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 <lxc/log.h>
  92 #include <lxc/conf.h>
  93 #include <lxc/confile.h>
  94 #include <lxc/start.h>
  95 #include <lxc/utils.h>
  96 #include <lxc/cgroup.h>
  97 #include <lxc/monitor.h>
  98
  99 #include "error.h"
 100 #include "af_unix.h"
 101 #include "mainloop.h"
 102 #include "commands.h"
 103
 104
 105 lxc_log_define(lxc_start, lxc);
 106
 107 LXC_TTY_HANDLER(SIGINT);
 108 LXC_TTY_HANDLER(SIGQUIT);
 109
 110 static int setup_sigchld_fd(sigset_t *oldmask)
 111 {
 112         sigset_t mask;
 113         int fd;
 114
 115         if (sigprocmask(SIG_BLOCK, NULL, &mask)) {
 116                 SYSERROR("failed to get mask signal");
 117                 return -1;
 118         }
 119
 120         if (sigaddset(&mask, SIGCHLD) || sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 121                 SYSERROR("failed to set mask signal");
 122                 return -1;
 123         }
 124
 125         fd = signalfd(-1, &mask, 0);
 126         if (fd < 0) {
 127                 SYSERROR("failed to create the signal fd");
 128                 return -1;
 129         }
 130
 131         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 132                 SYSERROR("failed to set sigfd to close-on-exec");
 133                 close(fd);
 134                 return -1;
 135         }
 136
 137         DEBUG("sigchild handler set");
 138
 139         return fd;
 140 }
 141
 142 static int sigchld_handler(int fd, void *data,
 143                            struct lxc_epoll_descr *descr)
 144 {
 145         DEBUG("child exited");
 146
 147         return 1;
 148 }
 149
 150 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 151 {
 152         handler->state = state;
 153         lxc_monitor_send_state(name, state);
 154         return 0;
 155 }
 156
 157 int lxc_poll(const char *name, struct lxc_handler *handler)
 158 {
 159         int sigfd = handler->sigfd;
 160         int pid = handler->pid;
 161         struct lxc_epoll_descr descr;
 162
 163         if (lxc_mainloop_open(&descr)) {
 164                 ERROR("failed to create mainloop");
 165                 goto out_sigfd;
 166         }
 167
 168         if (lxc_mainloop_add_handler(&descr, sigfd, sigchld_handler, &pid)) {
 169                 ERROR("failed to add handler for the signal");
 170                 goto out_mainloop_open;
 171         }
 172
 173         if (lxc_command_mainloop_add(name, &descr, handler))
 174                 goto out_mainloop_open;
 175
 176         return lxc_mainloop(&descr);
 177
 178 out_mainloop_open:
 179         lxc_mainloop_close(&descr);
 180 out_sigfd:
 181         close(sigfd);
 182         return -1;
 183 }
 184
 185 static int fdname(int fd, char *name, size_t size)
 186 {
 187         char path[MAXPATHLEN];
 188         ssize_t len;
 189
 190         snprintf(path, MAXPATHLEN, "/proc/self/fd/%d", fd);
 191
 192         len = readlink(path, name, size);
 193         if (len >  0)
 194                 path[len] = '\0';
 195
 196         return (len <= 0) ? -1 : 0;
 197 }
 198
 199 static int console_init(char *console, size_t size)
 200 {
 201         struct stat stat;
 202         int i;
 203
 204         for (i = 0; i < 3; i++) {
 205                 if (!isatty(i))
 206                         continue;
 207
 208                 if (ttyname_r(i, console, size)) {
 209                         SYSERROR("failed to retrieve tty name");
 210                         return -1;
 211                 }
 212
 213                 return 0;
 214         }
 215
 216         if (!fstat(0, &stat)) {
 217                 if (S_ISREG(stat.st_mode) || S_ISCHR(stat.st_mode) ||
 218                     S_ISFIFO(stat.st_mode) || S_ISLNK(stat.st_mode))
 219                         return fdname(0, console, size);
 220         }
 221
 222         console[0] = '\0';
 223
 224         DEBUG("console initialized");
 225
 226         return 0;
 227 }
 228
 229 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf)
 230 {
 231         struct lxc_handler *handler;
 232
 233         handler = malloc(sizeof(*handler));
 234         if (!handler)
 235                 return NULL;
 236
 237         memset(handler, 0, sizeof(*handler));
 238
 239         handler->conf = conf;
 240
 241         /* Begin the set the state to STARTING*/
 242         if (lxc_set_state(name, handler, STARTING)) {
