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_SYS_CAPABILITY_H
  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 const struct ns_info ns_info[LXC_NS_MAX] = {
  88         [LXC_NS_MNT] = {"mnt", CLONE_NEWNS},
  89         [LXC_NS_PID] = {"pid", CLONE_NEWPID},
  90         [LXC_NS_UTS] = {"uts", CLONE_NEWUTS},
  91         [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC},
  92         [LXC_NS_USER] = {"user", CLONE_NEWUSER},
  93         [LXC_NS_NET] = {"net", CLONE_NEWNET},
  94         [LXC_NS_CGROUP] = {"cgroup", CLONE_NEWCGROUP}
  95 };
  96
  97 extern void mod_all_rdeps(struct lxc_container *c, bool inc);
  98 static bool do_destroy_container(struct lxc_conf *conf);
  99 static int lxc_rmdir_onedev_wrapper(void *data);
 100 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
 101                                             const char *name);
 102
 103 static void print_top_failing_dir(const char *path)
 104 {
 105         size_t len = strlen(path);
 106         char *copy = alloca(len + 1), *p, *e, saved;
 107         strcpy(copy, path);
 108
 109         p = copy;
 110         e = copy + len;
 111         while (p < e) {
 112                 while (p < e && *p == '/')
 113                         p++;
 114                 while (p < e && *p != '/')
 115                         p++;
 116                 saved = *p;
 117                 *p = '\0';
 118                 if (access(copy, X_OK)) {
 119                         SYSERROR("could not access %s.  Please grant it 'x' " \
 120                               "access, or add an ACL for the container root.",
 121                               copy);
 122                         return;
 123                 }
 124                 *p = saved;
 125         }
 126 }
 127
 128 static void close_ns(int ns_fd[LXC_NS_MAX]) {
 129         int i;
 130
 131         for (i = 0; i < LXC_NS_MAX; i++) {
 132                 if (ns_fd[i] > -1) {
 133                         close(ns_fd[i]);
 134                         ns_fd[i] = -1;
 135                 }
 136         }
 137 }
 138
 139 /*
 140  * preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
 141  * in clone_flags.
 142  * Return true on success, false on failure.  On failure, leave an error
 143  * message in *errmsg, which caller must free.
 144  */
 145 static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid,
 146                         char **errmsg)
 147 {
 148         int i, ret;
 149         char path[MAXPATHLEN];
 150
 151         for (i = 0; i < LXC_NS_MAX; i++)
 152                 ns_fd[i] = -1;
 153
 154         snprintf(path, MAXPATHLEN, "/proc/%d/ns", pid);
 155         if (access(path, X_OK)) {
 156                 if (asprintf(errmsg, "Kernel does not support setns.") == -1)
 157                         *errmsg = NULL;
 158                 return false;
 159         }
 160
 161         for (i = 0; i < LXC_NS_MAX; i++) {
 162                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 163                         continue;
 164                 snprintf(path, MAXPATHLEN, "/proc/%d/ns/%s", pid,
 165                          ns_info[i].proc_name);
 166                 ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC);
 167                 if (ns_fd[i] < 0)
 168                         goto error;
 169         }
 170
 171         return true;
 172
 173 error:
 174         if (errno == ENOENT) {
 175                 ret = asprintf(errmsg, "Kernel does not support setns for %s",
 176                         ns_info[i].proc_name);
 177         } else {
 178                 ret = asprintf(errmsg, "Failed to open %s: %s",
 179                         path, strerror(errno));
 180         }
 181         if (ret == -1)
 182                 *errmsg = NULL;
 183         close_ns(ns_fd);
 184         return false;
 185 }
 186
 187 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 188         int i;
 189
 190         for (i = 0; i < LXC_NS_MAX; i++) {
 191                 if (ns_fd[i] < 0)
 192                         continue;
 193
 194                 if (setns(ns_fd[i], 0) != 0)
 195                         goto error;
 196         }
 197         return 0;
 198
 199 error:
 200         SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name);
 201         return -1;
 202 }
 203
 204 static int match_fd(int fd)
 205 {
 206         return (fd == 0 || fd == 1 || fd == 2);
 207 }
 208
 209 /*
 210  * Check for any fds we need to close
 211  * * if fd_to_ignore != -1, then if we find that fd open we will ignore it.
 212  * * By default we warn about open fds we find.
 213  * * If closeall is true, we will close open fds.
 214  * * If lxc-start was passed "-C", then conf->close_all_fds will be true,
 215  *     in which case we also close all open fds.
 216  * * A daemonized container will always pass closeall=true.
 217  */
 218 int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore)
 219 {
 220         struct dirent *direntp;
 221         int fd, fddir;
 222         DIR *dir;
 223
 224         if (conf && conf->close_all_fds)
 225                 closeall = true;
 226
 227 restart:
 228         dir = opendir("/proc/self/fd");
 229         if (!dir) {
 230                 WARN("failed to open directory: %m");
 231                 return -1;
 232         }
 233
 234         fddir = dirfd(dir);
 235
 236         while ((direntp = readdir(dir))) {
 237                 if (!direntp)
 238                         break;
 239
 240                 if (!strcmp(direntp->d_name, "."))
 241                         continue;
 242
 243                 if (!strcmp(direntp->d_name, ".."))
 244                         continue;
 245
 246                 fd = atoi(direntp->d_name);
 247
 248                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 249                         continue;
 250
 251                 if (current_config && fd == current_config->logfd)
 252                         continue;
 253
 254                 if (match_fd(fd))
 255                         continue;
 256
 257                 if (closeall) {
 258                         close(fd);
 259                         closedir(dir);
 260                         INFO("closed inherited fd %d", fd);
 261                         goto restart;
 262                 }
 263                 WARN("inherited fd %d", fd);
 264         }
 265
 266         /*
 267          * only enable syslog at this point to avoid the above logging function
 268          * to open a new fd and make the check_inherited function enter an
 269          * infinite loop.
