src/lxc/start.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <dlezcano at fr.ibm.com>
   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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  22  */
  23
  24 #include "../config.h"
  25 #include <stdio.h>
  26 #undef _GNU_SOURCE
  27 #include <string.h>
  28 #include <stdlib.h>
  29 #include <dirent.h>
  30 #include <errno.h>
  31 #include <unistd.h>
  32 #include <signal.h>
  33 #include <fcntl.h>
  34 #include <termios.h>
  35 #include <namespace.h>
  36 #include <sys/param.h>
  37 #include <sys/file.h>
  38 #include <sys/mount.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/prctl.h>
  42 #include <sys/types.h>
  43 #include <sys/capability.h>
  44 #include <sys/wait.h>
  45 #include <sys/un.h>
  46 #include <sys/poll.h>
  47
  48 #ifdef HAVE_SYS_SIGNALFD_H
  49 #  include <sys/signalfd.h>
  50 #else
  51 #  ifndef __NR_signalfd4
  52 /* assume kernel headers are too old */
  53 #    if __i386__
  54 #      define __NR_signalfd4 327
  55 #    elif __x86_64__
  56 #      define __NR_signalfd4 289
  57 #    elif __powerpc__
  58 #      define __NR_signalfd4 313
  59 #    elif __s390x__
  60 #      define __NR_signalfd4 322
  61 #    endif
  62 #endif
  63
  64 #  ifndef __NR_signalfd
  65 /* assume kernel headers are too old */
  66 #    if __i386__
  67 #      define __NR_signalfd 321
  68 #    elif __x86_64__
  69 #      define __NR_signalfd 282
  70 #    elif __powerpc__
  71 #      define __NR_signalfd 305
  72 #    elif __s390x__
  73 #      define __NR_signalfd 316
  74 #    endif
  75 #endif
  76
  77 int signalfd(int fd, const sigset_t *mask, int flags)
  78 {
  79         int retval;
  80
  81         retval = syscall (__NR_signalfd4, fd, mask, _NSIG / 8, flags);
  82         if (errno == ENOSYS && flags == 0)
  83                 retval = syscall (__NR_signalfd, fd, mask, _NSIG / 8);
  84         return retval;
  85 }
  86 #endif
  87
  88 #if !HAVE_DECL_PR_CAPBSET_DROP
  89 #define PR_CAPBSET_DROP 24
  90 #endif
  91
  92 #include "start.h"
  93 #include "conf.h"
  94 #include "cgroup.h"
  95 #include "log.h"
  96 #include "error.h"
  97 #include "af_unix.h"
  98 #include "mainloop.h"
  99 #include "utils.h"
 100 #include "monitor.h"
 101 #include "commands.h"
 102 #include "console.h"
 103
 104 lxc_log_define(lxc_start, lxc);
 105
 106 LXC_TTY_HANDLER(SIGINT);
 107 LXC_TTY_HANDLER(SIGQUIT);
 108
 109 static int setup_sigchld_fd(sigset_t *oldmask)
 110 {
 111         sigset_t mask;
 112         int fd;
 113
 114         if (sigprocmask(SIG_BLOCK, NULL, &mask)) {
 115                 SYSERROR("failed to get mask signal");
 116                 return -1;
 117         }
 118
 119         if (sigaddset(&mask, SIGCHLD) || sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 120                 SYSERROR("failed to set mask signal");
 121                 return -1;
 122         }
 123
 124         fd = signalfd(-1, &mask, 0);
 125         if (fd < 0) {
 126                 SYSERROR("failed to create the signal fd");
 127                 return -1;
 128         }
 129
 130         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 131                 SYSERROR("failed to set sigfd to close-on-exec");
 132                 close(fd);
 133                 return -1;
 134         }
 135
 136         DEBUG("sigchild handler set");
 137
 138         return fd;
 139 }
 140
 141 static int sigchld_handler(int fd, void *data,
 142                            struct lxc_epoll_descr *descr)
 143 {
 144         DEBUG("child exited");
 145
 146         return 1;
 147 }
 148
 149 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 150 {
 151         handler->state = state;
 152         lxc_monitor_send_state(name, state);
 153         return 0;
 154 }
 155
 156 int lxc_poll(const char *name, struct lxc_handler *handler)
 157 {
 158         int sigfd = handler->sigfd;
 159         int pid = handler->pid;
 160         struct lxc_epoll_descr descr;
 161
 162         if (lxc_mainloop_open(&descr)) {
 163                 ERROR("failed to create mainloop");
 164                 goto out_sigfd;
 165         }
 166
 167         if (lxc_mainloop_add_handler(&descr, sigfd, sigchld_handler, &pid)) {
 168                 ERROR("failed to add handler for the signal");
 169                 goto out_mainloop_open;
 170         }
 171
 172         if (lxc_console_mainloop_add(&descr, handler)) {
 173                 ERROR("failed to add console handler to mainloop");
 174                 goto out_mainloop_open;
 175         }
 176
 177         if (lxc_command_mainloop_add(name, &descr, handler))
 178                 goto out_mainloop_open;
 179
 180         return lxc_mainloop(&descr);
 181
 182 out_mainloop_open:
 183         lxc_mainloop_close(&descr);
 184 out_sigfd:
 185         close(sigfd);