 243                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 244                 goto out_free;
 245         }
 246
 247         if (console_init(conf->console, sizeof(conf->console))) {
 248                 ERROR("failed to initialize the console");
 249                 goto out_aborting;
 250         }
 251
 252         if (lxc_create_tty(name, conf)) {
 253                 ERROR("failed to create the ttys");
 254                 goto out_aborting;
 255         }
 256
 257         /* the signal fd has to be created before forking otherwise
 258          * if the child process exits before we setup the signal fd,
 259          * the event will be lost and the command will be stuck */
 260         handler->sigfd = setup_sigchld_fd(&handler->oldmask);
 261         if (handler->sigfd < 0) {
 262                 ERROR("failed to set sigchild fd handler");
 263                 goto out_delete_tty;
 264         }
 265
 266         /* Avoid signals from terminal */
 267         LXC_TTY_ADD_HANDLER(SIGINT);
 268         LXC_TTY_ADD_HANDLER(SIGQUIT);
 269
 270         INFO("'%s' is initialized", name);
 271         return handler;
 272
 273 out_delete_tty:
 274         lxc_delete_tty(&conf->tty_info);
 275 out_aborting:
 276         lxc_set_state(name, handler, ABORTING);
 277 out_free:
 278         free(handler);
 279         return NULL;
 280 }
 281
 282 void lxc_fini(const char *name, struct lxc_handler *handler)
 283 {
 284         /* The STOPPING state is there for future cleanup code
 285          * which can take awhile
 286          */
 287         lxc_set_state(name, handler, STOPPING);
 288         lxc_set_state(name, handler, STOPPED);
 289         lxc_unlink_nsgroup(name);
 290
 291         lxc_delete_tty(&handler->conf->tty_info);
 292         free(handler);
 293
 294         LXC_TTY_DEL_HANDLER(SIGQUIT);
 295         LXC_TTY_DEL_HANDLER(SIGINT);
 296 }
 297
 298 void lxc_abort(const char *name, struct lxc_handler *handler)
 299 {
 300         lxc_set_state(name, handler, ABORTING);
 301         kill(handler->pid, SIGKILL);
 302 }
 303
 304 struct start_arg {
 305         const char *name;
 306         char *const *argv;
 307         struct lxc_handler *handler;
 308         int *sv;
 309 };
 310
 311 static int do_start(void *arg)
 312 {
 313         struct start_arg *start_arg = arg;
 314         struct lxc_handler *handler = start_arg->handler;
 315         const char *name = start_arg->name;
 316         char *const *argv = start_arg->argv;
 317         int *sv = start_arg->sv;
 318         int err = -1, sync;
 319
 320         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 321                 SYSERROR("failed to set sigprocmask");
 322                 return -1;
 323         }
 324
 325         close(sv[1]);
 326
 327         /* Be sure we don't inherit this after the exec */
 328         fcntl(sv[0], F_SETFD, FD_CLOEXEC);
 329
 330         /* Tell our father he can begin to configure the container */
 331         if (write(sv[0], &sync, sizeof(sync)) < 0) {
 332                 SYSERROR("failed to write socket");
 333                 return -1;
 334         }
 335
 336         /* Wait for the father to finish the configuration */
 337         if (read(sv[0], &sync, sizeof(sync)) < 0) {
 338                 SYSERROR("failed to read socket");
 339                 return -1;
 340         }
 341
 342         /* Setup the container, ip, names, utsname, ... */
 343         if (lxc_setup(name, handler->conf)) {
 344                 ERROR("failed to setup the container");
 345                 goto out_warn_father;
 346         }
 347
 348         if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 349                 SYSERROR("failed to remove CAP_SYS_BOOT capability");
 350                 return -1;
 351         }
 352
 353         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 354                 SYSERROR("failed to set pdeath signal");
 355                 return -1;
 356         }
 357
 358         NOTICE("exec'ing '%s'", argv[0]);
 359
 360         execvp(argv[0], argv);
 361         SYSERROR("failed to exec %s", argv[0]);
 362
 363 out_warn_father:
 364         /* If the exec fails, tell that to our father */
 365         if (write(sv[0], &err, sizeof(err)) < 0)
 366                 SYSERROR("failed to write the socket");
 367         return -1;
 368 }
 369
 370 int lxc_spawn(const char *name, struct lxc_handler *handler, char *const argv[])