 270          */
 271         lxc_log_enable_syslog();
 272
 273         closedir(dir); /* cannot fail */
 274         return 0;
 275 }
 276
 277 static int setup_signal_fd(sigset_t *oldmask)
 278 {
 279         sigset_t mask;
 280         int fd;
 281
 282         /* Block everything except serious error signals */
 283         if (sigfillset(&mask) ||
 284             sigdelset(&mask, SIGILL) ||
 285             sigdelset(&mask, SIGSEGV) ||
 286             sigdelset(&mask, SIGBUS) ||
 287             sigdelset(&mask, SIGWINCH) ||
 288             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 289                 SYSERROR("failed to set signal mask");
 290                 return -1;
 291         }
 292
 293         fd = signalfd(-1, &mask, 0);
 294         if (fd < 0) {
 295                 SYSERROR("failed to create the signal fd");
 296                 return -1;
 297         }
 298
 299         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 300                 SYSERROR("failed to set sigfd to close-on-exec");
 301                 close(fd);
 302                 return -1;
 303         }
 304
 305         DEBUG("sigchild handler set");
 306
 307         return fd;
 308 }
 309
 310 static int signal_handler(int fd, uint32_t events, void *data,
 311                           struct lxc_epoll_descr *descr)
 312 {
 313         struct signalfd_siginfo siginfo;
 314         siginfo_t info;
 315         int ret;
 316         pid_t *pid = data;
 317         bool init_died = false;
 318
 319         ret = read(fd, &siginfo, sizeof(siginfo));
 320         if (ret < 0) {
 321                 ERROR("failed to read signal info");
 322                 return -1;
 323         }
 324
 325         if (ret != sizeof(siginfo)) {
 326                 ERROR("unexpected siginfo size");
 327                 return -1;
 328         }
 329
 330         // check whether init is running
 331         info.si_pid = 0;
 332         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 333         if (ret == 0 && info.si_pid == *pid) {
 334                 init_died = true;
 335         }
 336
 337         if (siginfo.ssi_signo != SIGCHLD) {
 338                 kill(*pid, siginfo.ssi_signo);
 339                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 340                 return init_died ? 1 : 0;
 341         }
 342
 343         if (siginfo.ssi_code == CLD_STOPPED ||
 344             siginfo.ssi_code == CLD_CONTINUED) {
 345                 INFO("container init process was stopped/continued");
 346                 return init_died ? 1 : 0;
 347         }
 348
 349         /* more robustness, protect ourself from a SIGCHLD sent
 350          * by a process different from the container init
 351          */
 352         if (siginfo.ssi_pid != *pid) {
 353                 WARN("invalid pid for SIGCHLD");
 354                 return init_died ? 1 : 0;
 355         }
 356
 357         DEBUG("container init process exited");
 358         return 1;
 359 }
 360
 361 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 362 {
 363         handler->state = state;
 364         lxc_monitor_send_state(name, state, handler->lxcpath);
 365         return 0;
 366 }
 367
 368 int lxc_poll(const char *name, struct lxc_handler *handler)
 369 {
 370         int sigfd = handler->sigfd;
 371         int pid = handler->pid;
 372         struct lxc_epoll_descr descr;
 373
 374         if (lxc_mainloop_open(&descr)) {
 375                 ERROR("failed to create mainloop");
 376                 goto out_sigfd;
 377         }
 378
 379         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 380                 ERROR("failed to add handler for the signal");
 381                 goto out_mainloop_open;
 382         }
 383
 384         if (lxc_console_mainloop_add(&descr, handler->conf)) {
 385                 ERROR("failed to add console handler to mainloop");
 386                 goto out_mainloop_open;
 387         }
 388
 389         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 390                 ERROR("failed to add command handler to mainloop");
 391                 goto out_mainloop_open;
 392         }
 393
 394         if (handler->conf->need_utmp_watch) {
 395                 #if HAVE_SYS_CAPABILITY_H
 396                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 397                         ERROR("failed to add utmp handler to mainloop");
 398                         goto out_mainloop_open;
 399                 }
 400                 #else
 401                         DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
 402                 #endif
 403         }
 404
 405         return lxc_mainloop(&descr, -1);
 406
 407 out_mainloop_open:
 408         lxc_mainloop_close(&descr);
 409 out_sigfd:
 410         close(sigfd);
 411         return -1;
 412 }
 413
 414 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath)
 415 {
 416         int i;
 417         struct lxc_handler *handler;
 418
 419         handler = malloc(sizeof(*handler));
 420         if (!handler)
 421                 return NULL;
 422
 423         memset(handler, 0, sizeof(*handler));
 424
 425         handler->ttysock[0] = handler->ttysock[1] = -1;
 426         handler->conf = conf;
 427         handler->lxcpath = lxcpath;
 428         handler->pinfd = -1;
 429
 430         for (i = 0; i < LXC_NS_MAX; i++)
 431                 handler->nsfd[i] = -1;
 432
 433         lsm_init();
 434
 435         handler->name = strdup(name);
 436         if (!handler->name) {
 437                 ERROR("failed to allocate memory");
 438                 goto out_free;
 439         }
 440
 441         if (lxc_cmd_init(name, handler, lxcpath))
 442                 goto out_free_name;
 443
 444         if (lxc_read_seccomp_config(conf) != 0) {
 445                 ERROR("failed loading seccomp policy");
 446                 goto out_close_maincmd_fd;
 447         }
 448
 449         /* Begin by setting the state to STARTING */
 450         if (lxc_set_state(name, handler, STARTING)) {
 451                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 452                 goto out_close_maincmd_fd;
 453         }
 454
 455         /* Start of environment variable setup for hooks */
 456         if (name && setenv("LXC_NAME", name, 1)) {
 457                 SYSERROR("failed to set environment variable for container name");
 458         }
 459         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) {
