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 /* assume kernel headers are too old */
  52 #include <stdint.h>
  53 struct signalfd_siginfo
  54 {
  55         uint32_t ssi_signo;
  56         int32_t ssi_errno;
  57         int32_t ssi_code;
  58         uint32_t ssi_pid;
  59         uint32_t ssi_uid;
  60         int32_t ssi_fd;
  61         uint32_t ssi_tid;
  62         uint32_t ssi_band;
  63         uint32_t ssi_overrun;
  64         uint32_t ssi_trapno;
  65         int32_t ssi_status;
  66         int32_t ssi_int;
  67         uint64_t ssi_ptr;
  68         uint64_t ssi_utime;
  69         uint64_t ssi_stime;
  70         uint64_t ssi_addr;
  71         uint8_t __pad[48];
  72 };
  73
  74 #  ifndef __NR_signalfd4
  75 /* assume kernel headers are too old */
  76 #    if __i386__
  77 #      define __NR_signalfd4 327
  78 #    elif __x86_64__
  79 #      define __NR_signalfd4 289
  80 #    elif __powerpc__
  81 #      define __NR_signalfd4 313
  82 #    elif __s390x__
  83 #      define __NR_signalfd4 322
  84 #    endif
  85 #endif
  86
  87 #  ifndef __NR_signalfd
  88 /* assume kernel headers are too old */
  89 #    if __i386__
  90 #      define __NR_signalfd 321
  91 #    elif __x86_64__
  92 #      define __NR_signalfd 282
  93 #    elif __powerpc__
  94 #      define __NR_signalfd 305
  95 #    elif __s390x__
  96 #      define __NR_signalfd 316
  97 #    endif
  98 #endif
  99
 100 int signalfd(int fd, const sigset_t *mask, int flags)
 101 {
 102         int retval;
 103
 104         retval = syscall (__NR_signalfd4, fd, mask, _NSIG / 8, flags);
 105         if (errno == ENOSYS && flags == 0)
 106                 retval = syscall (__NR_signalfd, fd, mask, _NSIG / 8);
 107         return retval;
 108 }
 109 #endif
 110
 111 #if !HAVE_DECL_PR_CAPBSET_DROP
 112 #define PR_CAPBSET_DROP 24
 113 #endif
 114
 115 #include "start.h"
 116 #include "conf.h"
 117 #include "log.h"
 118 #include "cgroup.h"
 119 #include "error.h"
 120 #include "af_unix.h"
 121 #include "mainloop.h"
 122 #include "utils.h"
 123 #include "utmp.h"
 124 #include "monitor.h"
 125 #include "commands.h"
 126 #include "console.h"
 127 #include "sync.h"
 128
 129 lxc_log_define(lxc_start, lxc);
 130
 131 LXC_TTY_HANDLER(SIGINT);
 132 LXC_TTY_HANDLER(SIGQUIT);
 133
 134 static int match_fd(int fd)
 135 {
 136         return (fd == 0 || fd == 1 || fd == 2);
 137 }
 138
 139 int lxc_check_inherited(int fd_to_ignore)
 140 {
 141         struct dirent dirent, *direntp;
 142         int fd, fddir;
 143         DIR *dir;
 144
 145         dir = opendir("/proc/self/fd");
 146         if (!dir) {
 147                 WARN("failed to open directory: %m");
 148                 return -1;
 149         }
 150
 151         fddir = dirfd(dir);
 152
 153         while (!readdir_r(dir, &dirent, &direntp)) {
 154                 if (!direntp)
 155                         break;
 156
 157                 if (!strcmp(direntp->d_name, "."))
 158                         continue;
 159
 160                 if (!strcmp(direntp->d_name, ".."))
