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 "cgroup.h"
  97 #include "error.h"
  98 #include "af_unix.h"
  99 #include "mainloop.h"
 100 #include "utils.h"
 101 #include "utmp.h"
 102 #include "monitor.h"
 103 #include "commands.h"
 104 #include "console.h"
 105
 106 lxc_log_define(lxc_start, lxc);
 107
 108 LXC_TTY_HANDLER(SIGINT);
 109 LXC_TTY_HANDLER(SIGQUIT);
 110
 111 static int match_fd(int fd)
 112 {
 113         return (fd == 0 || fd == 1 || fd == 2);
 114 }
 115
 116 int lxc_check_inherited(void)
 117 {
 118         struct dirent dirent, *direntp;
 119         int fd, fddir;
 120         DIR *dir;
 121         int ret = 0;
 122
 123         dir = opendir("/proc/self/fd");
 124         if (!dir) {
 125                 WARN("failed to open directory: %m");
 126                 return -1;
 127         }
 128
 129         fddir = dirfd(dir);
 130
 131         while (!readdir_r(dir, &dirent, &direntp)) {
 132                 char procpath[64];
 133                 char path[PATH_MAX];
 134
 135                 if (!direntp)
 136                         break;
 137
 138                 if (!strcmp(direntp->d_name, "."))
 139                         continue;
 140
 141                 if (!strcmp(direntp->d_name, ".."))
 142                         continue;
 143
 144                 fd = atoi(direntp->d_name);
 145
 146                 if (fd == fddir || fd == lxc_log_fd)
 147                         continue;
 148
 149                 if (match_fd(fd))
 150                         continue;
 151                 /*
 152                  * found inherited fd
 153                  */
 154                 ret = -1;
 155
 156                 snprintf(procpath, sizeof(procpath), "/proc/self/fd/%d", fd);
 157
 158                 if (readlink(procpath, path, sizeof(path)) == -1)
 159                         ERROR("readlink(%s) failed : %m", procpath);
 160                 else
 161                         ERROR("inherited fd %d on %s", fd, path);
 162         }
 163
 164         if (closedir(dir))
 165                 ERROR("failed to close directory");
 166         return ret;
 167 }
 168
 169 static int setup_sigchld_fd(sigset_t *oldmask)
 170 {
 171         sigset_t mask;
 172         int fd;
 173
 174         if (sigprocmask(SIG_BLOCK, NULL, &mask)) {
 175                 SYSERROR("failed to get mask signal");
 176                 return -1;
 177         }
 178
 179         if (sigaddset(&mask, SIGCHLD) || sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 180                 SYSERROR("failed to set mask signal");
 181                 return -1;
 182         }
 183
 184         fd = signalfd(-1, &mask, 0);
 185         if (fd < 0) {
 186                 SYSERROR("failed to create the signal fd");
 187                 return -1;
 188         }
 189
 190         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 191                 SYSERROR("failed to set sigfd to close-on-exec");
 192                 close(fd);
 193                 return -1;
 194         }
 195
 196         DEBUG("sigchild handler set");
 197
 198         return fd;
 199 }
 200
 201 static int sigchld_handler(int fd, void *data,
 202                            struct lxc_epoll_descr *descr)
 203 {
 204         DEBUG("child exited");
 205
 206         return 1;
 207 }
 208
 209 int lxc_pid_callback(int fd, struct lxc_request *request,
 210                      struct lxc_handler *handler)
 211 {
 212         struct lxc_answer answer;
 213         int ret;
 214
 215         answer.pid = handler->pid;
 216         answer.ret = 0;
 217
 218         ret = send(fd, &answer, sizeof(answer), 0);
 219         if (ret < 0) {
 220                 WARN("failed to send answer to the peer");
 221                 return -1;
 222         }
 223
 224         if (ret != sizeof(answer)) {
 225                 ERROR("partial answer sent");
 226                 return -1;
 227         }
 228
 229         return 0;
 230 }
 231
 232 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 233 {
 234         handler->state = state;
 235         lxc_monitor_send_state(name, state);
 236         return 0;
 237 }
 238
 239 int lxc_poll(const char *name, struct lxc_handler *handler)
 240 {
 241         int sigfd = handler->sigfd;
 242         int pid = handler->pid;
 243         struct lxc_epoll_descr descr;
 244
 245         if (lxc_mainloop_open(&descr)) {
 246                 ERROR("failed to create mainloop");
 247                 goto out_sigfd;
 248         }
 249
 250         if (lxc_mainloop_add_handler(&descr, sigfd, sigchld_handler, &pid)) {
 251                 ERROR("failed to add handler for the signal");
 252                 goto out_mainloop_open;
 253         }
 254
 255         if (lxc_console_mainloop_add(&descr, handler)) {
 256                 ERROR("failed to add console handler to mainloop");
 257                 goto out_mainloop_open;
 258         }
 259
 260         if (lxc_command_mainloop_add(name, &descr, handler)) {
 261                 ERROR("failed to add command handler to mainloop");
 262                 goto out_mainloop_open;
