 764                         ERROR("Failed to clone (%#x): %s (includes CLONE_NEWUSER).", flags, strerror(errno));
 765                 else
 766                         ERROR("Failed to clone (%#x): %s.", flags, strerror(errno));
 767                 return -1;
 768         }
 769         if (wait(&status) < 0) {
 770                 SYSERROR("Unexpected wait error: %s.", strerror(errno));
 771                 return -1;
 772         }
 773
 774         if (WEXITSTATUS(status) != 1)
 775                 return 1;
 776
 777         return 0;
 778 }
 779
 780 /* netpipe is used in the unprivileged case to transfer the ifindexes from
 781  * parent to child
 782  */
 783 static int netpipe = -1;
 784
 785 static inline int count_veths(struct lxc_list *network)
 786 {
 787         struct lxc_list *iterator;
 788         struct lxc_netdev *netdev;
 789         int count = 0;
 790
 791         lxc_list_for_each(iterator, network) {
 792                 netdev = iterator->elem;
 793                 if (netdev->type != LXC_NET_VETH)
 794                         continue;
 795                 count++;
 796         }
 797         return count;
 798 }
 799
 800 static int read_unpriv_netifindex(struct lxc_list *network)
 801 {
 802         struct lxc_list *iterator;
 803         struct lxc_netdev *netdev;
 804
 805         if (netpipe == -1)
 806                 return 0;
 807         lxc_list_for_each(iterator, network) {
 808                 netdev = iterator->elem;
 809                 if (netdev->type != LXC_NET_VETH)
 810                         continue;
 811                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 812                         ERROR("Out of memory.");
 813                         close(netpipe);
 814                         return -1;
 815                 }
 816                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 817                         close(netpipe);
 818                         return -1;
 819                 }
 820         }
 821         close(netpipe);
 822         return 0;
 823 }
 824
 825 static int do_start(void *data)
 826 {
 827         struct lxc_list *iterator;
 828         struct lxc_handler *handler = data;
 829         int devnull_fd = -1, ret;
 830         char path[PATH_MAX];
 831
 832         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 833                 SYSERROR("Failed to set signal mask.");
 834                 return -1;
 835         }
 836
 837         /* This prctl must be before the synchro, so if the parent dies before
 838          * we set the parent death signal, we will detect its death with the
 839          * synchro right after, otherwise we have a window where the parent can
 840          * exit before we set the pdeath signal leading to a unsupervized
 841          * container.
 842          */
 843         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 844                 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL.");
 845                 return -1;
 846         }
 847
 848         lxc_sync_fini_parent(handler);
 849
 850         /* Don't leak the pinfd to the container. */
 851         if (handler->pinfd >= 0) {
 852                 close(handler->pinfd);
 853         }
 854
 855         if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP))
 856                 return -1;
 857
 858         /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
 859          * https://github.com/lxc/lxd/issues/1978.
 860          */
 861         if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) ==
 862             (CLONE_NEWNET | CLONE_NEWUSER)) {
 863                 ret = unshare(CLONE_NEWNET);
 864                 if (ret < 0) {
 865                         SYSERROR("Failed to unshare CLONE_NEWNET.");
 866                         goto out_warn_father;
 867                 }
 868                 INFO("Unshared CLONE_NEWNET.");
 869         }
 870
 871         /* Tell the parent task it can begin to configure the container and wait
 872          * for it to finish.
 873          */
 874         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 875                 return -1;
 876
 877         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 878                 goto out_warn_father;
 879
 880         /* If we are in a new user namespace, become root there to have
 881          * privilege over our namespace.
 882          */
 883         if (!lxc_list_empty(&handler->conf->id_map)) {
 884                 if (lxc_switch_uid_gid(0, 0) < 0)
 885                         goto out_warn_father;
 886
 887                 /* Drop groups only after we switched to a valid gid in the new
 888                  * user namespace.