 161                         continue;
 162
 163                 fd = atoi(direntp->d_name);
 164
 165                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 166                         continue;
 167
 168                 if (match_fd(fd))
 169                         continue;
 170
 171                 WARN("inherited fd %d", fd);
 172         }
 173
 174         closedir(dir); /* cannot fail */
 175         return 0;
 176 }
 177
 178 static int setup_signal_fd(sigset_t *oldmask)
 179 {
 180         sigset_t mask;
 181         int fd;
 182
 183         /* Block everything except serious error signals */
 184         if (sigfillset(&mask) ||
 185             sigdelset(&mask, SIGILL) ||
 186             sigdelset(&mask, SIGSEGV) ||
 187             sigdelset(&mask, SIGBUS) ||
 188             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 189                 SYSERROR("failed to set signal mask");
 190                 return -1;
 191         }
 192
 193         fd = signalfd(-1, &mask, 0);
 194         if (fd < 0) {
 195                 SYSERROR("failed to create the signal fd");
 196                 return -1;
 197         }
 198
 199         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 200                 SYSERROR("failed to set sigfd to close-on-exec");
 201                 close(fd);
 202                 return -1;
 203         }
 204
 205         DEBUG("sigchild handler set");
 206
 207         return fd;
 208 }
 209
 210 static int signal_handler(int fd, void *data,
 211                            struct lxc_epoll_descr *descr)
 212 {
 213         struct signalfd_siginfo siginfo;
 214         int ret;
 215         pid_t *pid = data;
 216
 217         ret = read(fd, &siginfo, sizeof(siginfo));
 218         if (ret < 0) {
 219                 ERROR("failed to read signal info");
 220                 return -1;
 221         }
 222
 223         if (ret != sizeof(siginfo)) {
 224                 ERROR("unexpected siginfo size");
 225                 return -1;
 226         }
 227
 228         if (siginfo.ssi_signo != SIGCHLD) {
 229                 kill(*pid, siginfo.ssi_signo);
 230                 INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid);
 231                 return 0;
 232         }
 233
 234         if (siginfo.ssi_code == CLD_STOPPED ||
 235             siginfo.ssi_code == CLD_CONTINUED) {
 236                 INFO("container init process was stopped/continued");
 237                 return 0;
 238         }
 239
 240         /* more robustness, protect ourself from a SIGCHLD sent
 241          * by a process different from the container init
 242          */
 243         if (siginfo.ssi_pid != *pid) {
 244                 WARN("invalid pid for SIGCHLD");
 245                 return 0;
 246         }
 247
 248         DEBUG("container init process exited");
 249         return 1;
 250 }
 251
 252 int lxc_pid_callback(int fd, struct lxc_request *request,
 253                      struct lxc_handler *handler)
 254 {
 255         struct lxc_answer answer;
 256         int ret;
 257
 258         answer.pid = handler->pid;
 259         answer.ret = 0;
 260
 261         ret = send(fd, &answer, sizeof(answer), 0);
 262         if (ret < 0) {
 263                 WARN("failed to send answer to the peer");
 264                 return -1;
 265         }
 266
 267         if (ret != sizeof(answer)) {
 268                 ERROR("partial answer sent");
 269                 return -1;
 270         }
 271
 272         return 0;
 273 }
 274
 275 int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state)
 276 {
 277         handler->state = state;
 278         lxc_monitor_send_state(name, state);
 279         return 0;
 280 }
 281
 282 int lxc_poll(const char *name, struct lxc_handler *handler)
 283 {
 284         int sigfd = handler->sigfd;
 285         int pid = handler->pid;
 286         struct lxc_epoll_descr descr;
 287
 288         if (lxc_mainloop_open(&descr)) {
 289                 ERROR("failed to create mainloop");
 290                 goto out_sigfd;
 291         }
 292
 293         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 294                 ERROR("failed to add handler for the signal");
 295                 goto out_mainloop_open;
 296         }
 297
 298         if (lxc_console_mainloop_add(&descr, handler)) {
 299                 ERROR("failed to add console handler to mainloop");
 300                 goto out_mainloop_open;
 301         }
 302
 303         if (lxc_command_mainloop_add(name, &descr, handler)) {
 304                 ERROR("failed to add command handler to mainloop");
 305                 goto out_mainloop_open;
 306         }
 307
 308         if (lxc_utmp_mainloop_add(&descr, handler)) {
 309                 ERROR("failed to add utmp handler to mainloop");
 310                 goto out_mainloop_open;
 311         }
 312
 313         return lxc_mainloop(&descr);
 314
 315 out_mainloop_open:
 316         lxc_mainloop_close(&descr);
 317 out_sigfd:
 318         close(sigfd);
 319         return -1;
 320 }
 321
 322 struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf)
 323 {
 324         struct lxc_handler *handler;
 325
 326         handler = malloc(sizeof(*handler));
 327         if (!handler)
