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 #include <lxc/lxc.h>
  48 #include <lxc/confile.h>
  49
  50 #ifdef HAVE_SYS_SIGNALFD_H
  51 #  include <sys/signalfd.h>
  52 #else
  53 #  ifndef __NR_signalfd4
  54 /* assume kernel headers are too old */
  55 #    if __i386__
  56 #      define __NR_signalfd4 327
  57 #    elif __x86_64__
  58 #      define __NR_signalfd4 289
  59 #    elif __powerpc__
  60 #      define __NR_signalfd4 313
  61 #    elif __s390x__
  62 #      define __NR_signalfd4 322
  63 #    endif
  64 #endif
  65
  66 #  ifndef __NR_signalfd
  67 /* assume kernel headers are too old */
  68 #    if __i386__
  69 #      define __NR_signalfd 321
  70 #    elif __x86_64__
  71 #      define __NR_signalfd 282
  72 #    elif __powerpc__
  73 #      define __NR_signalfd 305
  74 #    elif __s390x__
  75 #      define __NR_signalfd 316
  76 #    endif
  77 #endif
  78
  79 int signalfd(int fd, const sigset_t *mask, int flags)
  80 {
  81         int retval;
  82
  83         retval = syscall (__NR_signalfd4, fd, mask, _NSIG / 8, flags);
  84         if (errno == ENOSYS && flags == 0)
  85                 retval = syscall (__NR_signalfd, fd, mask, _NSIG / 8);
  86         return retval;
  87 }
  88 #endif
  89
  90 #if !HAVE_DECL_PR_CAPBSET_DROP
  91 #define PR_CAPBSET_DROP 24
  92 #endif
  93
  94 #include "error.h"
  95 #include "af_unix.h"
  96 #include "mainloop.h"
  97 #include "commands.h"
  98
  99 #include <lxc/lxc.h>
 100 #include <lxc/log.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 static int 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         int ret = -1;
 159         struct lxc_epoll_descr descr;
 160
 161         if (lxc_mainloop_open(&descr)) {
 162                 ERROR("failed to create mainloop");
 163                 goto out_sigfd;
 164         }
 165
 166         if (lxc_mainloop_add_handler(&descr, sigfd, sigchld_handler, &pid)) {
 167                 ERROR("failed to add handler for the signal");
 168                 goto out_mainloop_open;
 169         }
 170
 171         if (lxc_command_mainloop_add(name, &descr, handler))
 172                 goto out_mainloop_open;
 173
 174         ret = lxc_mainloop(&descr);
 175
 176 out:
 177         return ret;
 178
 179 out_mainloop_open:
 180         lxc_mainloop_close(&descr);
 181 out_sigfd:
 182         close(sigfd);
 183         goto out;
 184 }
 185
 186 static void remove_init_pid(const char *name, pid_t pid)
 187 {
 188         char init[MAXPATHLEN];
 189
 190         snprintf(init, MAXPATHLEN, LXCPATH "/%s/init", name);
 191         unlink(init);
 192 }
 193
 194 static int fdname(int fd, char *name, size_t size)
 195 {
 196         char path[MAXPATHLEN];
 197         ssize_t len;
 198
 199         snprintf(path, MAXPATHLEN, "/proc/self/fd/%d", fd);
 200
 201         len = readlink(path, name, size);
 202         if (len >  0)
 203                 path[len] = '\0';
 204
 205         return (len <= 0) ? -1 : 0;
 206 }
 207
 208 static int console_init(char *console, size_t size)
 209 {
 210         struct stat stat;
 211         int i;
 212
 213         for (i = 0; i < 3; i++) {
 214                 if (!isatty(i))
 215                         continue;
 216
 217                 if (ttyname_r(i, console, size)) {
 218                         SYSERROR("failed to retrieve tty name");
 219                         return -1;
 220                 }
 221
 222                 return 0;
 223         }
 224
 225         if (!fstat(0, &stat)) {
 226                 if (S_ISREG(stat.st_mode) || S_ISCHR(stat.st_mode) ||
 227                     S_ISFIFO(stat.st_mode) || S_ISLNK(stat.st_mode))
 228                         return fdname(0, console, size);
 229         }
 230
 231         console[0] = '\0';
 232
 233         DEBUG("console initialized");
 234
 235         return 0;
 236 }
 237
 238 struct lxc_handler *lxc_init(const char *name)
 239 {
 240         struct lxc_handler *handler;
 241         char path[MAXPATHLEN];
 242
 243         handler = malloc(sizeof(*handler));
 244         if (!handler)
 245                 return NULL;
 246
 247         memset(handler, 0, sizeof(*handler));
 248
 249         /* Begin the set the state to STARTING*/
 250         if (set_state(name, handler, STARTING)) {
 251                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 252                 goto out_free;
 253         }
 254
 255         if (lxc_conf_init(&handler->conf)) {
 256                 ERROR("failed to initialize the configuration");
 257                 goto out_aborting;
 258         }
 259
 260         snprintf(path, sizeof(path), LXCPATH "/%s/config", name);