 889                  */
 890                 if (lxc_setgroups(0, NULL) < 0)
 891                         goto out_warn_father;
 892         }
 893
 894         if (access(handler->lxcpath, X_OK)) {
 895                 print_top_failing_dir(handler->lxcpath);
 896                 goto out_warn_father;
 897         }
 898
 899         #if HAVE_LIBCAP
 900         if (handler->conf->need_utmp_watch) {
 901                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 902                         SYSERROR("Failed to remove the CAP_SYS_BOOT capability.");
 903                         goto out_warn_father;
 904                 }
 905                 DEBUG("Dropped the CAP_SYS_BOOT capability.");
 906         }
 907         #endif
 908
 909         ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount);
 910         if (ret < 0 || ret >= sizeof(path))
 911                 goto out_warn_father;
 912
 913         /* In order to checkpoint restore, we need to have everything in the
 914          * same mount namespace. However, some containers may not have a
 915          * reasonable /dev (in particular, they may not have /dev/null), so we
 916          * can't set init's std fds to /dev/null by opening it from inside the
 917          * container.
 918          *
 919          * If that's the case, fall back to using the host's /dev/null. This
 920          * means that migration won't work, but at least we won't spew output
 921          * where it isn't wanted.
 922          */
 923         if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) {
 924                 devnull_fd = open_devnull();
 925
 926                 if (devnull_fd < 0)
 927                         goto out_warn_father;
 928                 WARN("Using /dev/null from the host for container init's "
 929                      "standard file descriptors. Migration will not work.");
 930         }
 931
 932         /* Setup the container, ip, names, utsname, ... */
 933         if (lxc_setup(handler)) {
 934                 ERROR("Failed to setup container \"%s\".", handler->name);
 935                 goto out_warn_father;
 936         }
 937
 938         /* Ask father to setup cgroups and wait for him to finish. */
 939         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 940                 goto out_error;
 941
 942         /* Unshare cgroup namespace after we have setup our cgroups. If we do it
 943          * earlier we end up with a wrong view of /proc/self/cgroup. For
 944          * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
 945          * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
 946          * /proc/self/cgroup would show us:
 947          *
 948          *      8:cpuset:/lxc/c
 949          *
 950          * whereas it should actually show
 951          *
 952          *      8:cpuset:/
 953          */
 954         if (cgns_supported()) {
 955                 if (unshare(CLONE_NEWCGROUP) < 0) {
 956                         INFO("Failed to unshare CLONE_NEWCGROUP.");
 957                         goto out_warn_father;
 958                 }
 959                 INFO("Unshared CLONE_NEWCGROUP.");
 960         }
 961
 962         /* Set the label to change to when we exec(2) the container's init. */
 963         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 964                 goto out_warn_father;
 965
 966         /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
 967          * before we aren't allowed anymore.
 968          */
 969         if (handler->conf->no_new_privs) {
 970                 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
 971                         SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 972                         goto out_warn_father;
 973                 }
 974                 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 975         }
 976
 977         /* Some init's such as busybox will set sane tty settings on stdin,
 978          * stdout, stderr which it thinks is the console. We already set them
 979          * the way we wanted on the real terminal, and we want init to do its
 980          * setup on its console ie. the pty allocated in lxc_console_create() so
 981          * make sure that that pty is stdin,stdout,stderr.
 982          */
 983         if (lxc_console_set_stdfds(handler->conf->console.slave) < 0)
 984                 goto out_warn_father;
 985
 986         /* If we mounted a temporary proc, then unmount it now. */
 987         tmp_proc_unmount(handler->conf);
 988
 989         if (lxc_seccomp_load(handler->conf) != 0)
 990                 goto out_warn_father;
 991
 992         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 993                 ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name);
 994                 goto out_warn_father;
 995         }
 996
 997         /* The container has been setup. We can now switch to an unprivileged
 998          * uid/gid.