 460                 SYSERROR("failed to set environment variable for config path");
 461         }
 462         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) {
 463                 SYSERROR("failed to set environment variable for rootfs mount");
 464         }
 465         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) {
 466                 SYSERROR("failed to set environment variable for rootfs mount");
 467         }
 468         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) {
 469                 SYSERROR("failed to set environment variable for console path");
 470         }
 471         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 472                 SYSERROR("failed to set environment variable for console log");
 473         }
 474         if (setenv("LXC_CGNS_AWARE", "1", 1)) {
 475                 SYSERROR("failed to set LXC_CGNS_AWARE environment variable");
 476         }
 477         /* End of environment variable setup for hooks */
 478
 479         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 480                 ERROR("failed to run pre-start hooks for container '%s'.", name);
 481                 goto out_aborting;
 482         }
 483
 484         /* the signal fd has to be created before forking otherwise
 485          * if the child process exits before we setup the signal fd,
 486          * the event will be lost and the command will be stuck */
 487         handler->sigfd = setup_signal_fd(&handler->oldmask);
 488         if (handler->sigfd < 0) {
 489                 ERROR("failed to set sigchild fd handler");
 490                 goto out_delete_tty;
 491         }
 492
 493         /* do this after setting up signals since it might unblock SIGWINCH */
 494         if (lxc_console_create(conf)) {
 495                 ERROR("failed to create console");
 496                 goto out_restore_sigmask;
 497         }
 498
 499         if (ttys_shift_ids(conf) < 0) {
 500                 ERROR("Failed to shift tty into container");
 501                 goto out_restore_sigmask;
 502         }
 503
 504         INFO("'%s' is initialized", name);
 505         return handler;
 506
 507 out_restore_sigmask:
 508         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 509 out_delete_tty:
 510         lxc_delete_tty(&conf->tty_info);
 511 out_aborting:
 512         lxc_set_state(name, handler, ABORTING);
 513 out_close_maincmd_fd:
 514         close(conf->maincmd_fd);
 515         conf->maincmd_fd = -1;
 516 out_free_name:
 517         free(handler->name);
 518         handler->name = NULL;
 519 out_free:
 520         free(handler);
 521         return NULL;
 522 }
 523
 524 void lxc_fini(const char *name, struct lxc_handler *handler)
 525 {
 526         int i, rc;
 527         pid_t self = getpid();
 528         char *namespaces[LXC_NS_MAX+1];
 529         size_t namespace_count = 0;
 530
 531         /* The STOPPING state is there for future cleanup code
 532          * which can take awhile
 533          */
 534         lxc_set_state(name, handler, STOPPING);
 535
 536         for (i = 0; i < LXC_NS_MAX; i++) {
 537                 if (handler->nsfd[i] != -1) {
 538                         rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d",
 539                                       ns_info[i].proc_name, self, handler->nsfd[i]);
 540                         if (rc == -1) {
 541                                 SYSERROR("failed to allocate memory");
 542                                 break;
 543                         }
 544                         ++namespace_count;
 545                 }
 546         }
 547         namespaces[namespace_count] = NULL;
 548
 549         if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) {
 550                 SYSERROR("failed to set environment variable for stop target");
 551         }
 552         if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) {
 553                 SYSERROR("failed to set environment variable for stop target");
 554         }
 555
 556         if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces))
 557                 ERROR("failed to run stop hooks for container '%s'.", name);
 558
 559         while (namespace_count--)
 560                 free(namespaces[namespace_count]);
 561         for (i = 0; i < LXC_NS_MAX; i++) {
 562                 if (handler->nsfd[i] != -1) {
 563                         close(handler->nsfd[i]);
 564                         handler->nsfd[i] = -1;
 565                 }
 566         }
 567         lxc_set_state(name, handler, STOPPED);
 568
 569         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) {
 570                 ERROR("failed to run post-stop hooks for container '%s'.", name);
 571                 if (handler->conf->reboot) {
 572                         WARN("Container will be stopped instead of rebooted.");
 573                         handler->conf->reboot = 0;
 574                         setenv("LXC_TARGET", "stop", 1);
 575                 }
 576         }
 577
 578         /* reset mask set by setup_signal_fd */
 579         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 580                 WARN("failed to restore sigprocmask");
 581
 582         lxc_console_delete(&handler->conf->console);
 583         lxc_delete_tty(&handler->conf->tty_info);
 584         close(handler->conf->maincmd_fd);
 585         handler->conf->maincmd_fd = -1;
 586         free(handler->name);
 587         if (handler->ttysock[0] != -1) {
 588                 close(handler->ttysock[0]);
 589                 close(handler->ttysock[1]);
 590         }
 591
 592         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1)
 593                 lxc_destroy_container_on_signal(handler, name);
 594
 595         cgroup_destroy(handler);
 596         free(handler);
 597 }
 598
 599 void lxc_abort(const char *name, struct lxc_handler *handler)
 600 {
 601         int ret, status;
 602
 603         lxc_set_state(name, handler, ABORTING);
 604         if (handler->pid > 0)
 605                 kill(handler->pid, SIGKILL);
 606         while ((ret = waitpid(-1, &status, 0)) > 0) ;
 607 }
 608
 609 #include <sys/reboot.h>
 610 #include <linux/reboot.h>
 611
 612 /*
 613  * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
 614  * in a child pid namespace if container reboot support exists.
 615  * Otherwise, it will either succeed or return -EPERM.