 261
 262         if (!access(path, F_OK) && lxc_config_read(path, &handler->conf)) {
 263                 ERROR("failed to read the configuration file");
 264                 goto out_aborting;
 265         }
 266
 267         if (console_init(handler->conf.console,
 268                          sizeof(handler->conf.console))) {
 269                 ERROR("failed to initialize the console");
 270                 goto out_aborting;
 271         }
 272
 273         if (lxc_create_tty(name, &handler->conf)) {
 274                 ERROR("failed to create the ttys");
 275                 goto out_aborting;
 276         }
 277
 278         /* the signal fd has to be created before forking otherwise
 279          * if the child process exits before we setup the signal fd,
 280          * the event will be lost and the command will be stuck */
 281         handler->sigfd = setup_sigchld_fd(&handler->oldmask);
 282         if (handler->sigfd < 0) {
 283                 ERROR("failed to set sigchild fd handler");
 284                 goto out_delete_tty;
 285         }
 286
 287         /* Avoid signals from terminal */
 288         LXC_TTY_ADD_HANDLER(SIGINT);
 289         LXC_TTY_ADD_HANDLER(SIGQUIT);
 290
 291 out:
 292         if (handler)
 293                 INFO("'%s' is initialized", name);
 294
 295         return handler;
 296
 297 out_delete_tty:
 298         lxc_delete_tty(&handler->conf.tty_info);
 299 out_aborting:
 300         set_state(name, handler, ABORTING);
 301 out_free:
 302         free(handler);
 303         handler = NULL;
 304         goto out;
 305 }
 306
 307 void lxc_fini(const char *name, struct lxc_handler *handler)
 308 {
 309         /* The STOPPING state is there for future cleanup code
 310          * which can take awhile
 311          */
 312         set_state(name, handler, STOPPING);
 313         set_state(name, handler, STOPPED);
 314         lxc_unlink_nsgroup(name);
 315
 316         if (handler) {
 317                 remove_init_pid(name, handler->pid);
 318                 lxc_delete_tty(&handler->conf.tty_info);
 319                 free(handler);
 320         }
 321
 322         LXC_TTY_DEL_HANDLER(SIGQUIT);
 323         LXC_TTY_DEL_HANDLER(SIGINT);
 324 }
 325
 326 void lxc_abort(const char *name, struct lxc_handler *handler)
 327 {
 328         set_state(name, handler, ABORTING);
 329         kill(handler->pid, SIGKILL);
 330 }
 331
 332 struct start_arg {
 333         const char *name;
 334         char *const *argv;
 335         struct lxc_handler *handler;
 336         int *sv;
 337 };
 338
 339 static int do_start(void *arg)
 340 {
 341         struct start_arg *start_arg = arg;
 342         struct lxc_handler *handler = start_arg->handler;
 343         const char *name = start_arg->name;
 344         char *const *argv = start_arg->argv;
 345         int *sv = start_arg->sv;
 346         int err = -1, sync;
 347
 348         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 349                 SYSERROR("failed to set sigprocmask");
 350                 goto out_child;
 351         }
 352
 353         close(sv[1]);
 354
 355         /* Be sure we don't inherit this after the exec */
 356         fcntl(sv[0], F_SETFD, FD_CLOEXEC);
 357
 358         /* Tell our father he can begin to configure the container */
 359         if (write(sv[0], &sync, sizeof(sync)) < 0) {
 360                 SYSERROR("failed to write socket");
 361                 goto out_child;
 362         }
 363
 364         /* Wait for the father to finish the configuration */
 365         if (read(sv[0], &sync, sizeof(sync)) < 0) {
 366                 SYSERROR("failed to read socket");
 367                 goto out_child;
 368         }
 369
 370         /* Setup the container, ip, names, utsname, ... */
 371         if (lxc_setup(name, &handler->conf)) {
 372                 ERROR("failed to setup the container");
 373                 goto out_warn_father;
 374         }
 375
 376         if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 377                 SYSERROR("failed to remove CAP_SYS_BOOT capability");
 378                 goto out_child;
 379         }
 380
 381         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 382                 SYSERROR("failed to set pdeath signal");
 383                 goto out_child;
 384         }
 385
 386         NOTICE("exec'ing '%s'", argv[0]);
 387
 388         execvp(argv[0], argv);
 389         SYSERROR("failed to exec %s", argv[0]);
 390
 391 out_warn_father:
 392         /* If the exec fails, tell that to our father */