 999          */
1000         if (handler->conf->is_execute) {
1001                 bool have_cap_setgid;
1002                 uid_t new_uid = handler->conf->init_uid;
1003                 gid_t new_gid = handler->conf->init_gid;
1004
1005                 /* If we are in a new user namespace we already dropped all
1006                  * groups when we switched to root in the new user namespace
1007                  * further above. Only drop groups if we can, so ensure that we
1008                  * have necessary privilege.
1009                  */
1010                 #if HAVE_LIBCAP
1011                 have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE);
1012                 #else
1013                 have_cap_setgid = false;
1014                 #endif
1015                 if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) {
1016                         if (lxc_setgroups(0, NULL) < 0)
1017                                 goto out_warn_father;
1018                 }
1019
1020                 if (lxc_switch_uid_gid(new_uid, new_gid) < 0)
1021                         goto out_warn_father;
1022         }
1023
1024         /* The clearenv() and putenv() calls have been moved here to allow us to
1025          * use environment variables passed to the various hooks, such as the
1026          * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS
1027          * are valid in this context but others are.
1028          */
1029         if (clearenv()) {
1030                 SYSERROR("Failed to clear environment.");
1031                 /* Don't error out though. */
1032         }
1033
1034         lxc_list_for_each(iterator, &handler->conf->environment) {
1035                 if (putenv((char *)iterator->elem)) {
1036                         SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem);
1037                         goto out_warn_father;
1038                 }
1039         }
1040
1041         if (putenv("container=lxc")) {
1042                 SYSERROR("Failed to set environment variable: container=lxc.");
1043                 goto out_warn_father;
1044         }
1045
1046         if (handler->conf->pty_names) {
1047                 if (putenv(handler->conf->pty_names)) {
1048                         SYSERROR("Failed to set environment variable for container ptys.");
1049                         goto out_warn_father;
1050                 }
1051         }
1052
1053         close(handler->sigfd);
1054
1055         if (devnull_fd < 0) {
1056                 devnull_fd = open_devnull();
1057
1058                 if (devnull_fd < 0)
1059                         goto out_warn_father;
1060         }
1061
1062         if (handler->backgrounded && set_stdfds(devnull_fd))
1063                 goto out_warn_father;
1064
1065         if (devnull_fd >= 0) {
1066                 close(devnull_fd);
1067                 devnull_fd = -1;
1068         }
1069
1070         setsid();
1071
1072         /* After this call, we are in error because this ops should not return
1073          * as it execs.
1074          */
1075         handler->ops->start(handler, handler->data);
1076
1077 out_warn_father:
1078         /* We want the parent to know something went wrong, so we return a
1079          * special error code.
1080          */
1081         lxc_sync_wake_parent(handler, LXC_SYNC_ERROR);
1082
1083 out_error:
1084         if (devnull_fd >= 0)
1085                 close(devnull_fd);
1086
1087         return -1;
1088 }
1089
1090 static int save_phys_nics(struct lxc_conf *conf)
1091 {
1092         struct lxc_list *iterator;
1093         int am_root = (getuid() == 0);
1094
1095         if (!am_root)
1096                 return 0;
1097
1098         lxc_list_for_each(iterator, &conf->network) {
1099                 struct lxc_netdev *netdev = iterator->elem;
1100
1101                 if (netdev->type != LXC_NET_PHYS)
1102                         continue;
1103                 conf->saved_nics = realloc(conf->saved_nics,
1104                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
1105                 if (!conf->saved_nics)
1106                         return -1;
1107                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
1108                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
1109                 if (!conf->saved_nics[conf->num_savednics].orig_name)
1110                         return -1;
1111                 INFO("Stored saved_nic #%d idx %d name %s.", conf->num_savednics,
1112                         conf->saved_nics[conf->num_savednics].ifindex,