 328                 return NULL;
 329
 330         memset(handler, 0, sizeof(*handler));
 331
 332         handler->conf = conf;
 333
 334         handler->name = strdup(name);
 335         if (!handler->name) {
 336                 ERROR("failed to allocate memory");
 337                 goto out_free;
 338         }
 339
 340         /* Begin the set the state to STARTING*/
 341         if (lxc_set_state(name, handler, STARTING)) {
 342                 ERROR("failed to set state '%s'", lxc_state2str(STARTING));
 343                 goto out_free_name;
 344         }
 345
 346         if (lxc_create_tty(name, conf)) {
 347                 ERROR("failed to create the ttys");
 348                 goto out_aborting;
 349         }
 350
 351         if (lxc_create_console(conf)) {
 352                 ERROR("failed to create console");
 353                 goto out_delete_tty;
 354         }
 355
 356         /* the signal fd has to be created before forking otherwise
 357          * if the child process exits before we setup the signal fd,
 358          * the event will be lost and the command will be stuck */
 359         handler->sigfd = setup_signal_fd(&handler->oldmask);
 360         if (handler->sigfd < 0) {
 361                 ERROR("failed to set sigchild fd handler");
 362                 goto out_delete_console;
 363         }
 364
 365         INFO("'%s' is initialized", name);
 366         return handler;
 367
 368 out_delete_console:
 369         lxc_delete_console(&conf->console);
 370 out_delete_tty:
 371         lxc_delete_tty(&conf->tty_info);
 372 out_aborting:
 373         lxc_set_state(name, handler, ABORTING);
 374 out_free_name:
 375         free(handler->name);
 376         handler->name = NULL;
 377 out_free:
 378         free(handler);
 379         return NULL;
 380 }
 381
 382 void lxc_fini(const char *name, struct lxc_handler *handler)
 383 {
 384         /* The STOPPING state is there for future cleanup code
 385          * which can take awhile
 386          */
 387         lxc_set_state(name, handler, STOPPING);
 388         lxc_set_state(name, handler, STOPPED);
 389
 390         /* reset mask set by setup_signal_fd */
 391         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 392                 WARN("failed to restore sigprocmask");
 393
 394         lxc_delete_console(&handler->conf->console);
 395         lxc_delete_tty(&handler->conf->tty_info);
 396         free(handler->name);
 397         free(handler);
 398 }
 399
 400 void lxc_abort(const char *name, struct lxc_handler *handler)
 401 {
 402         lxc_set_state(name, handler, ABORTING);
 403         if (handler->pid > 0)
 404                 kill(handler->pid, SIGKILL);
 405 }
 406
 407 static int do_start(void *data)
 408 {
 409         struct lxc_handler *handler = data;
 410
 411         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 412                 SYSERROR("failed to set sigprocmask");
 413                 return -1;
 414         }
 415
 416         /* This prctl must be before the synchro, so if the parent
 417          * dies before we set the parent death signal, we will detect
 418          * its death with the synchro right after, otherwise we have
 419          * a window where the parent can exit before we set the pdeath
 420          * signal leading to a unsupervized container.
 421          */
 422         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 423                 SYSERROR("failed to set pdeath signal");
 424                 return -1;
 425         }
 426
 427         lxc_sync_fini_parent(handler);
 428
 429         /* Tell the parent task it can begin to configure the
 430          * container and wait for it to finish
 431          */
 432         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 433                 return -1;
 434
 435         /* Setup the container, ip, names, utsname, ... */
 436         if (lxc_setup(handler->name, handler->conf)) {
 437                 ERROR("failed to setup the container");
 438                 goto out_warn_father;
 439         }
 440
 441         if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 442                 SYSERROR("failed to remove CAP_SYS_BOOT capability");
 443                 return -1;
 444         }
 445
 446         close(handler->sigfd);
 447
 448         /* after this call, we are in error because this
 449          * ops should not return as it execs */
 450         if (handler->ops->start(handler, handler->data))
 451                 return -1;
 452
 453 out_warn_father:
 454         lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE);
 455         return -1;
 456 }
 457
 458 int lxc_spawn(struct lxc_handler *handler)
 459 {
 460         int clone_flags;
 461         int failed_before_rename = 0;
 462         const char *name = handler->name;
 463
 464         if (lxc_sync_init(handler))
 465                 return -1;
 466
 467         clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS;
 468         if (!lxc_list_empty(&handler->conf->network)) {
 469
 470                 clone_flags |= CLONE_NEWNET;
 471
 472                 /* Find gateway addresses from the link device, which is
 473                  * no longer accessible inside the container. Do this