 393         if (write(sv[0], &err, sizeof(err)) < 0)
 394                 SYSERROR("failed to write the socket");
 395 out_child:
 396         return -1;
 397 }
 398
 399 int lxc_spawn(const char *name, struct lxc_handler *handler, char *const argv[])
 400 {
 401         int sv[2];
 402         int clone_flags;
 403         int err = -1, sync;
 404
 405         struct start_arg start_arg = {
 406                 .name = name,
 407                 .argv = argv,
 408                 .handler = handler,
 409                 .sv = sv,
 410         };
 411
 412         /* Synchro socketpair */
 413         if (socketpair(AF_LOCAL, SOCK_STREAM, 0, sv)) {
 414                 SYSERROR("failed to create communication socketpair");
 415                 goto out;
 416         }
 417
 418         clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 419         if (!lxc_list_empty(&handler->conf.network)) {
 420
 421                 clone_flags |= CLONE_NEWNET;
 422
 423                 /* that should be done before the clone because we will
 424                  * fill the netdev index and use them in the child
 425                  */
 426                 if (lxc_create_network(&handler->conf.network)) {
 427                         ERROR("failed to create the network");
 428                         goto out_close;
 429                 }
 430         }
 431
 432         /* Create a process in a new set of namespaces */
 433         handler->pid = lxc_clone(do_start, &start_arg, clone_flags);
 434         if (handler->pid < 0) {
 435                 SYSERROR("failed to fork into a new namespace");
 436                 goto out_close;
 437         }
 438
 439         close(sv[0]);
 440
 441         /* Wait for the child to be ready */
 442         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 443                 SYSERROR("failed to read the socket");
 444                 goto out_abort;
 445         }
 446
 447         if (lxc_rename_nsgroup(name, handler))
 448                 goto out_abort;
 449
 450         /* Create the network configuration */
 451         if (clone_flags & CLONE_NEWNET) {
 452                 if (lxc_assign_network(&handler->conf.network, handler->pid)) {
 453                         ERROR("failed to create the configured network");
 454                         goto out_abort;
 455                 }
 456         }
 457
 458         /* Tell the child to continue its initialization */
 459         if (write(sv[1], &sync, sizeof(sync)) < 0) {
 460                 SYSERROR("failed to write the socket");
 461                 goto out_abort;
 462         }
 463
 464         /* Wait for the child to exec or returning an error */
 465         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 466                 ERROR("failed to read the socket");
 467                 goto out_abort;
 468         }
 469
 470         if (set_state(name, handler, RUNNING)) {
 471                 ERROR("failed to set state to %s",
 472                               lxc_state2str(RUNNING));
 473                 goto out_abort;
 474         }
 475
 476         err = 0;
 477
 478         NOTICE("'%s' started with pid '%d'", argv[0], handler->pid);
 479
 480 out_close:
 481         close(sv[0]);
 482         close(sv[1]);
 483 out:
 484         return err;
 485
 486 out_abort:
 487         lxc_abort(name, handler);
 488         goto out_close;
 489 }
 490
 491 int lxc_start(const char *name, char *const argv[])
 492 {
 493         struct lxc_handler *handler;
 494         int err = -1;
 495         int status;
 496
 497         handler = lxc_init(name);
 498         if (!handler) {
 499                 ERROR("failed to initialize the container");
 500                 return -1;
 501         }
 502
 503         err = lxc_spawn(name, handler, argv);
 504         if (err) {
 505                 ERROR("failed to spawn '%s'", argv[0]);
 506                 goto out;
 507         }
 508
 509         err = lxc_close_all_inherited_fd();
 510         if (err) {
 511                 ERROR("unable to close inherited fds");
 512                 goto out_abort;
 513         }
 514
 515         err = lxc_poll(name, handler);
 516         if (err) {
 517                 ERROR("mainloop exited with an error");
 518                 goto out_abort;
 519         }
 520
 521         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 522                 continue;
 523
 524         err =  lxc_error_set_and_log(handler->pid, status);
 525 out:
 526         lxc_fini(name, handler);
 527         return err;
 528
 529 out_abort:
 530         lxc_abort(name, handler);
 531         goto out;
 532 }