1113                         conf->saved_nics[conf->num_savednics].orig_name);
1114                 conf->num_savednics++;
1115         }
1116
1117         return 0;
1118 }
1119
1120 static int lxc_recv_ttys_from_child(struct lxc_handler *handler)
1121 {
1122         int i;
1123         int *ttyfds;
1124         struct lxc_pty_info *pty_info;
1125         int ret = -1;
1126         int sock = handler->ttysock[1];
1127         struct lxc_conf *conf = handler->conf;
1128         struct lxc_tty_info *tty_info = &conf->tty_info;
1129         size_t num_ttyfds = (2 * conf->tty);
1130
1131         if (!conf->tty)
1132                 return 0;
1133
1134         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty);
1135         if (!tty_info->pty_info)
1136                 return -1;
1137
1138         ttyfds = malloc(num_ttyfds * sizeof(int));
1139         if (!ttyfds)
1140                 return -1;
1141
1142         ret = lxc_abstract_unix_recv_fds(sock, ttyfds, num_ttyfds, NULL, 0);
1143         for (i = 0; (ret >= 0 && *ttyfds != -1) && (i < num_ttyfds); i++) {
1144                 pty_info = &tty_info->pty_info[i / 2];
1145                 pty_info->busy = 0;
1146                 pty_info->slave = ttyfds[i++];
1147                 pty_info->master = ttyfds[i];
1148                 TRACE("received pty with master fd %d and slave fd %d from "
1149                       "parent", pty_info->master, pty_info->slave);
1150         }
1151
1152         tty_info->nbtty = conf->tty;
1153
1154         free(ttyfds);
1155
1156         if (ret < 0)
1157                 ERROR("failed to receive %d ttys from child: %s", conf->tty,
1158                       strerror(errno));
1159         else
1160                 TRACE("received %d ttys from child", conf->tty);
1161
1162         return ret;
1163 }
1164
1165 void resolve_clone_flags(struct lxc_handler *handler)
1166 {
1167         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
1168
1169         if (!lxc_list_empty(&handler->conf->id_map))
1170                 handler->clone_flags |= CLONE_NEWUSER;
1171
1172         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
1173                 if (!lxc_requests_empty_network(handler))
1174                         handler->clone_flags |= CLONE_NEWNET;
1175         } else {
1176                 INFO("Inheriting a NET namespace.");
1177         }
1178
1179         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1)
1180                 handler->clone_flags |= CLONE_NEWIPC;
1181         else
1182                 INFO("Inheriting an IPC namespace.");
1183
1184         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1)
1185                 handler->clone_flags |= CLONE_NEWUTS;
1186         else
1187                 INFO("Inheriting a UTS namespace.");
1188 }
1189
1190 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1191  * exec()s the requested container binary.
1192  * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1193  * right here should be double checked if they'd pose a security risk. (For
1194  * example, any {u}mount() operations performed here will be reflected on the
1195  * host!)
1196  */
1197 static int lxc_spawn(struct lxc_handler *handler)
1198 {
1199         int failed_before_rename = 0;
1200         const char *name = handler->name;
1201         bool cgroups_connected = false;
1202         int saved_ns_fd[LXC_NS_MAX];
1203         int preserve_mask = 0, i, flags;
1204         int netpipepair[2], nveths;
1205         bool wants_to_map_ids;
1206         struct lxc_list *id_map;
1207
1208         netpipe = -1;
1209         id_map = &handler->conf->id_map;
1210         wants_to_map_ids = !lxc_list_empty(id_map);
1211
1212         for (i = 0; i < LXC_NS_MAX; i++)
1213                 if (handler->conf->inherit_ns_fd[i] != -1)
1214                         preserve_mask |= ns_info[i].clone_flag;
1215
1216         if (lxc_sync_init(handler))
1217                 return -1;
1218
1219         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
1220                 lxc_sync_fini(handler);
1221                 return -1;
1222         }
1223
1224         resolve_clone_flags(handler);
1225
1226         if (handler->clone_flags & CLONE_NEWNET) {
1227                 if (!lxc_list_empty(&handler->conf->network)) {
1228
1229                         /* Find gateway addresses from the link device, which is
1230                          * no longer accessible inside the container. Do this
1231                          * before creating network interfaces, since goto
1232                          * out_delete_net does not work before lxc_clone.
