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         handler->lock = lxc_get_lock(name);
 250         if (handler->lock < 0)
 251                 goto out_free;
 252
 253         /* Begin the set the state to STARTING*/
 254         if (set_state(name, handler, STARTING)) {
 255                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 256                 goto out_put_lock;
 257         }
 258
 259         if (lxc_conf_init(&handler->lxc_conf)) {
 260                 ERROR("failed to initialize the configuration");
 261                 goto out_aborting;
 262         }
 263
 264         snprintf(path, sizeof(path), LXCPATH "/%s/config", name);
 265
 266         if (!access(path, F_OK)) {
 267
 268                 if (lxc_config_read(path, &handler->lxc_conf)) {
 269                         ERROR("failed to read the configuration file");
 270                         goto out_aborting;
 271                 }
 272         }
 273
 274         if (console_init(handler->tty, sizeof(handler->tty))) {
 275                 ERROR("failed to initialize the console");
 276                 goto out_aborting;
 277         }
 278
 279         if (lxc_create_tty(name, &handler->tty_info)) {
 280                 ERROR("failed to create the ttys");
 281                 goto out_aborting;
 282         }
 283
 284         /* the signal fd has to be created before forking otherwise
 285          * if the child process exits before we setup the signal fd,
 286          * the event will be lost and the command will be stuck */
 287         handler->sigfd = setup_sigchld_fd(&handler->oldmask);
 288         if (handler->sigfd < 0) {
 289                 ERROR("failed to set sigchild fd handler");
 290                 goto out_delete_tty;
 291         }
 292
 293         /* Avoid signals from terminal */
 294         LXC_TTY_ADD_HANDLER(SIGINT);
 295         LXC_TTY_ADD_HANDLER(SIGQUIT);
 296
 297 out:
 298         if (handler)
 299                 INFO("'%s' is initialized", name);
 300
 301         return handler;
 302
 303 out_delete_tty:
 304         lxc_delete_tty(&handler->tty_info);
 305 out_aborting:
 306         set_state(name, handler, ABORTING);
 307 out_put_lock:
 308         lxc_put_lock(handler->lock);
 309 out_free:
 310         free(handler);
 311         handler = NULL;
 312         goto out;
 313 }
 314
 315 void lxc_fini(const char *name, struct lxc_handler *handler)
 316 {
 317         /* The STOPPING state is there for future cleanup code
 318          * which can take awhile
 319          */
 320         set_state(name, handler, STOPPING);
 321         set_state(name, handler, STOPPED);
 322         lxc_unlink_nsgroup(name);
 323
 324         if (handler) {
 325                 remove_init_pid(name, handler->pid);
 326                 lxc_delete_tty(&handler->tty_info);
 327                 lxc_put_lock(handler->lock);
 328                 free(handler);
 329         }
 330
 331         LXC_TTY_DEL_HANDLER(SIGQUIT);
 332         LXC_TTY_DEL_HANDLER(SIGINT);
 333 }
 334
 335 void lxc_abort(const char *name, struct lxc_handler *handler)
 336 {
 337         set_state(name, handler, ABORTING);
 338         kill(handler->pid, SIGKILL);
 339 }
 340
 341 struct start_arg {
 342         const char *name;
 343         char *const *argv;
 344         struct lxc_handler *handler;
 345         int *sv;
 346 };
 347
 348 static int do_start(void *arg)
 349 {
 350         struct start_arg *start_arg = arg;
 351         struct lxc_handler *handler = start_arg->handler;
 352         const char *name = start_arg->name;
 353         char *const *argv = start_arg->argv;
 354         int *sv = start_arg->sv;
 355         int err = -1, sync;
 356
 357         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 358                 SYSERROR("failed to set sigprocmask");
 359                 goto out_child;
 360         }
 361
 362         close(sv[1]);
 363
 364         /* Be sure we don't inherit this after the exec */
 365         fcntl(sv[0], F_SETFD, FD_CLOEXEC);
 366
 367         /* Tell our father he can begin to configure the container */
 368         if (write(sv[0], &sync, sizeof(sync)) < 0) {
 369                 SYSERROR("failed to write socket");
 370                 goto out_child;
 371         }
 372
 373         /* Wait for the father to finish the configuration */
 374         if (read(sv[0], &sync, sizeof(sync)) < 0) {
 375                 SYSERROR("failed to read socket");
 376                 goto out_child;
 377         }
 378
 379         /* Setup the container, ip, names, utsname, ... */
 380         if (lxc_setup(name, handler)) {
 381                 ERROR("failed to setup the container");
 382                 goto out_warn_father;
 383         }
 384
