src/nspawn/nspawn.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <signal.h>
  23 #include <sched.h>
  24 #include <unistd.h>
  25 #include <sys/types.h>
  26 #include <sys/syscall.h>
  27 #include <sys/mount.h>
  28 #include <sys/wait.h>
  29 #include <stdlib.h>
  30 #include <string.h>
  31 #include <stdio.h>
  32 #include <errno.h>
  33 #include <sys/prctl.h>
  34 #include <sys/capability.h>
  35 #include <getopt.h>
  36 #include <sys/epoll.h>
  37 #include <termios.h>
  38 #include <sys/signalfd.h>
  39 #include <grp.h>
  40 #include <linux/fs.h>
  41 #include <sys/un.h>
  42 #include <sys/socket.h>
  43
  44 #include <systemd/sd-daemon.h>
  45
  46 #include "log.h"
  47 #include "util.h"
  48 #include "mkdir.h"
  49 #include "audit.h"
  50 #include "missing.h"
  51 #include "cgroup-util.h"
  52 #include "strv.h"
  53 #include "path-util.h"
  54 #include "loopback-setup.h"
  55
  56 static char *arg_directory = NULL;
  57 static char *arg_user = NULL;
  58 static char **arg_controllers = NULL;
  59 static char *arg_uuid = NULL;
  60 static bool arg_private_network = false;
  61 static bool arg_read_only = false;
  62 static bool arg_boot = false;
  63
  64 static int help(void) {
  65
  66         printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
  67                "Spawn a minimal namespace container for debugging, testing and building.\n\n"
  68                "  -h --help             Show this help\n"
  69                "  -D --directory=NAME   Root directory for the container\n"
  70                "  -b --boot             Boot up full system (i.e. invoke init)\n"
  71                "  -u --user=USER        Run the command under specified user or uid\n"
  72                "  -C --controllers=LIST Put the container in specified comma-separated cgroup hierarchies\n"
  73                "     --uuid=UUID        Set a specific machine UUID for the container\n"
  74                "     --private-network  Disable network in container\n"
  75                "     --read-only        Mount the root directory read-only\n",
  76                program_invocation_short_name);
  77
  78         return 0;
  79 }
  80
  81 static int parse_argv(int argc, char *argv[]) {
  82
  83         enum {
  84                 ARG_PRIVATE_NETWORK = 0x100,
  85                 ARG_UUID,
  86                 ARG_READ_ONLY
  87         };
  88
  89         static const struct option options[] = {
  90                 { "help",            no_argument,       NULL, 'h'                 },
  91                 { "directory",       required_argument, NULL, 'D'                 },
  92                 { "user",            required_argument, NULL, 'u'                 },
  93                 { "controllers",     required_argument, NULL, 'C'                 },
  94                 { "private-network", no_argument,       NULL, ARG_PRIVATE_NETWORK },
  95                 { "boot",            no_argument,       NULL, 'b'                 },
  96                 { "uuid",            required_argument, NULL, ARG_UUID            },
  97                 { "read-only",       no_argument,       NULL, ARG_READ_ONLY       },
  98                 { NULL,              0,                 NULL, 0                   }
  99         };
 100
 101         int c;
 102
 103         assert(argc >= 0);
 104         assert(argv);
 105
 106         while ((c = getopt_long(argc, argv, "+hD:u:C:b", options, NULL)) >= 0) {
 107
 108                 switch (c) {
 109
 110                 case 'h':
 111                         help();
 112                         return 0;
 113
 114                 case 'D':
 115                         free(arg_directory);
 116                         arg_directory = canonicalize_file_name(optarg);
 117                         if (!arg_directory) {
 118                                 log_error("Failed to canonicalize root directory.");
 119                                 return -ENOMEM;
 120                         }
 121
 122                         break;
 123
 124                 case 'u':
 125                         free(arg_user);
 126                         if (!(arg_user = strdup(optarg))) {
 127                                 log_error("Failed to duplicate user name.");
 128                                 return -ENOMEM;
 129                         }
 130
 131                         break;
 132
 133                 case 'C':
 134                         strv_free(arg_controllers);
 135                         arg_controllers = strv_split(optarg, ",");
 136                         if (!arg_controllers) {
 137                                 log_error("Failed to split controllers list.");
 138                                 return -ENOMEM;
 139                         }
 140                         strv_uniq(arg_controllers);
 141
 142                         break;
 143
 144                 case ARG_PRIVATE_NETWORK:
 145                         arg_private_network = true;
 146                         break;
 147
 148                 case 'b':
 149                         arg_boot = true;
 150                         break;
 151
 152                 case ARG_UUID:
 153                         arg_uuid = optarg;
 154                         break;
 155
 156                 case ARG_READ_ONLY:
 157                         arg_read_only = true;
 158                         break;
 159
 160                 case '?':
 161                         return -EINVAL;
 162
 163                 default:
 164                         log_error("Unknown option code %c", c);
 165                         return -EINVAL;
 166                 }
 167         }
 168
 169         return 1;
 170 }
 171
 172 static int mount_all(const char *dest) {
 173
 174         typedef struct MountPoint {
 175                 const char *what;
 176                 const char *where;
 177                 const char *type;
 178                 const char *options;
 179                 unsigned long flags;
 180                 bool fatal;
 181         } MountPoint;
 182
 183         static const MountPoint mount_table[] = {
 184                 { "proc",      "/proc",     "proc",  NULL,       MS_NOSUID|MS_NOEXEC|MS_NODEV, true  },
 185                 { "/proc/sys", "/proc/sys", "bind",  NULL,       MS_BIND, true                       },   /* Bind mount first */
 186                 { "/proc/sys", "/proc/sys", "bind",  NULL,       MS_BIND|MS_RDONLY|MS_REMOUNT, true  },   /* Then, make it r/o */
 187                 { "/sys",      "/sys",      "bind",  NULL,       MS_BIND,                      true  },   /* Bind mount first */
 188                 { "/sys",      "/sys",      "bind",  NULL,       MS_BIND|MS_RDONLY|MS_REMOUNT, true  },   /* Then, make it r/o */
 189                 { "tmpfs",     "/dev",      "tmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,     true  },
 190                 { "/dev/pts",  "/dev/pts",  "bind",  NULL,       MS_BIND,                      true  },
 191                 { "tmpfs",     "/run",      "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true  },
 192 #ifdef HAVE_SELINUX
 193                 { "/sys/fs/selinux", "/sys/fs/selinux", "bind", NULL, MS_BIND,                      false },  /* Bind mount first */
 194                 { "/sys/fs/selinux", "/sys/fs/selinux", "bind", NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, false },  /* Then, make it r/o */
 195 #endif
 196         };
 197
 198         unsigned k;
 199         int r = 0;
 200         char *where;
 201
 202         for (k = 0; k < ELEMENTSOF(mount_table); k++) {
 203                 int t;
 204
 205                 if (asprintf(&where, "%s/%s", dest, mount_table[k].where) < 0) {
 206                         log_error("Out of memory");
 207
 208                         if (r == 0)
 209                                 r = -ENOMEM;
 210
 211                         break;
 212                 }
 213
 214                 t = path_is_mount_point(where, false);
 215                 if (t < 0) {
 216                         log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
 217                         free(where);
 218
 219                         if (r == 0)
 220                                 r = t;
 221
 222                         continue;
 223                 }
 224
 225                 mkdir_p(where, 0755);
 226
 227                 if (mount(mount_table[k].what,
 228                           where,
 229                           mount_table[k].type,
 230                           mount_table[k].flags,
 231                           mount_table[k].options) < 0 &&
 232                     mount_table[k].fatal) {
 233
 234                         log_error("mount(%s) failed: %m", where);
 235
 236                         if (r == 0)
 237                                 r = -errno;
 238                 }
 239
 240                 free(where);
 241         }
 242
 243         return r;
 244 }
 245
 246 static int setup_timezone(const char *dest) {
 247         char *where;
 248
 249         assert(dest);
 250
 251         /* Fix the timezone, if possible */
 252         if (asprintf(&where, "%s/etc/localtime", dest) < 0) {
 253                 log_error("Out of memory");
 254                 return -ENOMEM;
 255         }
 256
 257         if (mount("/etc/localtime", where, "bind", MS_BIND, NULL) >= 0)
 258                 mount("/etc/localtime", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
 259
 260         free(where);
 261
 262         if (asprintf(&where, "%s/etc/timezone", dest) < 0) {
 263                 log_error("Out of memory");
 264                 return -ENOMEM;
 265         }
 266
 267         if (mount("/etc/timezone", where, "bind", MS_BIND, NULL) >= 0)
 268                 mount("/etc/timezone", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
 269
 270         free(where);
 271
 272         return 0;
 273 }
 274
 275 static int setup_resolv_conf(const char *dest) {
 276         char *where;
 277
 278         assert(dest);
 279
 280         if (arg_private_network)
 281                 return 0;
 282
 283         /* Fix resolv.conf, if possible */
 284         if (asprintf(&where, "%s/etc/resolv.conf", dest) < 0) {
 285                 log_error("Out of memory");
 286                 return -ENOMEM;
 287         }
 288
 289         if (mount("/etc/resolv.conf", where, "bind", MS_BIND, NULL) >= 0)
 290                 mount("/etc/resolv.conf", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
 291
 292         free(where);
 293
 294         return 0;
 295 }
 296
 297 static int copy_devnodes(const char *dest) {
 298
 299         static const char devnodes[] =
 300                 "null\0"
 301                 "zero\0"
 302                 "full\0"
 303                 "random\0"
 304                 "urandom\0"
 305                 "tty\0"
 306                 "ptmx\0"
 307                 "rtc0\0";
 308
 309         const char *d;
 310         int r = 0;
 311         mode_t u;
 312
 313         assert(dest);