 616  */
 617 static int container_reboot_supported(void *arg)
 618 {
 619         int *cmd = arg;
 620         int ret;
 621
 622         ret = reboot(*cmd);
 623         if (ret == -1 && errno == EINVAL)
 624                 return 1;
 625         return 0;
 626 }
 627
 628 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 629 {
 630         FILE *f;
 631         int ret, cmd, v, flags;
 632         long stack_size = 4096;
 633         void *stack = alloca(stack_size);
 634         int status;
 635         pid_t pid;
 636
 637         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 638         if (!f) {
 639                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 640                 return 1;
 641         }
 642
 643         ret = fscanf(f, "%d", &v);
 644         fclose(f);
 645         if (ret != 1) {
 646                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
 647                 return 1;
 648         }
 649         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 650
 651         flags = CLONE_NEWPID | SIGCHLD;
 652         if (!lxc_list_empty(&conf->id_map))
 653                 flags |= CLONE_NEWUSER;
 654
 655 #ifdef __ia64__
 656         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 657 #else
 658         stack += stack_size;
 659         pid = clone(container_reboot_supported, stack, flags, &cmd);
 660 #endif
 661         if (pid < 0) {
 662                 if (flags & CLONE_NEWUSER)
 663                         ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno));
 664                 else
 665                         ERROR("failed to clone (%#x): %s", flags, strerror(errno));
 666                 return -1;
 667         }
 668         if (wait(&status) < 0) {
 669                 SYSERROR("unexpected wait error: %m");
 670                 return -1;
 671         }
 672
 673         if (WEXITSTATUS(status) != 1)
 674                 return 1;
 675
 676         return 0;
 677 }
 678
 679 /*
 680  * netpipe is used in the unprivileged case to transfer the ifindexes
 681  * from parent to child
 682  */
 683 static int netpipe = -1;
 684
 685 static inline int count_veths(struct lxc_list *network)
 686 {
 687         struct lxc_list *iterator;
 688         struct lxc_netdev *netdev;
 689         int count = 0;
 690
 691         lxc_list_for_each(iterator, network) {
 692                 netdev = iterator->elem;
 693                 if (netdev->type != LXC_NET_VETH)
 694                         continue;
 695                 count++;
 696         }
 697         return count;
 698 }
 699
 700 static int read_unpriv_netifindex(struct lxc_list *network)
 701 {
 702         struct lxc_list *iterator;
 703         struct lxc_netdev *netdev;
 704
 705         if (netpipe == -1)
 706                 return 0;
 707         lxc_list_for_each(iterator, network) {
 708                 netdev = iterator->elem;
 709                 if (netdev->type != LXC_NET_VETH)
 710                         continue;
 711                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 712                         ERROR("Out of memory");
 713                         close(netpipe);
 714                         return -1;
 715                 }
 716                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 717                         close(netpipe);
 718                         return -1;
 719                 }
 720         }
 721         close(netpipe);
 722         return 0;
 723 }
 724
 725 static int do_start(void *data)
 726 {
 727         struct lxc_list *iterator;
 728         struct lxc_handler *handler = data;
 729         int devnull_fd = -1, ret;
 730         char path[PATH_MAX];
 731
 732         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 733                 SYSERROR("failed to set sigprocmask");
 734                 return -1;
 735         }
 736
 737         /* This prctl must be before the synchro, so if the parent
 738          * dies before we set the parent death signal, we will detect
 739          * its death with the synchro right after, otherwise we have
 740          * a window where the parent can exit before we set the pdeath
 741          * signal leading to a unsupervized container.
 742          */
 743         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 744                 SYSERROR("failed to set pdeath signal");
 745                 return -1;
 746         }
 747
 748         lxc_sync_fini_parent(handler);
 749
 750         /* don't leak the pinfd to the container */
 751         if (handler->pinfd >= 0) {
 752                 close(handler->pinfd);
 753         }
 754
 755         if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP))
 756                 return -1;
 757
 758         /* Unshare CLONE_NEWNET after CLONE_NEWUSER  - see
 759           https://github.com/lxc/lxd/issues/1978 */
 760         if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) ==
 761                         (CLONE_NEWNET | CLONE_NEWUSER)) {
 762                 ret = unshare(CLONE_NEWNET);
 763                 if (ret < 0) {
 764                         SYSERROR("Error unsharing network namespace");
 765                         goto out_warn_father;
 766                 }
 767         }
 768
 769         /* Tell the parent task it can begin to configure the
 770          * container and wait for it to finish
 771          */
 772         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 773                 return -1;
 774
 775         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 776                 goto out_warn_father;
 777
 778         /*
 779          * if we are in a new user namespace, become root there to have
 780          * privilege over our namespace. When using lxc-execute we default to root,
 781          * but this can be overriden using the lxc.init_uid and lxc.init_gid
 782          * configuration options.
 783          */
 784         if (!lxc_list_empty(&handler->conf->id_map)) {
 785                 gid_t new_gid = 0;
 786                 if (handler->conf->is_execute && handler->conf->init_gid)
 787                         new_gid = handler->conf->init_gid;
 788
 789                 uid_t new_uid = 0;
 790                 if (handler->conf->is_execute && handler->conf->init_uid)
 791                         new_uid = handler->conf->init_uid;
 792
 793                 NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid);
 794                 if (setgid(new_gid)) {
 795                         SYSERROR("setgid");
 796                         goto out_warn_father;
 797                 }
 798                 if (setuid(new_uid)) {
 799                         SYSERROR("setuid");
 800                         goto out_warn_father;
 801                 }
 802                 if (setgroups(0, NULL)) {
 803                         SYSERROR("setgroups");
 804                         goto out_warn_father;
 805                 }
 806         }
 807
 808         if (access(handler->lxcpath, X_OK)) {
 809                 print_top_failing_dir(handler->lxcpath);
 810                 goto out_warn_father;
 811         }
 812
 813         #if HAVE_SYS_CAPABILITY_H
 814         if (handler->conf->need_utmp_watch) {
 815                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 816                         SYSERROR("failed to remove CAP_SYS_BOOT capability");
 817                         goto out_warn_father;
 818                 }
 819                 DEBUG("Dropped cap_sys_boot");
 820         }
 821         #endif
 822
 823         ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount);
 824         if (ret < 0 || ret >= sizeof(path)) {
 825                 SYSERROR("sprintf'd too many chars");
 826                 goto out_warn_father;
 827         }
 828
 829         /* In order to checkpoint restore, we need to have everything in the
 830          * same mount namespace. However, some containers may not have a
 831          * reasonable /dev (in particular, they may not have /dev/null), so we
 832          * can't set init's std fds to /dev/null by opening it from inside the
 833          * container.