1233                          */
1234                         if (lxc_find_gateway_addresses(handler)) {
1235                                 ERROR("Failed to find gateway addresses.");
1236                                 lxc_sync_fini(handler);
1237                                 return -1;
1238                         }
1239
1240                         /* That should be done before the clone because we will
1241                          * fill the netdev index and use them in the child.
1242                          */
1243                         if (lxc_setup_networks_in_parent_namespaces(handler)) {
1244                                 ERROR("Failed to create the network.");
1245                                 lxc_sync_fini(handler);
1246                                 return -1;
1247                         }
1248                 }
1249
1250                 if (save_phys_nics(handler->conf)) {
1251                         ERROR("Failed to save physical nic info.");
1252                         goto out_abort;
1253                 }
1254         }
1255
1256         if (!cgroup_init(handler)) {
1257                 ERROR("Failed initializing cgroup support.");
1258                 goto out_delete_net;
1259         }
1260
1261         cgroups_connected = true;
1262
1263         if (!cgroup_create(handler)) {
1264                 ERROR("Failed creating cgroups.");
1265                 goto out_delete_net;
1266         }
1267
1268         /* If the rootfs is not a blockdev, prevent the container from marking
1269          * it readonly.
1270          * If the container is unprivileged then skip rootfs pinning.
1271          */
1272         if (!wants_to_map_ids) {
1273                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
1274                 if (handler->pinfd == -1)
1275                         INFO("Failed to pin the rootfs for container \"%s\".", handler->name);
1276         }
1277
1278         if (!preserve_ns(saved_ns_fd, preserve_mask, getpid()))
1279                 goto out_delete_net;
1280
1281         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
1282                 goto out_delete_net;
1283
1284         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
1285                 if (pipe(netpipepair) < 0) {
1286                         SYSERROR("Failed to create pipe.");
1287                         goto out_delete_net;
1288                 }
1289                 /* Store netpipe in the global var for do_start's use. */
1290                 netpipe = netpipepair[0];
1291         }
1292
1293         /* Create a process in a new set of namespaces. */
1294         flags = handler->clone_flags;
1295         if (handler->clone_flags & CLONE_NEWUSER) {
1296                 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1297                  * clone a new user namespace first and only later unshare our
1298                  * network namespace to ensure that network devices ownership is
1299                  * set up correctly.
1300                  */
1301                 flags &= ~CLONE_NEWNET;
1302         }
1303         handler->pid = lxc_clone(do_start, handler, flags);
1304         if (handler->pid < 0) {
1305                 SYSERROR("Failed to clone a new set of namespaces.");
1306                 goto out_delete_net;
1307         }
1308         for (i = 0; i < LXC_NS_MAX; i++)
1309                 if (flags & ns_info[i].clone_flag)
1310                         INFO("Cloned %s.", ns_info[i].flag_name);
1311
1312         if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid))
1313                 INFO("Failed to preserve namespace for lxc.hook.stop.");
1314
1315         if (attach_ns(saved_ns_fd))
1316                 WARN("Failed to restore saved namespaces.");
1317
1318         lxc_sync_fini_child(handler);
1319
1320         /* Map the container uids. The container became an invalid userid the
1321          * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1322          * anything immediately, but allows the container to setuid(0) (0 being
1323          * mapped to something else on the host.) later to become a valid uid
1324          * again.