 314
 315         u = umask(0000);
 316
 317         NULSTR_FOREACH(d, devnodes) {
 318                 struct stat st;
 319                 char *from = NULL, *to = NULL;
 320
 321                 asprintf(&from, "/dev/%s", d);
 322                 asprintf(&to, "%s/dev/%s", dest, d);
 323
 324                 if (!from || !to) {
 325                         log_error("Failed to allocate devnode path");
 326
 327                         free(from);
 328                         free(to);
 329
 330                         from = to = NULL;
 331
 332                         if (r == 0)
 333                                 r = -ENOMEM;
 334
 335                         break;
 336                 }
 337
 338                 if (stat(from, &st) < 0) {
 339
 340                         if (errno != ENOENT) {
 341                                 log_error("Failed to stat %s: %m", from);
 342                                 if (r == 0)
 343                                         r = -errno;
 344                         }
 345
 346                 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
 347
 348                         log_error("%s is not a char or block device, cannot copy.", from);
 349                         if (r == 0)
 350                                 r = -EIO;
 351
 352                 } else if (mknod(to, st.st_mode, st.st_rdev) < 0) {
 353
 354                         log_error("mknod(%s) failed: %m", dest);
 355                         if (r == 0)
 356                                 r = -errno;
 357                 }
 358
 359                 free(from);
 360                 free(to);
 361         }
 362
 363         umask(u);
 364
 365         return r;
 366 }
 367
 368 static int setup_dev_console(const char *dest, const char *console) {
 369         struct stat st;
 370         char *to = NULL;
 371         int r;
 372         mode_t u;
 373
 374         assert(dest);
 375         assert(console);
 376
 377         u = umask(0000);
 378
 379         if (stat(console, &st) < 0) {
 380                 log_error("Failed to stat %s: %m", console);
 381                 r = -errno;
 382                 goto finish;
 383
 384         } else if (!S_ISCHR(st.st_mode)) {
 385                 log_error("/dev/console is not a char device.");
 386                 r = -EIO;
 387                 goto finish;
 388         }
 389
 390         r = chmod_and_chown(console, 0600, 0, 0);
 391         if (r < 0) {
 392                 log_error("Failed to correct access mode for TTY: %s", strerror(-r));
 393                 goto finish;
 394         }
 395
 396         if (asprintf(&to, "%s/dev/console", dest) < 0) {
 397                 log_error("Out of memory");
 398                 r = -ENOMEM;
 399                 goto finish;
 400         }
 401
 402         /* We need to bind mount the right tty to /dev/console since
 403          * ptys can only exist on pts file systems. To have something
 404          * to bind mount things on we create a device node first, that
 405          * has the right major/minor (note that the major minor
 406          * doesn't actually matter here, since we mount it over
 407          * anyway). */
 408
 409         if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0) {
 410                 log_error("mknod() for /dev/console failed: %m");
 411                 r = -errno;
 412                 goto finish;
 413         }
 414
 415         if (mount(console, to, "bind", MS_BIND, NULL) < 0) {
 416                 log_error("Bind mount for /dev/console failed: %m");
 417                 r = -errno;
 418                 goto finish;
 419         }
 420
 421 finish:
 422         free(to);
 423         umask(u);
 424
 425         return r;
 426 }
 427
 428 static int setup_kmsg(const char *dest, int kmsg_socket) {
 429         char *from = NULL, *to = NULL;
 430         int r, fd, k;
 431         mode_t u;
 432         union {
 433                 struct cmsghdr cmsghdr;
 434                 uint8_t buf[CMSG_SPACE(sizeof(int))];
 435         } control;
 436         struct msghdr mh;
 437         struct cmsghdr *cmsg;
 438
 439         assert(dest);
 440         assert(kmsg_socket >= 0);
 441
 442         u = umask(0000);
 443
 444         /* We create the kmsg FIFO as /dev/kmsg, but immediately
 445          * delete it after bind mounting it to /proc/kmsg. While FIFOs
 446          * on the reading side behave very similar to /proc/kmsg,
 447          * their writing side behaves differently from /dev/kmsg in
 448          * that writing blocks when nothing is reading. In order to
 449          * avoid any problems with containers deadlocking due to this
 450          * we simply make /dev/kmsg unavailable to the container. */
 451         if (asprintf(&from, "%s/dev/kmsg", dest) < 0) {
 452                 log_error("Out of memory");
 453                 r = -ENOMEM;
 454                 goto finish;
 455         }
 456
 457         if (asprintf(&to, "%s/proc/kmsg", dest) < 0) {
 458                 log_error("Out of memory");
 459                 r = -ENOMEM;
 460                 goto finish;
 461         }
 462
 463         if (mkfifo(from, 0600) < 0) {
 464                 log_error("mkfifo() for /dev/kmsg failed: %m");
 465                 r = -errno;
 466                 goto finish;
 467         }
 468
 469         r = chmod_and_chown(from, 0600, 0, 0);
 470         if (r < 0) {
 471                 log_error("Failed to correct access mode for /dev/kmsg: %s", strerror(-r));