 834          *
 835          * If that's the case, fall back to using the host's /dev/null. This
 836          * means that migration won't work, but at least we won't spew output
 837          * where it isn't wanted.
 838          */
 839         if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) {
 840                 devnull_fd = open_devnull();
 841
 842                 if (devnull_fd < 0)
 843                         goto out_warn_father;
 844                 WARN("using host's /dev/null for container init's std fds, migraiton won't work");
 845         }
 846
 847         /* Setup the container, ip, names, utsname, ... */
 848         if (lxc_setup(handler)) {
 849                 ERROR("failed to setup the container");
 850                 goto out_warn_father;
 851         }
 852
 853         /* ask father to setup cgroups and wait for him to finish */
 854         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 855                 goto out_error;
 856
 857         /* Set the label to change to when we exec(2) the container's init */
 858         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 859                 goto out_warn_father;
 860
 861         /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
 862          * before we aren't allowed anymore. */
 863         if (handler->conf->no_new_privs) {
 864                 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
 865                         SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 866                         goto out_warn_father;
 867                 }
 868                 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 869         }
 870
 871         /* Some init's such as busybox will set sane tty settings on stdin,
 872          * stdout, stderr which it thinks is the console. We already set them
 873          * the way we wanted on the real terminal, and we want init to do its
 874          * setup on its console ie. the pty allocated in lxc_console_create()
 875          * so make sure that that pty is stdin,stdout,stderr.
 876          */
 877         if (lxc_console_set_stdfds(handler->conf->console.slave) < 0)
 878                 goto out_warn_father;
 879
 880         /* If we mounted a temporary proc, then unmount it now */
 881         tmp_proc_unmount(handler->conf);
 882
 883         if (lxc_seccomp_load(handler->conf) != 0)
 884                 goto out_warn_father;
 885
 886         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 887                 ERROR("failed to run start hooks for container '%s'.", handler->name);
 888                 goto out_warn_father;
 889         }
 890
 891         /* The clearenv() and putenv() calls have been moved here
 892          * to allow us to use environment variables passed to the various
 893          * hooks, such as the start hook above.  Not all of the
 894          * variables like CONFIG_PATH or ROOTFS are valid in this
 895          * context but others are. */
 896         if (clearenv()) {
 897                 SYSERROR("failed to clear environment");
 898                 /* don't error out though */
 899         }
 900
 901         lxc_list_for_each(iterator, &handler->conf->environment) {
 902                 if (putenv((char *)iterator->elem)) {
 903                         SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem);
 904                         goto out_warn_father;
 905                 }
 906         }
 907
 908         if (putenv("container=lxc")) {
 909                 SYSERROR("failed to set environment variable 'container=lxc'");
 910                 goto out_warn_father;
 911         }
 912
 913         if (handler->conf->pty_names) {
 914                 if (putenv(handler->conf->pty_names)) {
 915                         SYSERROR("failed to set environment variable for container ptys");
 916                         goto out_warn_father;
 917                 }
 918         }
 919
 920         close(handler->sigfd);
 921
 922         if (devnull_fd < 0) {
 923                 devnull_fd = open_devnull();
 924
 925                 if (devnull_fd < 0)
 926                         goto out_warn_father;
 927         }
 928
 929         if (handler->backgrounded && set_stdfds(devnull_fd))
 930                 goto out_warn_father;
 931
 932         if (devnull_fd >= 0) {
 933                 close(devnull_fd);
 934                 devnull_fd = -1;
 935         }
 936
 937         if (cgns_supported() && unshare(CLONE_NEWCGROUP) != 0) {
 938                 SYSERROR("Failed to unshare cgroup namespace");
 939                 goto out_warn_father;
 940         }
 941
 942         setsid();
 943
 944         /* after this call, we are in error because this
 945          * ops should not return as it execs */
 946         handler->ops->start(handler, handler->data);
 947
 948 out_warn_father:
 949         /* we want the parent to know something went wrong, so we return a special
 950          * error code. */
 951         lxc_sync_wake_parent(handler, LXC_SYNC_ERROR);
 952
 953 out_error:
 954         if (devnull_fd >= 0)
 955                 close(devnull_fd);
 956
 957         return -1;
 958 }
 959
 960 static int save_phys_nics(struct lxc_conf *conf)
 961 {
 962         struct lxc_list *iterator;
 963         int am_root = (getuid() == 0);
 964
 965         if (!am_root)
 966                 return 0;
 967
 968         lxc_list_for_each(iterator, &conf->network) {
 969                 struct lxc_netdev *netdev = iterator->elem;
 970
 971                 if (netdev->type != LXC_NET_PHYS)
 972                         continue;
 973                 conf->saved_nics = realloc(conf->saved_nics,
 974                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
 975                 if (!conf->saved_nics) {
 976                         SYSERROR("failed to allocate memory");
 977                         return -1;
 978                 }
 979                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