1325          */
1326         if (wants_to_map_ids && lxc_map_ids(id_map, handler->pid)) {
1327                 ERROR("Failed to set up id mapping.");
1328                 goto out_delete_net;
1329         }
1330
1331         if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) {
1332                 failed_before_rename = 1;
1333                 goto out_delete_net;
1334         }
1335
1336         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) {
1337                 failed_before_rename = 1;
1338                 goto out_delete_net;
1339         }
1340
1341         if (!cgroup_create_legacy(handler)) {
1342                 ERROR("Failed to setup legacy cgroups for container \"%s\".", name);
1343                 goto out_delete_net;
1344         }
1345         if (!cgroup_setup_limits(handler, false)) {
1346                 ERROR("Failed to setup cgroup limits for container \"%s\".", name);
1347                 goto out_delete_net;
1348         }
1349
1350         if (!cgroup_enter(handler))
1351                 goto out_delete_net;
1352
1353         if (!cgroup_chown(handler))
1354                 goto out_delete_net;
1355
1356         if (failed_before_rename)
1357                 goto out_delete_net;
1358
1359         /* Create the network configuration. */
1360         if (handler->clone_flags & CLONE_NEWNET) {
1361                 if (lxc_assign_network(handler->lxcpath, handler->name,
1362                                        &handler->conf->network, handler->pid)) {
1363                         ERROR("Failed to create the configured network.");
1364                         goto out_delete_net;
1365                 }
1366         }
1367
1368         if (netpipe != -1) {
1369                 struct lxc_list *iterator;
1370                 struct lxc_netdev *netdev;
1371
1372                 close(netpipe);
1373                 lxc_list_for_each(iterator, &handler->conf->network) {
1374                         netdev = iterator->elem;
1375                         if (netdev->type != LXC_NET_VETH)
1376                                 continue;
1377                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1378                                 ERROR("Error writing veth name to container.");
1379                                 goto out_delete_net;
1380                         }
1381                 }
1382                 close(netpipepair[1]);
1383         }
1384
1385         /* Tell the child to continue its initialization. We'll get
1386          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1387          */
1388         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1389                 goto out_delete_net;
1390
1391         if (!lxc_list_empty(&handler->conf->limits) && setup_resource_limits(&handler->conf->limits, handler->pid)) {
1392                 ERROR("failed to setup resource limits for '%s'", name);
1393                 goto out_delete_net;
1394         }
1395
1396         if (!cgroup_setup_limits(handler, true)) {
1397                 ERROR("Failed to setup the devices cgroup for container \"%s\".", name);
1398                 goto out_delete_net;
1399         }
1400
1401         cgroup_disconnect();
1402         cgroups_connected = false;
1403
1404         /* Read tty fds allocated by child. */
1405         if (lxc_recv_ttys_from_child(handler) < 0) {
1406                 ERROR("Failed to receive tty info from child process.");
1407                 goto out_delete_net;
1408         }
1409
1410         /* Tell the child to complete its initialization and wait for it to exec
1411          * or return an error. (The child will never return
1412          * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing
1413          * lxc_sync_barrier_child to return success, or return a different
1414          * value, causing us to error out).