 472                 goto finish;
 473         }
 474
 475         if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
 476                 log_error("Bind mount for /proc/kmsg failed: %m");
 477                 r = -errno;
 478                 goto finish;
 479         }
 480
 481         fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
 482         if (fd < 0) {
 483                 log_error("Failed to open fifo: %m");
 484                 r = -errno;
 485                 goto finish;
 486         }
 487
 488         zero(mh);
 489         zero(control);
 490
 491         mh.msg_control = &control;
 492         mh.msg_controllen = sizeof(control);
 493
 494         cmsg = CMSG_FIRSTHDR(&mh);
 495         cmsg->cmsg_level = SOL_SOCKET;
 496         cmsg->cmsg_type = SCM_RIGHTS;
 497         cmsg->cmsg_len = CMSG_LEN(sizeof(int));
 498         memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
 499
 500         mh.msg_controllen = cmsg->cmsg_len;
 501
 502         /* Store away the fd in the socket, so that it stays open as
 503          * long as we run the child */
 504         k = sendmsg(kmsg_socket, &mh, MSG_DONTWAIT|MSG_NOSIGNAL);
 505         close_nointr_nofail(fd);
 506
 507         if (k < 0) {
 508                 log_error("Failed to send FIFO fd: %m");
 509                 r = -errno;
 510                 goto finish;
 511         }
 512
 513         /* And now make the FIFO unavailable as /dev/kmsg... */
 514         unlink(from);
 515
 516 finish:
 517         free(from);
 518         free(to);
 519         umask(u);
 520
 521         return r;
 522 }
 523
 524 static int setup_hostname(void) {
 525         char *hn;
 526         int r = 0;
 527
 528         hn = path_get_file_name(arg_directory);
 529         if (hn) {
 530                 hn = strdup(hn);
 531                 if (!hn)
 532                         return -ENOMEM;
 533
 534                 hostname_cleanup(hn);
 535
 536                 if (!isempty(hn))
 537                         if (sethostname(hn, strlen(hn)) < 0)
 538                                 r = -errno;
 539
 540                 free(hn);
 541         }
 542
 543         return r;
 544 }
 545
 546 static int drop_capabilities(void) {
 547         static const unsigned long retain[] = {
 548                 CAP_CHOWN,
 549                 CAP_DAC_OVERRIDE,
 550                 CAP_DAC_READ_SEARCH,
 551                 CAP_FOWNER,
 552                 CAP_FSETID,
 553                 CAP_IPC_OWNER,
 554                 CAP_KILL,
 555                 CAP_LEASE,
 556                 CAP_LINUX_IMMUTABLE,
 557                 CAP_NET_BIND_SERVICE,
 558                 CAP_NET_BROADCAST,
 559                 CAP_NET_RAW,
 560                 CAP_SETGID,
 561                 CAP_SETFCAP,
 562                 CAP_SETPCAP,
 563                 CAP_SETUID,
 564                 CAP_SYS_ADMIN,
 565                 CAP_SYS_CHROOT,
 566                 CAP_SYS_NICE,
 567                 CAP_SYS_PTRACE,
 568                 CAP_SYS_TTY_CONFIG
 569         };
 570
 571         unsigned long l;
 572
 573         for (l = 0; l <= cap_last_cap(); l++) {
 574                 unsigned i;
 575
 576                 for (i = 0; i < ELEMENTSOF(retain); i++)
 577                         if (retain[i] == l)
 578                                 break;
 579
 580                 if (i < ELEMENTSOF(retain))
 581                         continue;
 582
 583                 if (prctl(PR_CAPBSET_DROP, l) < 0) {
 584                         log_error("PR_CAPBSET_DROP failed: %m");
 585                         return -errno;
 586                 }
 587         }
 588
 589         return 0;
 590 }
 591
 592 static int is_os_tree(const char *path) {
 593         int r;
 594         char *p;
 595         /* We use /bin/sh as flag file if something is an OS */
 596
 597         if (asprintf(&p, "%s/bin/sh", path) < 0)
 598                 return -ENOMEM;
 599
 600         r = access(p, F_OK);
 601         free(p);
 602
 603         return r < 0 ? 0 : 1;
 604 }
 605
 606 static int process_pty(int master, sigset_t *mask) {
 607
 608         char in_buffer[LINE_MAX], out_buffer[LINE_MAX];
 609         size_t in_buffer_full = 0, out_buffer_full = 0;
 610         struct epoll_event stdin_ev, stdout_ev, master_ev, signal_ev;
 611         bool stdin_readable = false, stdout_writable = false, master_readable = false, master_writable = false;
 612         int ep = -1, signal_fd = -1, r;
 613
 614         fd_nonblock(STDIN_FILENO, 1);
 615         fd_nonblock(STDOUT_FILENO, 1);
 616         fd_nonblock(master, 1);
 617
 618         if ((signal_fd = signalfd(-1, mask, SFD_NONBLOCK|SFD_CLOEXEC)) < 0) {
 619                 log_error("signalfd(): %m");
 620                 r = -errno;
 621                 goto finish;
 622         }
 623
 624         if ((ep = epoll_create1(EPOLL_CLOEXEC)) < 0) {
 625                 log_error("Failed to create epoll: %m");
 626                 r = -errno;
 627                 goto finish;
 628         }
 629
 630         zero(stdin_ev);
 631         stdin_ev.events = EPOLLIN|EPOLLET;
 632         stdin_ev.data.fd = STDIN_FILENO;
 633
 634         zero(stdout_ev);
 635         stdout_ev.events = EPOLLOUT|EPOLLET;
 636         stdout_ev.data.fd = STDOUT_FILENO;
 637
 638         zero(master_ev);
 639         master_ev.events = EPOLLIN|EPOLLOUT|EPOLLET;
 640         master_ev.data.fd = master;
 641
 642         zero(signal_ev);
 643         signal_ev.events = EPOLLIN;