 980                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
 981                 if (!conf->saved_nics[conf->num_savednics].orig_name) {
 982                         SYSERROR("failed to allocate memory");
 983                         return -1;
 984                 }
 985                 INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics,
 986                         conf->saved_nics[conf->num_savednics].ifindex,
 987                         conf->saved_nics[conf->num_savednics].orig_name);
 988                 conf->num_savednics++;
 989         }
 990
 991         return 0;
 992 }
 993
 994 static int recv_fd(int sock, int *fd)
 995 {
 996         if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) {
 997                 SYSERROR("Error receiving tty fd from child");
 998                 return -1;
 999         }
1000         if (*fd == -1)
1001                 return -1;
1002         return 0;
1003 }
1004
1005 static int recv_ttys_from_child(struct lxc_handler *handler)
1006 {
1007         struct lxc_conf *conf = handler->conf;
1008         int i, sock = handler->ttysock[1];
1009         struct lxc_tty_info *tty_info = &conf->tty_info;
1010
1011         if (!conf->tty)
1012                 return 0;
1013
1014         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty);
1015         if (!tty_info->pty_info) {
1016                 SYSERROR("failed to allocate pty_info");
1017                 return -1;
1018         }
1019
1020         for (i = 0; i < conf->tty; i++) {
1021                 struct lxc_pty_info *pty_info = &tty_info->pty_info[i];
1022                 pty_info->busy = 0;
1023                 if (recv_fd(sock, &pty_info->slave) < 0 ||
1024                                 recv_fd(sock, &pty_info->master) < 0) {
1025                         ERROR("Error receiving tty info from child");
1026                         return -1;
1027                 }
1028         }
1029         tty_info->nbtty = conf->tty;
1030
1031         return 0;
1032 }
1033
1034 void resolve_clone_flags(struct lxc_handler *handler)
1035 {
1036         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
1037
1038         if (!lxc_list_empty(&handler->conf->id_map)) {
1039                 INFO("Cloning a new user namespace");
1040                 handler->clone_flags |= CLONE_NEWUSER;
1041         }
1042
1043         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
1044                 if (!lxc_requests_empty_network(handler))
1045                         handler->clone_flags |= CLONE_NEWNET;
1046         } else {
1047                 INFO("Inheriting a net namespace");
1048         }
1049
1050         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) {
1051                 handler->clone_flags |= CLONE_NEWIPC;
1052         } else {
1053                 INFO("Inheriting an IPC namespace");
1054         }
1055
1056         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
1057                 handler->clone_flags |= CLONE_NEWUTS;
1058         } else {
1059                 INFO("Inheriting a UTS namespace");
1060         }
1061 }
1062
1063 static int lxc_spawn(struct lxc_handler *handler)
1064 {
1065         int failed_before_rename = 0;
1066         const char *name = handler->name;
1067         char *errmsg = NULL;
1068         bool cgroups_connected = false;
1069         int saved_ns_fd[LXC_NS_MAX];
1070         int preserve_mask = 0, i, flags;
1071         int netpipepair[2], nveths;
1072
1073         netpipe = -1;
1074
1075         for (i = 0; i < LXC_NS_MAX; i++)
1076                 if (handler->conf->inherit_ns_fd[i] != -1)
1077                         preserve_mask |= ns_info[i].clone_flag;
1078
1079         if (lxc_sync_init(handler))
1080                 return -1;
1081
1082         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
1083                 lxc_sync_fini(handler);
1084                 return -1;
1085         }
1086
1087         resolve_clone_flags(handler);
1088
1089         if (handler->clone_flags & CLONE_NEWNET) {
1090                 if (!lxc_list_empty(&handler->conf->network)) {
1091
1092                         /* Find gateway addresses from the link device, which is
1093                          * no longer accessible inside the container. Do this
1094                          * before creating network interfaces, since goto
1095                          * out_delete_net does not work before lxc_clone. */
1096                         if (lxc_find_gateway_addresses(handler)) {
1097                                 ERROR("failed to find gateway addresses");
1098                                 lxc_sync_fini(handler);
1099                                 return -1;
1100                         }
1101
1102                         /* that should be done before the clone because we will
1103                          * fill the netdev index and use them in the child
1104                          */
1105                         if (lxc_create_network(handler)) {
1106                                 ERROR("failed to create the network");
1107                                 lxc_sync_fini(handler);
1108                                 return -1;
1109                         }
1110                 }
1111
1112                 if (save_phys_nics(handler->conf)) {
1113                         ERROR("failed to save physical nic info");
1114                         goto out_abort;
1115                 }
1116         }
1117
1118         if (!cgroup_init(handler)) {
1119                 ERROR("failed initializing cgroup support");
1120                 goto out_delete_net;
1121         }
1122
1123         cgroups_connected = true;
1124
1125         if (!cgroup_create(handler)) {
1126                 ERROR("failed creating cgroups");
1127                 goto out_delete_net;
1128         }
1129
1130         /*
1131          * if the rootfs is not a blockdev, prevent the container from
1132          * marking it readonly.
