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