 644         signal_ev.data.fd = signal_fd;
 645
 646         if (epoll_ctl(ep, EPOLL_CTL_ADD, STDIN_FILENO, &stdin_ev) < 0 ||
 647             epoll_ctl(ep, EPOLL_CTL_ADD, STDOUT_FILENO, &stdout_ev) < 0 ||
 648             epoll_ctl(ep, EPOLL_CTL_ADD, master, &master_ev) < 0 ||
 649             epoll_ctl(ep, EPOLL_CTL_ADD, signal_fd, &signal_ev) < 0) {
 650                 log_error("Failed to regiser fds in epoll: %m");
 651                 r = -errno;
 652                 goto finish;
 653         }
 654
 655         for (;;) {
 656                 struct epoll_event ev[16];
 657                 ssize_t k;
 658                 int i, nfds;
 659
 660                 if ((nfds = epoll_wait(ep, ev, ELEMENTSOF(ev), -1)) < 0) {
 661
 662                         if (errno == EINTR || errno == EAGAIN)
 663                                 continue;
 664
 665                         log_error("epoll_wait(): %m");
 666                         r = -errno;
 667                         goto finish;
 668                 }
 669
 670                 assert(nfds >= 1);
 671
 672                 for (i = 0; i < nfds; i++) {
 673                         if (ev[i].data.fd == STDIN_FILENO) {
 674
 675                                 if (ev[i].events & (EPOLLIN|EPOLLHUP))
 676                                         stdin_readable = true;
 677
 678                         } else if (ev[i].data.fd == STDOUT_FILENO) {
 679
 680                                 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
 681                                         stdout_writable = true;
 682
 683                         } else if (ev[i].data.fd == master) {
 684
 685                                 if (ev[i].events & (EPOLLIN|EPOLLHUP))
 686                                         master_readable = true;
 687
 688                                 if (ev[i].events & (EPOLLOUT|EPOLLHUP))
 689                                         master_writable = true;
 690
 691                         } else if (ev[i].data.fd == signal_fd) {
 692                                 struct signalfd_siginfo sfsi;
 693                                 ssize_t n;
 694
 695                                 if ((n = read(signal_fd, &sfsi, sizeof(sfsi))) != sizeof(sfsi)) {
 696
 697                                         if (n >= 0) {
 698                                                 log_error("Failed to read from signalfd: invalid block size");
 699                                                 r = -EIO;
 700                                                 goto finish;
 701                                         }
 702
 703                                         if (errno != EINTR && errno != EAGAIN) {
 704                                                 log_error("Failed to read from signalfd: %m");
 705                                                 r = -errno;
 706                                                 goto finish;
 707                                         }
 708                                 } else {
 709
 710                                         if (sfsi.ssi_signo == SIGWINCH) {
 711                                                 struct winsize ws;
 712
 713                                                 /* The window size changed, let's forward that. */
 714                                                 if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
 715                                                         ioctl(master, TIOCSWINSZ, &ws);
 716                                         } else {
 717                                                 r = 0;
 718                                                 goto finish;
 719                                         }
 720                                 }
 721                         }
 722                 }
 723
 724                 while ((stdin_readable && in_buffer_full <= 0) ||
 725                        (master_writable && in_buffer_full > 0) ||
 726                        (master_readable && out_buffer_full <= 0) ||
 727                        (stdout_writable && out_buffer_full > 0)) {
 728
 729                         if (stdin_readable && in_buffer_full < LINE_MAX) {
 730
 731                                 if ((k = read(STDIN_FILENO, in_buffer + in_buffer_full, LINE_MAX - in_buffer_full)) < 0) {
 732
 733                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 734                                                 stdin_readable = false;
 735                                         else {
 736                                                 log_error("read(): %m");
 737                                                 r = -errno;
 738                                                 goto finish;
 739                                         }
 740                                 } else
 741                                         in_buffer_full += (size_t) k;
 742                         }
 743
 744                         if (master_writable && in_buffer_full > 0) {
 745
 746                                 if ((k = write(master, in_buffer, in_buffer_full)) < 0) {
 747
 748                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 749                                                 master_writable = false;
 750                                         else {
 751                                                 log_error("write(): %m");
 752                                                 r = -errno;
 753                                                 goto finish;