 186         return -1;
 187 }
 188
 189 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf)
 190 {
 191         struct lxc_handler *handler;
 192
 193         handler = malloc(sizeof(*handler));
 194         if (!handler)
 195                 return NULL;
 196
 197         memset(handler, 0, sizeof(*handler));
 198
 199         handler->conf = conf;
 200
 201         /* Begin the set the state to STARTING*/
 202         if (lxc_set_state(name, handler, STARTING)) {
 203                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 204                 goto out_free;
 205         }
 206
 207         if (lxc_create_tty(name, conf)) {
 208                 ERROR("failed to create the ttys");
 209                 goto out_aborting;
 210         }
 211
 212         if (lxc_create_console(conf)) {
 213                 ERROR("failed to create console");
 214                 goto out_delete_tty;
 215         }
 216
 217         /* the signal fd has to be created before forking otherwise
 218          * if the child process exits before we setup the signal fd,
 219          * the event will be lost and the command will be stuck */
 220         handler->sigfd = setup_sigchld_fd(&handler->oldmask);
 221         if (handler->sigfd < 0) {
 222                 ERROR("failed to set sigchild fd handler");
 223                 goto out_delete_console;
 224         }
 225
 226         /* Avoid signals from terminal */
 227         LXC_TTY_ADD_HANDLER(SIGINT);
 228         LXC_TTY_ADD_HANDLER(SIGQUIT);
 229
 230         INFO("'%s' is initialized", name);
 231         return handler;
 232
 233 out_delete_console:
 234         lxc_delete_console(&conf->console);
 235 out_delete_tty:
 236         lxc_delete_tty(&conf->tty_info);
 237 out_aborting:
 238         lxc_set_state(name, handler, ABORTING);
 239 out_free:
 240         free(handler);
 241         return NULL;
 242 }
 243
 244 void lxc_fini(const char *name, struct lxc_handler *handler)
 245 {
 246         /* The STOPPING state is there for future cleanup code
 247          * which can take awhile
 248          */
 249         lxc_set_state(name, handler, STOPPING);
 250         lxc_set_state(name, handler, STOPPED);
 251         lxc_unlink_nsgroup(name);
 252
 253         lxc_delete_console(&handler->conf->console);
 254         lxc_delete_tty(&handler->conf->tty_info);
 255         free(handler);
 256
 257         LXC_TTY_DEL_HANDLER(SIGQUIT);
 258         LXC_TTY_DEL_HANDLER(SIGINT);
 259 }
 260
 261 void lxc_abort(const char *name, struct lxc_handler *handler)
 262 {
 263         lxc_set_state(name, handler, ABORTING);
 264         kill(handler->pid, SIGKILL);
 265 }
 266
 267 struct start_arg {
 268         const char *name;
 269         char *const *argv;
 270         struct lxc_handler *handler;
 271         int *sv;
 272 };
 273
 274 static int do_start(void *arg)
 275 {
 276         struct start_arg *start_arg = arg;
 277         struct lxc_handler *handler = start_arg->handler;
 278         const char *name = start_arg->name;
 279         char *const *argv = start_arg->argv;
 280         int *sv = start_arg->sv;
 281         int err = -1, sync;
 282
 283         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 284                 SYSERROR("failed to set sigprocmask");
 285                 return -1;
 286         }
 287
 288         close(sv[1]);
 289
 290         /* Be sure we don't inherit this after the exec */
 291         fcntl(sv[0], F_SETFD, FD_CLOEXEC);
 292
 293         /* Tell our father he can begin to configure the container */
 294         if (write(sv[0], &sync, sizeof(sync)) < 0) {
 295                 SYSERROR("failed to write socket");
 296                 return -1;
 297         }
 298
 299         /* Wait for the father to finish the configuration */
 300         if (read(sv[0], &sync, sizeof(sync)) < 0) {
 301                 SYSERROR("failed to read socket");
 302                 return -1;
 303         }
 304
 305         /* Setup the container, ip, names, utsname, ... */
 306         if (lxc_setup(name, handler->conf)) {
 307                 ERROR("failed to setup the container");
 308                 goto out_warn_father;
 309         }
 310
 311         if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 312                 SYSERROR("failed to remove CAP_SYS_BOOT capability");
 313                 return -1;
 314         }
 315
 316         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 317                 SYSERROR("failed to set pdeath signal");
 318                 return -1;
 319         }
 320
 321         NOTICE("exec'ing '%s'", argv[0]);
 322
 323         execvp(argv[0], argv);
 324         SYSERROR("failed to exec %s", argv[0]);
 325
 326 out_warn_father:
 327         /* If the exec fails, tell that to our father */