 385         if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 386                 SYSERROR("failed to remove CAP_SYS_BOOT capability");
 387                 goto out_child;
 388         }
 389
 390         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 391                 SYSERROR("failed to set pdeath signal");
 392                 goto out_child;
 393         }
 394
 395         NOTICE("exec'ing '%s'", argv[0]);
 396
 397         execvp(argv[0], argv);
 398         SYSERROR("failed to exec %s", argv[0]);
 399
 400 out_warn_father:
 401         /* If the exec fails, tell that to our father */
 402         if (write(sv[0], &err, sizeof(err)) < 0)
 403                 SYSERROR("failed to write the socket");
 404 out_child:
 405         return -1;
 406 }
 407
 408 int lxc_spawn(const char *name, struct lxc_handler *handler, char *const argv[])
 409 {
 410         int sv[2];
 411         int clone_flags;
 412         int err = -1, sync;
 413
 414         struct start_arg start_arg = {
 415                 .name = name,
 416                 .argv = argv,
 417                 .handler = handler,
 418                 .sv = sv,
 419         };
 420
 421         /* Synchro socketpair */
 422         if (socketpair(AF_LOCAL, SOCK_STREAM, 0, sv)) {
 423                 SYSERROR("failed to create communication socketpair");
 424                 goto out;
 425         }
 426
 427         clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 428         if (conf_has_network(name))
 429                 clone_flags |= CLONE_NEWNET;
 430
 431         /* Create a process in a new set of namespaces */
 432         handler->pid = lxc_clone(do_start, &start_arg, clone_flags);
 433         if (handler->pid < 0) {
 434                 SYSERROR("failed to fork into a new namespace");
 435                 goto out_close;
 436         }
 437
 438         close(sv[0]);
 439
 440         /* Wait for the child to be ready */
 441         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 442                 SYSERROR("failed to read the socket");
 443                 goto out_abort;
 444         }
 445
 446         if (lxc_rename_nsgroup(name, handler))
 447                 goto out_abort;
 448
 449         /* Create the network configuration */
 450         if (clone_flags & CLONE_NEWNET &&
 451             conf_create_network(name, handler->pid)) {
 452                 ERROR("failed to create the configured network");
 453                 goto out_abort;
 454         }
 455
 456         /* Tell the child to continue its initialization */
 457         if (write(sv[1], &sync, sizeof(sync)) < 0) {
 458                 SYSERROR("failed to write the socket");
 459                 goto out_abort;
 460         }
 461
 462         /* Wait for the child to exec or returning an error */
 463         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 464                 ERROR("failed to read the socket");
 465                 goto out_abort;
 466         }
 467
 468         if (set_state(name, handler, RUNNING)) {
 469                 ERROR("failed to set state to %s",
 470                               lxc_state2str(RUNNING));
 471                 goto out_abort;
 472         }
 473
 474         err = 0;
 475
 476         NOTICE("'%s' started with pid '%d'", argv[0], handler->pid);
 477
 478 out_close:
 479         close(sv[0]);
 480         close(sv[1]);
 481 out:
 482         return err;
 483
 484 out_abort:
 485         lxc_abort(name, handler);
 486         goto out_close;
 487 }
 488
 489 int lxc_start(const char *name, char *const argv[])
 490 {
 491         struct lxc_handler *handler;
 492         int err = -1;
 493         int status;
 494
 495         handler = lxc_init(name);
 496         if (!handler) {
 497                 ERROR("failed to initialize the container");
 498                 return -1;
 499         }
 500
 501         err = lxc_spawn(name, handler, argv);
 502         if (err) {
 503                 ERROR("failed to spawn '%s'", argv[0]);
 504                 goto out;
 505         }
 506
 507         err = lxc_close_all_inherited_fd();
 508         if (err) {
 509                 ERROR("unable to close inherited fds");
 510                 goto out_abort;
 511         }
 512
 513         err = lxc_poll(name, handler);
 514         if (err) {
 515                 ERROR("mainloop exited with an error");
 516                 goto out_abort;
 517         }
 518
 519         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 520                 continue;
 521
 522         err =  lxc_error_set_and_log(handler->pid, status);
 523 out:
 524         lxc_fini(name, handler);
 525         return err;
 526
 527 out_abort:
 528         lxc_abort(name, handler);
 529         goto out;
 530 }