 754                                         }
 755
 756                                 } else {
 757                                         assert(in_buffer_full >= (size_t) k);
 758                                         memmove(in_buffer, in_buffer + k, in_buffer_full - k);
 759                                         in_buffer_full -= k;
 760                                 }
 761                         }
 762
 763                         if (master_readable && out_buffer_full < LINE_MAX) {
 764
 765                                 if ((k = read(master, out_buffer + out_buffer_full, LINE_MAX - out_buffer_full)) < 0) {
 766
 767                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 768                                                 master_readable = false;
 769                                         else {
 770                                                 log_error("read(): %m");
 771                                                 r = -errno;
 772                                                 goto finish;
 773                                         }
 774                                 }  else
 775                                         out_buffer_full += (size_t) k;
 776                         }
 777
 778                         if (stdout_writable && out_buffer_full > 0) {
 779
 780                                 if ((k = write(STDOUT_FILENO, out_buffer, out_buffer_full)) < 0) {
 781
 782                                         if (errno == EAGAIN || errno == EPIPE || errno == ECONNRESET || errno == EIO)
 783                                                 stdout_writable = false;
 784                                         else {
 785                                                 log_error("write(): %m");
 786                                                 r = -errno;
 787                                                 goto finish;
 788                                         }
 789
 790                                 } else {
 791                                         assert(out_buffer_full >= (size_t) k);
 792                                         memmove(out_buffer, out_buffer + k, out_buffer_full - k);
 793                                         out_buffer_full -= k;
 794                                 }
 795                         }
 796                 }
 797         }
 798
 799 finish:
 800         if (ep >= 0)
 801                 close_nointr_nofail(ep);
 802
 803         if (signal_fd >= 0)
 804                 close_nointr_nofail(signal_fd);
 805
 806         return r;
 807 }
 808
 809 int main(int argc, char *argv[]) {
 810         pid_t pid = 0;
 811         int r = EXIT_FAILURE, k;
 812         char *oldcg = NULL, *newcg = NULL;
 813         char **controller = NULL;
 814         int master = -1;
 815         const char *console = NULL;
 816         struct termios saved_attr, raw_attr;
 817         sigset_t mask;
 818         bool saved_attr_valid = false;
 819         struct winsize ws;
 820         int kmsg_socket_pair[2] = { -1, -1 };
 821
 822         log_parse_environment();
 823         log_open();
 824
 825         if ((r = parse_argv(argc, argv)) <= 0)
 826                 goto finish;
 827
 828         if (arg_directory) {
 829                 char *p;
 830
 831                 p = path_make_absolute_cwd(arg_directory);
 832                 free(arg_directory);
 833                 arg_directory = p;
 834         } else
 835                 arg_directory = get_current_dir_name();
 836
 837         if (!arg_directory) {
 838                 log_error("Failed to determine path");
 839                 goto finish;
 840         }
 841
 842         path_kill_slashes(arg_directory);
 843
 844         if (geteuid() != 0) {
 845                 log_error("Need to be root.");
 846                 goto finish;
 847         }
 848
 849         if (sd_booted() <= 0) {
 850                 log_error("Not running on a systemd system.");
 851                 goto finish;
 852         }
 853
 854         if (path_equal(arg_directory, "/")) {
 855                 log_error("Spawning container on root directory not supported.");
 856                 goto finish;
 857         }
 858
 859         if (is_os_tree(arg_directory) <= 0) {
 860                 log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
 861                 goto finish;
 862         }
 863
 864         if ((k = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &oldcg)) < 0) {
 865                 log_error("Failed to determine current cgroup: %s", strerror(-k));
 866                 goto finish;
 867         }
 868
 869         if (asprintf(&newcg, "%s/nspawn-%lu", oldcg, (unsigned long) getpid()) < 0) {
 870                 log_error("Failed to allocate cgroup path.");
 871                 goto finish;
 872         }
 873
 874         k = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, newcg, 0);
 875         if (k < 0)  {
 876                 log_error("Failed to create cgroup: %s", strerror(-k));
 877                 goto finish;
 878         }
 879
 880         STRV_FOREACH(controller,arg_controllers) {
 881                 k = cg_create_and_attach(*controller, newcg, 0);
 882                 if (k < 0)
 883                         log_warning("Failed to create cgroup in controller %s: %s", *controller, strerror(-k));
 884         }
 885
 886         if ((master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY)) < 0) {