 263         }
 264
 265         if (lxc_utmp_mainloop_add(&descr, handler)) {
 266                 ERROR("failed to add utmp handler to mainloop");
 267                 goto out_mainloop_open;
 268         }
 269
 270         return lxc_mainloop(&descr);
 271
 272 out_mainloop_open:
 273         lxc_mainloop_close(&descr);
 274 out_sigfd:
 275         close(sigfd);
 276         return -1;
 277 }
 278
 279 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf)
 280 {
 281         struct lxc_handler *handler;
 282
 283         handler = malloc(sizeof(*handler));
 284         if (!handler)
 285                 return NULL;
 286
 287         memset(handler, 0, sizeof(*handler));
 288
 289         handler->conf = conf;
 290
 291         handler->name = strdup(name);
 292         if (!handler->name) {
 293                 ERROR("failed to allocate memory");
 294                 goto out_free;
 295         }
 296
 297         /* Begin the set the state to STARTING*/
 298         if (lxc_set_state(name, handler, STARTING)) {
 299                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 300                 goto out_free_name;
 301         }
 302
 303         if (lxc_create_tty(name, conf)) {
 304                 ERROR("failed to create the ttys");
 305                 goto out_aborting;
 306         }
 307
 308         if (lxc_create_console(conf)) {
 309                 ERROR("failed to create console");
 310                 goto out_delete_tty;
 311         }
 312
 313         /* the signal fd has to be created before forking otherwise
 314          * if the child process exits before we setup the signal fd,
 315          * the event will be lost and the command will be stuck */
 316         handler->sigfd = setup_sigchld_fd(&handler->oldmask);
 317         if (handler->sigfd < 0) {
 318                 ERROR("failed to set sigchild fd handler");
 319                 goto out_delete_console;
 320         }
 321
 322         /* Avoid signals from terminal */
 323         LXC_TTY_ADD_HANDLER(SIGINT);
 324         LXC_TTY_ADD_HANDLER(SIGQUIT);
 325
 326         INFO("'%s' is initialized", name);
 327         return handler;
 328
 329 out_delete_console:
 330         lxc_delete_console(&conf->console);
 331 out_delete_tty:
 332         lxc_delete_tty(&conf->tty_info);
 333 out_aborting:
 334         lxc_set_state(name, handler, ABORTING);
 335 out_free_name:
 336         free(handler->name);
 337         handler->name = NULL;
 338 out_free:
 339         free(handler);
 340         return NULL;
 341 }
 342
 343 void lxc_fini(const char *name, struct lxc_handler *handler)
 344 {
 345         /* The STOPPING state is there for future cleanup code
 346          * which can take awhile
 347          */
 348         lxc_set_state(name, handler, STOPPING);
 349         lxc_set_state(name, handler, STOPPED);
 350         lxc_unlink_nsgroup(name);
 351
 352         lxc_delete_console(&handler->conf->console);
 353         lxc_delete_tty(&handler->conf->tty_info);
 354         free(handler->name);
 355         free(handler);
 356
 357         LXC_TTY_DEL_HANDLER(SIGQUIT);
 358         LXC_TTY_DEL_HANDLER(SIGINT);
 359 }
 360
 361 void lxc_abort(const char *name, struct lxc_handler *handler)
 362 {
 363         lxc_set_state(name, handler, ABORTING);
 364         if (handler->pid > 0)
 365                 kill(handler->pid, SIGKILL);
 366 }
 367
 368 struct start_arg {
 369         const char *name;
 370         char *const *argv;
 371         struct lxc_handler *handler;
 372         int *sv;
 373 };
 374
 375 static int do_start(void *arg)
 376 {
 377         struct start_arg *start_arg = arg;
 378         struct lxc_handler *handler = start_arg->handler;
 379         const char *name = start_arg->name;
 380         char *const *argv = start_arg->argv;
 381         int *sv = start_arg->sv;
 382         int err = -1, sync;
 383
 384         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 385                 SYSERROR("failed to set sigprocmask");
 386                 return -1;
 387         }
 388
 389         close(sv[1]);
 390
 391         /* Be sure we don't inherit this after the exec */
 392         fcntl(sv[0], F_SETFD, FD_CLOEXEC);
 393
 394         /* Tell our father he can begin to configure the container */
 395         if (write(sv[0], &sync, sizeof(sync)) < 0) {
 396                 SYSERROR("failed to write socket");
 397                 return -1;
 398         }
 399
 400         /* Wait for the father to finish the configuration */
 401         if (read(sv[0], &sync, sizeof(sync)) < 0) {
 402                 SYSERROR("failed to read socket");
 403                 return -1;
 404         }
 405
 406         /* Setup the container, ip, names, utsname, ... */
 407         if (lxc_setup(name, handler->conf)) {
 408                 ERROR("failed to setup the container");
 409                 goto out_warn_father;
 410         }
 411
 412         if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 413                 SYSERROR("failed to remove CAP_SYS_BOOT capability");
 414                 return -1;
 415         }