1415          */
1416         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1417                 return -1;
1418
1419         if (handler->ops->post_start(handler, handler->data))
1420                 goto out_abort;
1421
1422         if (lxc_set_state(name, handler, RUNNING)) {
1423                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name,
1424                       lxc_state2str(RUNNING));
1425                 goto out_abort;
1426         }
1427
1428         lxc_sync_fini(handler);
1429         handler->netnsfd = lxc_preserve_ns(handler->pid, "net");
1430
1431         return 0;
1432
1433 out_delete_net:
1434         if (cgroups_connected)
1435                 cgroup_disconnect();
1436         if (handler->clone_flags & CLONE_NEWNET)
1437                 lxc_delete_network(handler);
1438 out_abort:
1439         lxc_abort(name, handler);
1440         lxc_sync_fini(handler);
1441         if (handler->pinfd >= 0) {
1442                 close(handler->pinfd);
1443                 handler->pinfd = -1;
1444         }
1445
1446         return -1;
1447 }
1448
1449 int __lxc_start(const char *name, struct lxc_handler *handler,
1450                 struct lxc_operations* ops, void *data, const char *lxcpath,
1451                 bool backgrounded)
1452 {
1453         int status;
1454         int err = -1;
1455         bool removed_all_netdevs = true;
1456         struct lxc_conf *conf = handler->conf;
1457
1458         if (lxc_init(name, handler) < 0) {
1459                 ERROR("Failed to initialize container \"%s\".", name);
1460                 return -1;
1461         }
1462         handler->ops = ops;
1463         handler->data = data;
1464         handler->backgrounded = backgrounded;
1465         handler->netnsfd = -1;
1466
1467         if (must_drop_cap_sys_boot(handler->conf)) {
1468                 #if HAVE_LIBCAP
1469                 DEBUG("Dropping CAP_SYS_BOOT capability.");
1470                 #else
1471                 DEBUG("Not dropping CAP_SYS_BOOT capability as capabilities aren't supported.");
1472                 #endif
1473         } else {
1474                 DEBUG("Not dropping CAP_SYS_BOOT or watching utmp.");
1475                 handler->conf->need_utmp_watch = 0;
1476         }
1477
1478         if (!attach_block_device(handler->conf)) {
1479                 ERROR("Failed to attach block device.");
1480                 goto out_fini_nonet;
1481         }
1482
1483         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1484                 /* If the backing store is a device, mount it here and now. */
1485                 if (rootfs_is_blockdev(conf)) {
1486                         if (unshare(CLONE_NEWNS) < 0) {
1487                                 ERROR("Failed to unshare CLONE_NEWNS.");
1488                                 goto out_fini_nonet;
1489                         }
1490                         INFO("Unshared CLONE_NEWNS.");
1491
1492                         remount_all_slave();
1493                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1494                                 ERROR("Error setting up rootfs mount as root before spawn.");
1495                                 goto out_fini_nonet;
1496                         }
1497                         INFO("Set up container rootfs as host root.");
1498                 }
1499         }
1500
1501         err = lxc_spawn(handler);
1502         if (err) {
1503                 ERROR("Failed to spawn container \"%s\".", name);
1504                 goto out_detach_blockdev;
1505         }
1506
1507         handler->conf->reboot = 0;
1508
1509         err = lxc_poll(name, handler);
1510         if (err) {
1511                 ERROR("LXC mainloop exited with error: %d.", err);
1512                 if (handler->netnsfd >= 0) {
1513                         close(handler->netnsfd);
1514                         handler->netnsfd = -1;
1515                 }
1516                 goto out_abort;
1517         }
1518
1519         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1520                 continue;
1521
1522         /* If the child process exited but was not signaled, it didn't call
1523          * reboot. This should mean it was an lxc-execute which simply exited.
1524          * In any case, treat it as a 'halt'.