 887                 log_error("Failed to acquire pseudo tty: %m");
 888                 goto finish;
 889         }
 890
 891         if (!(console = ptsname(master))) {
 892                 log_error("Failed to determine tty name: %m");
 893                 goto finish;
 894         }
 895
 896         log_info("Spawning namespace container on %s (console is %s).", arg_directory, console);
 897
 898         if (ioctl(STDIN_FILENO, TIOCGWINSZ, &ws) >= 0)
 899                 ioctl(master, TIOCSWINSZ, &ws);
 900
 901         if (unlockpt(master) < 0) {
 902                 log_error("Failed to unlock tty: %m");
 903                 goto finish;
 904         }
 905
 906         if (tcgetattr(STDIN_FILENO, &saved_attr) < 0) {
 907                 log_error("Failed to get terminal attributes: %m");
 908                 goto finish;
 909         }
 910
 911         saved_attr_valid = true;
 912
 913         raw_attr = saved_attr;
 914         cfmakeraw(&raw_attr);
 915         raw_attr.c_lflag &= ~ECHO;
 916
 917         if (tcsetattr(STDIN_FILENO, TCSANOW, &raw_attr) < 0) {
 918                 log_error("Failed to set terminal attributes: %m");
 919                 goto finish;
 920         }
 921
 922         if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_NONBLOCK|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0) {
 923                 log_error("Failed to create kmsg socket pair");
 924                 goto finish;
 925         }
 926
 927         assert_se(sigemptyset(&mask) == 0);
 928         sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
 929         assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
 930
 931         pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWIPC|CLONE_NEWNS|CLONE_NEWPID|CLONE_NEWUTS|(arg_private_network ? CLONE_NEWNET : 0), NULL);
 932         if (pid < 0) {
 933                 if (errno == EINVAL)
 934                         log_error("clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
 935                 else
 936                         log_error("clone() failed: %m");
 937
 938                 goto finish;
 939         }
 940
 941         if (pid == 0) {
 942                 /* child */
 943
 944                 const char *home = NULL;
 945                 uid_t uid = (uid_t) -1;
 946                 gid_t gid = (gid_t) -1;
 947                 const char *envp[] = {
 948                         "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
 949                         "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
 950                         NULL, /* TERM */
 951                         NULL, /* HOME */
 952                         NULL, /* USER */
 953                         NULL, /* LOGNAME */
 954                         NULL, /* container_uuid */
 955                         NULL
 956                 };
 957
 958                 envp[2] = strv_find_prefix(environ, "TERM=");
 959
 960                 close_nointr_nofail(master);
 961
 962                 close_nointr(STDIN_FILENO);
 963                 close_nointr(STDOUT_FILENO);
 964                 close_nointr(STDERR_FILENO);
 965
 966                 close_all_fds(&kmsg_socket_pair[1], 1);
 967
 968                 reset_all_signal_handlers();
 969
 970                 assert_se(sigemptyset(&mask) == 0);
 971                 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 972
 973                 if (setsid() < 0)
 974                         goto child_fail;
 975
 976                 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
 977                         goto child_fail;
 978
 979                 /* Mark / as private, in case somebody marked it shared */
 980                 if (mount(NULL, "/", NULL, MS_PRIVATE|MS_REC, NULL) < 0)
 981                         goto child_fail;
 982
 983                 /* Turn directory into bind mount */
 984                 if (mount(arg_directory, arg_directory, "bind", MS_BIND, NULL) < 0) {
 985                         log_error("Failed to make bind mount.");
 986                         goto child_fail;
 987                 }
 988
 989                 if (arg_read_only)
 990                         if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL) < 0) {
 991                                 log_error("Failed to make read-only.");
 992                                 goto child_fail;
 993                         }
 994
 995                 if (mount_all(arg_directory) < 0)
 996                         goto child_fail;
 997
 998                 if (copy_devnodes(arg_directory) < 0)
 999                         goto child_fail;
1000
1001                 if (setup_dev_console(arg_directory, console) < 0)
1002                         goto child_fail;
1003
1004                 if (setup_kmsg(arg_directory, kmsg_socket_pair[1]) < 0)
1005                         goto child_fail;
1006
1007                 close_nointr_nofail(kmsg_socket_pair[1]);
1008
1009                 if (setup_timezone(arg_directory) < 0)
1010                         goto child_fail;
1011
1012                 if (setup_resolv_conf(arg_directory) < 0)
1013                         goto child_fail;
1014
1015                 if (chdir(arg_directory) < 0) {
1016                         log_error("chdir(%s) failed: %m", arg_directory);
1017                         goto child_fail;
1018                 }
1019
1020                 if (open_terminal("dev/console", O_RDWR) != STDIN_FILENO ||
1021                     dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