 416
 417         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 418                 SYSERROR("failed to set pdeath signal");
 419                 return -1;
 420         }
 421
 422         NOTICE("exec'ing '%s'", argv[0]);
 423
 424         execvp(argv[0], argv);
 425         SYSERROR("failed to exec %s", argv[0]);
 426
 427 out_warn_father:
 428         /* If the exec fails, tell that to our father */
 429         if (write(sv[0], &err, sizeof(err)) < 0)
 430                 SYSERROR("failed to write the socket");
 431         return -1;
 432 }
 433
 434 int lxc_spawn(const char *name, struct lxc_handler *handler, char *const argv[])
 435 {
 436         int sv[2];
 437         int clone_flags;
 438         int err = -1, sync;
 439
 440         struct start_arg start_arg = {
 441                 .name = name,
 442                 .argv = argv,
 443                 .handler = handler,
 444                 .sv = sv,
 445         };
 446
 447         /* Synchro socketpair */
 448         if (socketpair(AF_LOCAL, SOCK_STREAM, 0, sv)) {
 449                 SYSERROR("failed to create communication socketpair");
 450                 return -1;
 451         }
 452
 453         clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 454         if (!lxc_list_empty(&handler->conf->network)) {
 455
 456                 clone_flags |= CLONE_NEWNET;
 457
 458                 /* that should be done before the clone because we will
 459                  * fill the netdev index and use them in the child
 460                  */
 461                 if (lxc_create_network(&handler->conf->network)) {
 462                         ERROR("failed to create the network");
 463                         goto out_close;
 464                 }
 465         }
 466
 467         /* Create a process in a new set of namespaces */
 468         handler->pid = lxc_clone(do_start, &start_arg, clone_flags);
 469         if (handler->pid < 0) {
 470                 SYSERROR("failed to fork into a new namespace");
 471                 goto out_delete_net;
 472         }
 473
 474         close(sv[0]);
 475
 476         /* Wait for the child to be ready */
 477         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 478                 SYSERROR("failed to read the socket");
 479                 goto out_delete_net;
 480         }
 481
 482         if (lxc_rename_nsgroup(name, handler))
 483                 goto out_delete_net;
 484
 485         /* Create the network configuration */
 486         if (clone_flags & CLONE_NEWNET) {
 487                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 488                         ERROR("failed to create the configured network");
 489                         goto out_delete_net;
 490                 }
 491         }
 492
 493         /* Tell the child to continue its initialization */
 494         if (write(sv[1], &sync, sizeof(sync)) < 0) {
 495                 SYSERROR("failed to write the socket");
 496                 goto out_abort;
 497         }
 498
 499         /* Wait for the child to exec or returning an error */
 500         if (read(sv[1], &sync, sizeof(sync)) < 0) {
 501                 ERROR("failed to read the socket");
 502                 goto out_abort;
 503         }
 504
 505         if (lxc_set_state(name, handler, RUNNING)) {
 506                 ERROR("failed to set state to %s",
 507                               lxc_state2str(RUNNING));
 508                 goto out_abort;
 509         }
 510
 511         err = 0;
 512
 513         NOTICE("'%s' started with pid '%d'", argv[0], handler->pid);
 514
 515 out_close:
 516         close(sv[0]);
 517         close(sv[1]);
 518         return err;
 519
 520 out_delete_net:
 521         if (clone_flags & CLONE_NEWNET)
 522                 lxc_delete_network(&handler->conf->network);
 523 out_abort:
 524         lxc_abort(name, handler);
 525         close(sv[1]);
 526         return -1;
 527 }
 528
 529 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf)
 530 {
 531         struct lxc_handler *handler;
 532         int err = -1;
 533         int status;
 534
 535         if (lxc_check_inherited())
 536                 return -1;
 537
 538         handler = lxc_init(name, conf);
 539         if (!handler) {
 540                 ERROR("failed to initialize the container");
 541                 return -1;
 542         }
 543
 544         err = lxc_spawn(name, handler, argv);
 545         if (err) {
 546                 ERROR("failed to spawn '%s'", argv[0]);
 547                 goto out_fini;
 548         }
 549
 550         err = lxc_poll(name, handler);
 551         if (err) {
 552                 ERROR("mainloop exited with an error");
 553                 goto out_abort;
 554         }
 555
 556         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 557                 continue;
 558
 559         err =  lxc_error_set_and_log(handler->pid, status);
 560 out_fini:
 561         lxc_fini(name, handler);
 562         return err;
 563
 564 out_abort:
 565         lxc_abort(name, handler);
 566         goto out_fini;
 567 }