 371 {
 372         int sv[2];
 373         int clone_flags;
 374         int err = -1, sync;
 375
 376         struct start_arg start_arg = {
 377                 .name = name,
 378                 .argv = argv,
 379                 .handler = handler,
 380                 .sv = sv,
 381         };
 382
 383         /* Synchro socketpair */
 384         if (socketpair(AF_LOCAL, SOCK_STREAM, 0, sv)) {
 385                 SYSERROR("failed to create communication socketpair");
 386                 return -1;
 387         }
 388
 389         clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 390         if (!lxc_list_empty(&handler->conf->network)) {
 391
 392                 clone_flags |= CLONE_NEWNET;
 393
 394                 /* that should be done before the clone because we will
 395                  * fill the netdev index and use them in the child
 396                  */
 397                 if (lxc_create_network(&handler->conf->network)) {
 398                         ERROR("failed to create the network");
 399                         goto out_close;
 400                 }
 401         }
 402
 403         /* Create a process in a new set of namespaces */
 404         handler->pid = lxc_clone(do_start, &start_arg, clone_flags);
 405         if (handler->pid < 0) {
 406                 SYSERROR("failed to fork into a new namespace");
 407                 goto out_close;
 408         }
 409
 410         close(sv[0]);
 411
 412         /* Wait for the child to be ready */
 413         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 414                 SYSERROR("failed to read the socket");
 415                 goto out_abort;
 416         }
 417
 418         if (lxc_rename_nsgroup(name, handler))
 419                 goto out_abort;
 420
 421         /* Create the network configuration */
 422         if (clone_flags & CLONE_NEWNET) {
 423                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 424                         ERROR("failed to create the configured network");
 425                         goto out_abort;
 426                 }
 427         }
 428
 429         /* Tell the child to continue its initialization */
 430         if (write(sv[1], &sync, sizeof(sync)) < 0) {
 431                 SYSERROR("failed to write the socket");
 432                 goto out_abort;
 433         }
 434
 435         /* Wait for the child to exec or returning an error */
 436         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 437                 ERROR("failed to read the socket");
 438                 goto out_abort;
 439         }
 440
 441         if (lxc_set_state(name, handler, RUNNING)) {
 442                 ERROR("failed to set state to %s",
 443                               lxc_state2str(RUNNING));
 444                 goto out_abort;
 445         }
 446
 447         err = 0;
 448
 449         NOTICE("'%s' started with pid '%d'", argv[0], handler->pid);
 450
 451 out_close:
 452         close(sv[0]);
 453         close(sv[1]);
 454         return err;
 455
 456 out_abort:
 457         lxc_abort(name, handler);
 458         close(sv[1]);
 459         return -1;
 460 }
 461
 462 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf)
 463 {
 464         struct lxc_handler *handler;
 465         int err = -1;
 466         int status;
 467
 468         handler = lxc_init(name, conf);
 469         if (!handler) {
 470                 ERROR("failed to initialize the container");
 471                 return -1;
 472         }
 473
 474         err = lxc_spawn(name, handler, argv);
 475         if (err) {
 476                 ERROR("failed to spawn '%s'", argv[0]);
 477                 goto out_fini;
 478         }
 479
 480         err = lxc_close_all_inherited_fd();
 481         if (err) {
 482                 ERROR("unable to close inherited fds");
 483                 goto out_abort;
 484         }
 485
 486         err = lxc_poll(name, handler);
 487         if (err) {
 488                 ERROR("mainloop exited with an error");
 489                 goto out_abort;
 490         }
 491
 492         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 493                 continue;
 494
 495         err =  lxc_error_set_and_log(handler->pid, status);
 496 out_fini:
 497         lxc_fini(name, handler);
 498         return err;
 499
 500 out_abort:
 501         lxc_abort(name, handler);
 502         goto out_fini;
 503 }