1022                     dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
1023                         goto child_fail;
1024
1025                 if (mount(arg_directory, "/", "bind", MS_MOVE, NULL) < 0) {
1026                         log_error("mount(MS_BIND) failed: %m");
1027                         goto child_fail;
1028                 }
1029
1030                 if (chroot(".") < 0) {
1031                         log_error("chroot() failed: %m");
1032                         goto child_fail;
1033                 }
1034
1035                 if (chdir("/") < 0) {
1036                         log_error("chdir() failed: %m");
1037                         goto child_fail;
1038                 }
1039
1040                 umask(0022);
1041
1042                 loopback_setup();
1043
1044                 if (drop_capabilities() < 0)
1045                         goto child_fail;
1046
1047                 if (arg_user) {
1048
1049                         if (get_user_creds((const char**)&arg_user, &uid, &gid, &home) < 0) {
1050                                 log_error("get_user_creds() failed: %m");
1051                                 goto child_fail;
1052                         }
1053
1054                         if (mkdir_parents(home, 0775) < 0) {
1055                                 log_error("mkdir_parents() failed: %m");
1056                                 goto child_fail;
1057                         }
1058
1059                         if (safe_mkdir(home, 0775, uid, gid) < 0) {
1060                                 log_error("safe_mkdir() failed: %m");
1061                                 goto child_fail;
1062                         }
1063
1064                         if (initgroups((const char*)arg_user, gid) < 0) {
1065                                 log_error("initgroups() failed: %m");
1066                                 goto child_fail;
1067                         }
1068
1069                         if (setresgid(gid, gid, gid) < 0) {
1070                                 log_error("setregid() failed: %m");
1071                                 goto child_fail;
1072                         }
1073
1074                         if (setresuid(uid, uid, uid) < 0) {
1075                                 log_error("setreuid() failed: %m");
1076                                 goto child_fail;
1077                         }
1078                 }
1079
1080                 if ((asprintf((char**)(envp + 3), "HOME=%s", home ? home: "/root") < 0) ||
1081                     (asprintf((char**)(envp + 4), "USER=%s", arg_user ? arg_user : "root") < 0) ||
1082                     (asprintf((char**)(envp + 5), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)) {
1083                     log_error("Out of memory");
1084                     goto child_fail;
1085                 }
1086
1087                 if (arg_uuid) {
1088                         if (asprintf((char**)(envp + 6), "container_uuid=%s", arg_uuid) < 0) {
1089                                 log_error("Out of memory");
1090                                 goto child_fail;
1091                         }
1092                 }
1093
1094                 setup_hostname();
1095
1096                 if (arg_boot) {
1097                         char **a;
1098                         size_t l;
1099
1100                         /* Automatically search for the init system */
1101
1102                         l = 1 + argc - optind;
1103                         a = newa(char*, l + 1);
1104                         memcpy(a + 1, argv + optind, l * sizeof(char*));
1105
1106                         a[0] = (char*) "/usr/lib/systemd/systemd";
1107                         execve(a[0], a, (char**) envp);
1108
1109                         a[0] = (char*) "/lib/systemd/systemd";
1110                         execve(a[0], a, (char**) envp);
1111
1112                         a[0] = (char*) "/sbin/init";
1113                         execve(a[0], a, (char**) envp);
1114                 } else if (argc > optind)
1115                         execvpe(argv[optind], argv + optind, (char**) envp);
1116                 else {
1117                         chdir(home ? home : "/root");
1118                         execle("/bin/bash", "-bash", NULL, (char**) envp);
1119                 }
1120
1121                 log_error("execv() failed: %m");
1122
1123         child_fail:
1124                 _exit(EXIT_FAILURE);
1125         }
1126
1127         if (process_pty(master, &mask) < 0)
1128                 goto finish;
1129
1130         if (saved_attr_valid) {
1131                 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
1132                 saved_attr_valid = false;
1133         }
1134
1135         r = wait_for_terminate_and_warn(argc > optind ? argv[optind] : "bash", pid);
1136
1137         if (r < 0)
1138                 r = EXIT_FAILURE;
1139
1140 finish:
1141         if (saved_attr_valid)
1142                 tcsetattr(STDIN_FILENO, TCSANOW, &saved_attr);
1143
1144         if (master >= 0)
1145                 close_nointr_nofail(master);
1146
1147         close_pipe(kmsg_socket_pair);
1148
1149         if (oldcg)
1150                 cg_attach(SYSTEMD_CGROUP_CONTROLLER, oldcg, 0);
1151
1152         if (newcg)
1153                 cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, newcg, true);
1154
1155         free(arg_directory);
1156         strv_free(arg_controllers);
1157         free(oldcg);
1158         free(newcg);
1159
1160         return r;
1161 }