1525          */
1526         if (WIFSIGNALED(status)) {
1527                 switch(WTERMSIG(status)) {
1528                 case SIGINT: /* halt */
1529                         DEBUG("Container \"%s\" is halting.", name);
1530                         break;
1531                 case SIGHUP: /* reboot */
1532                         DEBUG("Container \"%s\" is rebooting.", name);
1533                         handler->conf->reboot = 1;
1534                         break;
1535                 case SIGSYS: /* seccomp */
1536                         DEBUG("Container \"%s\" violated its seccomp policy.", name);
1537                         break;
1538                 default:
1539                         DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status));
1540                         break;
1541                 }
1542         }
1543
1544         DEBUG("Pushing physical nics back to host namespace");
1545         lxc_restore_phys_nics_to_netns(handler->netnsfd, handler->conf);
1546
1547         DEBUG("Tearing down virtual network devices used by container \"%s\".", name);
1548         removed_all_netdevs = lxc_delete_network(handler);
1549
1550         if (handler->pinfd >= 0) {
1551                 close(handler->pinfd);
1552                 handler->pinfd = -1;
1553         }
1554
1555         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1556         err =  lxc_error_set_and_log(handler->pid, status);
1557 out_fini:
1558         if (!removed_all_netdevs) {
1559                 DEBUG("Failed tearing down network devices used by container. Trying again!");
1560                 removed_all_netdevs = lxc_delete_network(handler);
1561                 if (!removed_all_netdevs)
1562                         DEBUG("Failed tearing down network devices used by container. Not trying again!");
1563         }
1564
1565 out_detach_blockdev:
1566         detach_block_device(handler->conf);
1567
1568 out_fini_nonet:
1569         lxc_fini(name, handler);
1570         return err;
1571
1572 out_abort:
1573         lxc_abort(name, handler);
1574         goto out_fini;
1575 }
1576
1577 struct start_args {
1578         char *const *argv;
1579 };
1580
1581 static int start(struct lxc_handler *handler, void* data)
1582 {
1583         struct start_args *arg = data;
1584
1585         NOTICE("Exec'ing \"%s\".", arg->argv[0]);
1586
1587         execvp(arg->argv[0], arg->argv);
1588         SYSERROR("Failed to exec \"%s\".", arg->argv[0]);
1589         return 0;
1590 }
1591
1592 static int post_start(struct lxc_handler *handler, void* data)
1593 {
1594         struct start_args *arg = data;
1595
1596         NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid);
1597         return 0;
1598 }
1599
1600 static struct lxc_operations start_ops = {
1601         .start = start,
1602         .post_start = post_start
1603 };
1604
1605 int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler,
1606               const char *lxcpath, bool backgrounded)
1607 {
1608         struct start_args start_arg = {
1609                 .argv = argv,
1610         };
1611
1612         handler->conf->need_utmp_watch = 1;
1613         return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded);
1614 }
1615
1616 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1617                                             const char *name)
1618 {
1619         char destroy[MAXPATHLEN];
1620         bool bret = true;
1621         int ret = 0;
1622         struct lxc_container *c;
1623         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1624                 bret = do_destroy_container(handler->conf);
1625                 if (!bret) {
1626                         ERROR("Error destroying rootfs for container \"%s\".", name);
1627                         return;
1628                 }
1629         }
1630         INFO("Destroyed rootfs for container \"%s\".", name);
1631
1632         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1633         if (ret < 0 || ret >= MAXPATHLEN) {
1634                 ERROR("Error destroying directory for container \"%s\".", name);
1635                 return;
1636         }
1637
1638         c = lxc_container_new(name, handler->lxcpath);
1639         if (c) {
1640                 if (container_disk_lock(c)) {
1641                         INFO("Could not update lxc_snapshots file.");
1642                         lxc_container_put(c);
1643                 } else {
1644                         mod_all_rdeps(c, false);
1645                         container_disk_unlock(c);
1646                         lxc_container_put(c);
1647                 }
1648         }
1649
1650         if (am_unpriv())
1651                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper,
1652                                     destroy, "lxc_rmdir_onedev_wrapper");
1653         else
1654                 ret = lxc_rmdir_onedev(destroy, NULL);
1655
1656         if (ret < 0) {
1657                 ERROR("Error destroying directory for container \"%s\".", name);
1658                 return;
1659         }
1660         INFO("Destroyed directory for container \"%s\".", name);
1661 }
1662
1663 static int lxc_rmdir_onedev_wrapper(void *data)
1664 {
1665         char *arg = (char *) data;
1666         return lxc_rmdir_onedev(arg, NULL);
1667 }
1668
1669 static bool do_destroy_container(struct lxc_conf *conf) {
1670         if (am_unpriv()) {
1671                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf,
1672                                   "bdev_destroy_wrapper") < 0)
1673                         return false;
1674                 return true;
1675         }
1676         return bdev_destroy(conf);
1677 }