 328         if (write(sv[0], &err, sizeof(err)) < 0)
 329                 SYSERROR("failed to write the socket");
 330         return -1;
 331 }
 332
 333 int lxc_spawn(const char *name, struct lxc_handler *handler, char *const argv[])
 334 {
 335         int sv[2];
 336         int clone_flags;
 337         int err = -1, sync;
 338
 339         struct start_arg start_arg = {
 340                 .name = name,
 341                 .argv = argv,
 342                 .handler = handler,
 343                 .sv = sv,
 344         };
 345
 346         /* Synchro socketpair */
 347         if (socketpair(AF_LOCAL, SOCK_STREAM, 0, sv)) {
 348                 SYSERROR("failed to create communication socketpair");
 349                 return -1;
 350         }
 351
 352         clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 353         if (!lxc_list_empty(&handler->conf->network)) {
 354
 355                 clone_flags |= CLONE_NEWNET;
 356
 357                 /* that should be done before the clone because we will
 358                  * fill the netdev index and use them in the child
 359                  */
 360                 if (lxc_create_network(&handler->conf->network)) {
 361                         ERROR("failed to create the network");
 362                         goto out_close;
 363                 }
 364         }
 365
 366         /* Create a process in a new set of namespaces */
 367         handler->pid = lxc_clone(do_start, &start_arg, clone_flags);
 368         if (handler->pid < 0) {
 369                 SYSERROR("failed to fork into a new namespace");
 370                 goto out_delete_net;
 371         }
 372
 373         close(sv[0]);
 374
 375         /* Wait for the child to be ready */
 376         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 377                 SYSERROR("failed to read the socket");
 378                 goto out_delete_net;
 379         }
 380
 381         if (lxc_rename_nsgroup(name, handler))
 382                 goto out_delete_net;
 383
 384         /* Create the network configuration */
 385         if (clone_flags & CLONE_NEWNET) {
 386                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 387                         ERROR("failed to create the configured network");
 388                         goto out_delete_net;
 389                 }
 390         }
 391
 392         /* Tell the child to continue its initialization */
 393         if (write(sv[1], &sync, sizeof(sync)) < 0) {
 394                 SYSERROR("failed to write the socket");
 395                 goto out_abort;
 396         }
 397
 398         /* Wait for the child to exec or returning an error */
 399         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 400                 ERROR("failed to read the socket");
 401                 goto out_abort;
 402         }
 403
 404         if (lxc_set_state(name, handler, RUNNING)) {
 405                 ERROR("failed to set state to %s",
 406                               lxc_state2str(RUNNING));
 407                 goto out_abort;
 408         }
 409
 410         err = 0;
 411
 412         NOTICE("'%s' started with pid '%d'", argv[0], handler->pid);
 413
 414 out_close:
 415         close(sv[0]);
 416         close(sv[1]);
 417         return err;
 418
 419 out_delete_net:
 420         if (clone_flags & CLONE_NEWNET)
 421                 lxc_delete_network(&handler->conf->network);
 422 out_abort:
 423         lxc_abort(name, handler);
 424         close(sv[1]);
 425         return -1;
 426 }
 427
 428 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf)
 429 {
 430         struct lxc_handler *handler;
 431         int err = -1;
 432         int status;
 433
 434         handler = lxc_init(name, conf);
 435         if (!handler) {
 436                 ERROR("failed to initialize the container");
 437                 return -1;
 438         }
 439
 440         err = lxc_spawn(name, handler, argv);
 441         if (err) {
 442                 ERROR("failed to spawn '%s'", argv[0]);
 443                 goto out_fini;
 444         }
 445
 446         err = lxc_close_all_inherited_fd();
 447         if (err) {
 448                 ERROR("unable to close inherited fds");
 449                 goto out_abort;
 450         }
 451
 452         err = lxc_poll(name, handler);
 453         if (err) {
 454                 ERROR("mainloop exited with an error");
 455                 goto out_abort;
 456         }
 457
 458         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 459                 continue;
 460
 461         err =  lxc_error_set_and_log(handler->pid, status);
 462 out_fini:
 463         lxc_fini(name, handler);
 464         return err;
 465
 466 out_abort:
 467         lxc_abort(name, handler);
 468         goto out_fini;
 469 }