 474                  * before creating network interfaces, since goto
 475                  * out_delete_net does not work before lxc_clone. */
 476                 if (lxc_find_gateway_addresses(handler)) {
 477                         ERROR("failed to find gateway addresses");
 478                         lxc_sync_fini(handler);
 479                         return -1;
 480                 }
 481
 482                 /* that should be done before the clone because we will
 483                  * fill the netdev index and use them in the child
 484                  */
 485                 if (lxc_create_network(handler)) {
 486                         ERROR("failed to create the network");
 487                         lxc_sync_fini(handler);
 488                         return -1;
 489                 }
 490         }
 491
 492
 493         /* Create a process in a new set of namespaces */
 494         handler->pid = lxc_clone(do_start, handler, clone_flags);
 495         if (handler->pid < 0) {
 496                 SYSERROR("failed to fork into a new namespace");
 497                 goto out_delete_net;
 498         }
 499
 500         lxc_sync_fini_child(handler);
 501
 502         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE))
 503                 failed_before_rename = 1;
 504
 505         if (lxc_cgroup_create(name, handler->pid))
 506                 goto out_delete_net;
 507
 508         if (failed_before_rename)
 509                 goto out_delete_net;
 510
 511         /* Create the network configuration */
 512         if (clone_flags & CLONE_NEWNET) {
 513                 if (lxc_assign_network(&handler->conf->network, handler->pid)) {
 514                         ERROR("failed to create the configured network");
 515                         goto out_delete_net;
 516                 }
 517         }
 518
 519         /* Tell the child to continue its initialization and wait for
 520          * it to exec or return an error
 521          */
 522         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
 523                 return -1;
 524
 525         if (handler->ops->post_start(handler, handler->data))
 526                 goto out_abort;
 527
 528         if (lxc_set_state(name, handler, RUNNING)) {
 529                 ERROR("failed to set state to %s",
 530                               lxc_state2str(RUNNING));
 531                 goto out_abort;
 532         }
 533
 534         lxc_sync_fini(handler);
 535         return 0;
 536
 537 out_delete_net:
 538         if (clone_flags & CLONE_NEWNET)
 539                 lxc_delete_network(&handler->conf->network);
 540 out_abort:
 541         lxc_abort(name, handler);
 542         lxc_sync_fini(handler);
 543         return -1;
 544 }
 545
 546 int __lxc_start(const char *name, struct lxc_conf *conf,
 547                 struct lxc_operations* ops, void *data)
 548 {
 549         struct lxc_handler *handler;
 550         int err = -1;
 551         int status;
 552
 553         handler = lxc_init(name, conf);
 554         if (!handler) {
 555                 ERROR("failed to initialize the container");
 556                 return -1;
 557         }
 558         handler->ops = ops;
 559         handler->data = data;
 560
 561         err = lxc_spawn(handler);
 562         if (err) {
 563                 ERROR("failed to spawn '%s'", name);
 564                 goto out_fini;
 565         }
 566
 567         /* Avoid signals from terminal */
 568         LXC_TTY_ADD_HANDLER(SIGINT);
 569         LXC_TTY_ADD_HANDLER(SIGQUIT);
 570
 571         err = lxc_poll(name, handler);
 572         if (err) {
 573                 ERROR("mainloop exited with an error");
 574                 goto out_abort;
 575         }
 576
 577         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
 578                 continue;
 579
 580         err =  lxc_error_set_and_log(handler->pid, status);
 581 out_fini:
 582         LXC_TTY_DEL_HANDLER(SIGQUIT);
 583         LXC_TTY_DEL_HANDLER(SIGINT);
 584         lxc_cgroup_destroy(name);
 585         lxc_fini(name, handler);
 586         return err;
 587
 588 out_abort:
 589         lxc_abort(name, handler);
 590         goto out_fini;
 591 }
 592
 593 struct start_args {
 594         char *const *argv;
 595 };
 596
 597 static int start(struct lxc_handler *handler, void* data)
 598 {
 599         struct start_args *arg = data;
 600
 601         NOTICE("exec'ing '%s'", arg->argv[0]);
 602
 603         execvp(arg->argv[0], arg->argv);
 604         SYSERROR("failed to exec %s", arg->argv[0]);
 605         return 0;
 606 }
 607
 608 static int post_start(struct lxc_handler *handler, void* data)
 609 {
 610         struct start_args *arg = data;
 611
 612         NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid);
 613         return 0;
 614 }
 615
 616 static struct lxc_operations start_ops = {
 617         .start = start,
 618         .post_start = post_start
 619 };
 620
 621 int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf)
 622 {
 623         struct start_args start_arg = {
 624                 .argv = argv,
 625         };
 626
 627         if (lxc_check_inherited(-1))
 628                 return -1;
 629
 630         return __lxc_start(name, conf, &start_ops, &start_arg);
 631 }