1133          *
1134          * if the container is unprivileged then skip rootfs pinning
1135          */
1136         if (lxc_list_empty(&handler->conf->id_map)) {
1137                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
1138                 if (handler->pinfd == -1)
1139                         INFO("failed to pin the container's rootfs");
1140         }
1141
1142         if (!preserve_ns(saved_ns_fd, preserve_mask, getpid(), &errmsg)) {
1143                 SYSERROR("Failed to preserve requested namespaces: %s",
1144                         errmsg ? errmsg : "(Out of memory)");
1145                 free(errmsg);
1146                 goto out_delete_net;
1147         }
1148         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
1149                 goto out_delete_net;
1150
1151         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
1152                 if (pipe(netpipepair) < 0) {
1153                         SYSERROR("Error creating pipe");
1154                         goto out_delete_net;
1155                 }
1156                 /* store netpipe in the global var for do_start's use */
1157                 netpipe = netpipepair[0];
1158         }
1159
1160         /* Create a process in a new set of namespaces */
1161         flags = handler->clone_flags;
1162         if (handler->clone_flags & CLONE_NEWUSER)
1163                 flags &= ~CLONE_NEWNET;
1164         handler->pid = lxc_clone(do_start, handler, handler->clone_flags);
1165         if (handler->pid < 0) {
1166                 SYSERROR("failed to fork into a new namespace");
1167                 goto out_delete_net;
1168         }
1169
1170         if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid, &errmsg)) {
1171                 INFO("Failed to store namespace references for stop hook: %s",
1172                         errmsg ? errmsg : "(Out of memory)");
1173                 free(errmsg);
1174         }
1175
1176         if (attach_ns(saved_ns_fd))
1177                 WARN("failed to restore saved namespaces");
1178
1179         lxc_sync_fini_child(handler);
1180
1181         /* map the container uids - the container became an invalid
1182          * userid the moment it was cloned with CLONE_NEWUSER - this
1183          * call doesn't change anything immediately, but allows the
1184          * container to setuid(0) (0 being mapped to something else on
1185          * the host) later to become a valid uid again */
1186         if (lxc_map_ids(&handler->conf->id_map, handler->pid)) {
1187                 ERROR("failed to set up id mapping");
1188                 goto out_delete_net;
1189         }
1190
1191         if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) {
1192                 failed_before_rename = 1;
1193                 goto out_delete_net;
1194         }
1195
1196         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) {
1197                 failed_before_rename = 1;
1198                 goto out_delete_net;
1199         }
1200
1201         if (!cgroup_create_legacy(handler)) {
1202                 ERROR("failed to setup the legacy cgroups for %s", name);
1203                 goto out_delete_net;
1204         }
1205         if (!cgroup_setup_limits(handler, false)) {
1206                 ERROR("failed to setup the cgroup limits for '%s'", name);
1207                 goto out_delete_net;
1208         }
1209
1210         if (!cgroup_enter(handler))
1211                 goto out_delete_net;
1212
1213         if (!cgroup_chown(handler))
1214                 goto out_delete_net;
1215
1216         if (failed_before_rename)
1217                 goto out_delete_net;
1218
1219         /* Create the network configuration */
1220         if (handler->clone_flags & CLONE_NEWNET) {
1221                 if (lxc_assign_network(handler->lxcpath, handler->name,
1222                                         &handler->conf->network, handler->pid)) {
1223                         ERROR("failed to create the configured network");
1224                         goto out_delete_net;
1225                 }
1226         }
1227
1228         if (netpipe != -1) {
1229                 struct lxc_list *iterator;
1230                 struct lxc_netdev *netdev;
1231
1232                 close(netpipe);
1233                 lxc_list_for_each(iterator, &handler->conf->network) {
1234                         netdev = iterator->elem;
1235                         if (netdev->type != LXC_NET_VETH)
1236                                 continue;
1237                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1238                                 ERROR("Error writing veth name to container");
1239                                 goto out_delete_net;
1240                         }
1241                 }
1242                 close(netpipepair[1]);
1243         }
1244
1245         /* Tell the child to continue its initialization.  we'll get
1246          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1247          */
1248         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1249                 goto out_delete_net;
1250
1251         if (!cgroup_setup_limits(handler, true)) {
1252                 ERROR("failed to setup the devices cgroup for '%s'", name);
1253                 goto out_delete_net;
1254         }
1255
1256         cgroup_disconnect();
1257         cgroups_connected = false;
1258
1259         /* read tty fds allocated by child */
1260         if (recv_ttys_from_child(handler) < 0) {
1261                 ERROR("failed to receive tty info from child");
1262                 goto out_delete_net;
1263         }
1264
1265         /* Tell the child to complete its initialization and wait for
1266          * it to exec or return an error.  (the child will never
1267          * return LXC_SYNC_POST_CGROUP+1.  It will either close the
1268          * sync pipe, causing lxc_sync_barrier_child to return
1269          * success, or return a different value, causing us to error
1270          * out).
1271          */
1272         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1273                 return -1;
1274
1275         if (detect_shared_rootfs())
1276                 umount2(handler->conf->rootfs.mount, MNT_DETACH);
1277
1278         if (handler->ops->post_start(handler, handler->data))
1279                 goto out_abort;
1280
1281         if (lxc_set_state(name, handler, RUNNING)) {
1282                 ERROR("failed to set state to %s",
1283                               lxc_state2str(RUNNING));
1284                 goto out_abort;
1285         }
1286
1287         lxc_sync_fini(handler);
1288
1289         return 0;
1290
1291 out_delete_net:
1292         if (cgroups_connected)
1293                 cgroup_disconnect();
1294         if (handler->clone_flags & CLONE_NEWNET)
1295                 lxc_delete_network(handler);
1296 out_abort:
1297         lxc_abort(name, handler);
1298         lxc_sync_fini(handler);
1299         if (handler->pinfd >= 0) {
1300                 close(handler->pinfd);
1301                 handler->pinfd = -1;
1302         }
1303
1304         return -1;
1305 }
1306
1307 int get_netns_fd(int pid)
1308 {
1309         char path[MAXPATHLEN];
1310         int ret, fd;
1311
1312         ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid);
1313         if (ret < 0 || ret >= MAXPATHLEN) {
1314                 WARN("Failed to pin netns file for pid %d", pid);
1315                 return -1;
1316         }
1317
1318         fd = open(path, O_RDONLY);
1319         if (fd < 0) {
1320                 WARN("Failed to pin netns file %s for pid %d: %s",
1321                                 path, pid, strerror(errno));
1322                 return -1;
1323         }
1324         return fd;
1325 }
1326
1327 int __lxc_start(const char *name, struct lxc_conf *conf,
1328                 struct lxc_operations* ops, void *data, const char *lxcpath,
1329                 bool backgrounded)
1330 {
1331         struct lxc_handler *handler;
1332         int err = -1;
1333         int status;
1334         int netnsfd = -1;
1335
1336         handler = lxc_init(name, conf, lxcpath);
1337         if (!handler) {
1338                 ERROR("failed to initialize the container");
1339                 return -1;
1340         }
1341         handler->ops = ops;
1342         handler->data = data;
1343         handler->backgrounded = backgrounded;
1344
1345         if (must_drop_cap_sys_boot(handler->conf)) {
1346                 #if HAVE_SYS_CAPABILITY_H
1347                 DEBUG("Dropping cap_sys_boot");
1348                 #else
1349                 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1350                 #endif
1351         } else {
1352                 DEBUG("Not dropping cap_sys_boot or watching utmp");
1353                 handler->conf->need_utmp_watch = 0;
1354         }
1355
1356         if (!attach_block_device(handler->conf)) {
1357                 ERROR("Failure attaching block device");
1358                 goto out_fini_nonet;
1359         }
1360
1361         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1362                 /* if the backing store is a device, mount it here and now */
1363                 if (rootfs_is_blockdev(conf)) {
1364                         if (unshare(CLONE_NEWNS) < 0) {
1365                                 ERROR("Error unsharing mounts");
1366                                 goto out_fini_nonet;
1367                         }
1368                         remount_all_slave();
1369                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1370                                 ERROR("Error setting up rootfs mount as root before spawn");
1371                                 goto out_fini_nonet;
1372                         }
1373                         INFO("Set up container rootfs as host root");
1374                 }
1375         }
1376
1377         err = lxc_spawn(handler);
1378         if (err) {
1379                 ERROR("failed to spawn '%s'", name);
1380                 goto out_detach_blockdev;
1381         }
1382
1383         handler->conf->reboot = 0;
1384
1385         netnsfd = get_netns_fd(handler->pid);
1386
1387         err = lxc_poll(name, handler);
1388         if (err) {
1389                 ERROR("mainloop exited with an error");
1390                 if (netnsfd >= 0)
1391                         close(netnsfd);
1392                 goto out_abort;
1393         }
1394
1395         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1396                 continue;
1397
1398         /*
1399          * If the child process exited but was not signaled,
1400          * it didn't call reboot.  This should mean it was an
1401          * lxc-execute which simply exited.  In any case, treat
1402          * it as a 'halt'
1403          */
1404         if (WIFSIGNALED(status)) {
1405                 switch(WTERMSIG(status)) {
1406                 case SIGINT: /* halt */
1407                         DEBUG("Container halting");
1408                         break;
1409                 case SIGHUP: /* reboot */
1410                         DEBUG("Container rebooting");
1411                         handler->conf->reboot = 1;
1412                         break;
1413                 case SIGSYS: /* seccomp */
1414                         DEBUG("Container violated its seccomp policy");
1415                         break;
1416                 default:
1417                         DEBUG("unknown exit status for init: %d", WTERMSIG(status));
1418                         break;
1419                 }
1420         }
1421
1422         DEBUG("Pushing physical nics back to host namespace");
1423         lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf);
1424
1425         DEBUG("Tearing down virtual network devices used by container");
1426         lxc_delete_network(handler);
1427
1428         if (netnsfd >= 0)
1429                 close(netnsfd);
1430
1431         if (handler->pinfd >= 0) {
1432                 close(handler->pinfd);
1433                 handler->pinfd = -1;
1434         }
1435
1436         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1437         err =  lxc_error_set_and_log(handler->pid, status);
1438 out_fini:
1439         lxc_delete_network(handler);
1440
1441 out_detach_blockdev:
1442         detach_block_device(handler->conf);
1443
1444 out_fini_nonet:
1445         lxc_fini(name, handler);
1446         return err;
1447
1448 out_abort:
1449         lxc_abort(name, handler);
1450         goto out_fini;
1451 }
1452
1453 struct start_args {
1454         char *const *argv;
1455 };
1456
1457 static int start(struct lxc_handler *handler, void* data)
1458 {
1459         struct start_args *arg = data;
1460
1461         NOTICE("exec'ing '%s'", arg->argv[0]);
1462
1463         execvp(arg->argv[0], arg->argv);
1464         SYSERROR("failed to exec %s", arg->argv[0]);
1465         return 0;
1466 }
1467
1468 static int post_start(struct lxc_handler *handler, void* data)
1469 {
1470         struct start_args *arg = data;
1471
1472         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
1473         return 0;
1474 }
1475
1476 static struct lxc_operations start_ops = {
1477         .start = start,
1478         .post_start = post_start
1479 };
1480
1481 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf,
1482               const char *lxcpath, bool backgrounded)
1483 {
1484         struct start_args start_arg = {
1485                 .argv = argv,
1486         };
1487
1488         conf->need_utmp_watch = 1;
1489         return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded);
1490 }
1491
1492 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1493                                             const char *name)
1494 {
1495         char destroy[MAXPATHLEN];
1496         bool bret = true;
1497         int ret = 0;
1498         struct lxc_container *c;
1499         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1500                 bret = do_destroy_container(handler->conf);
1501                 if (!bret) {
1502                         ERROR("Error destroying rootfs for %s", name);
1503                         return;
1504                 }
1505         }
1506         INFO("Destroyed rootfs for %s", name);
1507
1508         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1509         if (ret < 0 || ret >= MAXPATHLEN) {
1510                 ERROR("Error printing path for %s", name);
1511                 ERROR("Error destroying directory for %s", name);
1512                 return;
1513         }
1514
1515         c = lxc_container_new(name, handler->lxcpath);
1516         if (c) {
1517                 if (container_disk_lock(c)) {
1518                         INFO("Could not update lxc_snapshots file");
1519                         lxc_container_put(c);
1520                 } else {
1521                         mod_all_rdeps(c, false);
1522                         container_disk_unlock(c);
1523                         lxc_container_put(c);
1524                 }
1525         }
1526
1527         if (am_unpriv())
1528                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, destroy);
1529         else
1530                 ret = lxc_rmdir_onedev(destroy, NULL);
1531
1532         if (ret < 0) {
1533                 ERROR("Error destroying directory for %s", name);
1534                 return;
1535         }
1536         INFO("Destroyed directory for %s", name);
1537 }
1538
1539 static int lxc_rmdir_onedev_wrapper(void *data)
1540 {
1541         char *arg = (char *) data;
1542         return lxc_rmdir_onedev(arg, NULL);
1543 }
1544
1545 static bool do_destroy_container(struct lxc_conf *conf) {
1546         if (am_unpriv()) {
1547                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf) < 0)
1548                         return false;
1549                 return true;
1550         }
1551         return bdev_destroy(conf);
1552 }