src/execute.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 General Public License as published by
  10   the Free Software Foundation; either version 2 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   General Public License for more details.
  17
  18   You should have received a copy of the GNU General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <assert.h>
  23 #include <dirent.h>
  24 #include <errno.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include <string.h>
  28 #include <signal.h>
  29 #include <sys/socket.h>
  30 #include <sys/un.h>
  31 #include <sys/prctl.h>
  32 #include <linux/sched.h>
  33 #include <sys/types.h>
  34 #include <sys/stat.h>
  35 #include <grp.h>
  36 #include <pwd.h>
  37 #include <sys/mount.h>
  38 #include <linux/fs.h>
  39 #include <linux/oom.h>
  40
  41 #ifdef HAVE_PAM
  42 #include <security/pam_appl.h>
  43 #endif
  44
  45 #include "execute.h"
  46 #include "strv.h"
  47 #include "macro.h"
  48 #include "util.h"
  49 #include "log.h"
  50 #include "ioprio.h"
  51 #include "securebits.h"
  52 #include "cgroup.h"
  53 #include "namespace.h"
  54 #include "tcpwrap.h"
  55 #include "exit-status.h"
  56 #include "missing.h"
  57 #include "utmp-wtmp.h"
  58 #include "def.h"
  59 #include "loopback-setup.h"
  60
  61 /* This assumes there is a 'tty' group */
  62 #define TTY_MODE 0620
  63
  64 static int shift_fds(int fds[], unsigned n_fds) {
  65         int start, restart_from;
  66
  67         if (n_fds <= 0)
  68                 return 0;
  69
  70         /* Modifies the fds array! (sorts it) */
  71
  72         assert(fds);
  73
  74         start = 0;
  75         for (;;) {
  76                 int i;
  77
  78                 restart_from = -1;
  79
  80                 for (i = start; i < (int) n_fds; i++) {
  81                         int nfd;
  82
  83                         /* Already at right index? */
  84                         if (fds[i] == i+3)
  85                                 continue;
  86
  87                         if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
  88                                 return -errno;
  89
  90                         close_nointr_nofail(fds[i]);
  91                         fds[i] = nfd;
  92
  93                         /* Hmm, the fd we wanted isn't free? Then
  94                          * let's remember that and try again from here*/
  95                         if (nfd != i+3 && restart_from < 0)
  96                                 restart_from = i;
  97                 }
  98
  99                 if (restart_from < 0)
 100                         break;
 101
 102                 start = restart_from;
 103         }
 104
 105         return 0;
 106 }
 107
 108 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
 109         unsigned i;
 110         int r;
 111
 112         if (n_fds <= 0)
 113                 return 0;
 114
 115         assert(fds);
 116
 117         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
 118
 119         for (i = 0; i < n_fds; i++) {
 120
 121                 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
 122                         return r;
 123
 124                 /* We unconditionally drop FD_CLOEXEC from the fds,
 125                  * since after all we want to pass these fds to our
 126                  * children */
 127
 128                 if ((r = fd_cloexec(fds[i], false)) < 0)
 129                         return r;
 130         }
 131
 132         return 0;
 133 }
 134
 135 static const char *tty_path(const ExecContext *context) {
 136         assert(context);
 137
 138         if (context->tty_path)
 139                 return context->tty_path;
 140
 141         return "/dev/console";
 142 }
 143
 144 void exec_context_tty_reset(const ExecContext *context) {
 145         assert(context);
 146
 147         if (context->tty_vhangup)
 148                 terminal_vhangup(tty_path(context));
 149
 150         if (context->tty_reset)
 151                 reset_terminal(tty_path(context));
 152
 153         if (context->tty_vt_disallocate && context->tty_path)
 154                 vt_disallocate(context->tty_path);
 155 }
 156
 157 static int open_null_as(int flags, int nfd) {
 158         int fd, r;
 159
 160         assert(nfd >= 0);
 161
 162         if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
 163                 return -errno;
 164
 165         if (fd != nfd) {
 166                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 167                 close_nointr_nofail(fd);
 168         } else
 169                 r = nfd;
 170
 171         return r;
 172 }
 173
 174 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
 175         int fd, r;
 176         union sockaddr_union sa;
 177
 178         assert(context);
 179         assert(output < _EXEC_OUTPUT_MAX);
 180         assert(ident);
 181         assert(nfd >= 0);
 182
 183         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 184         if (fd < 0)
 185                 return -errno;
 186
 187         zero(sa);
 188         sa.un.sun_family = AF_UNIX;
 189         strncpy(sa.un.sun_path, "/run/systemd/journal/stdout", sizeof(sa.un.sun_path));
 190
 191         r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
 192         if (r < 0) {
 193                 close_nointr_nofail(fd);
 194                 return -errno;
 195         }
 196
 197         if (shutdown(fd, SHUT_RD) < 0) {
 198                 close_nointr_nofail(fd);
 199                 return -errno;
 200         }
 201
 202         dprintf(fd,
 203                 "%s\n"
 204                 "%i\n"
 205                 "%i\n"
 206                 "%i\n"
 207                 "%i\n"
 208                 "%i\n",
 209                 context->syslog_identifier ? context->syslog_identifier : ident,
 210                 context->syslog_priority,
 211                 !!context->syslog_level_prefix,
 212                 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
 213                 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
 214                 output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || output == EXEC_OUTPUT_KMSG_AND_CONSOLE || output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE);
 215
 216         if (fd != nfd) {
 217                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 218                 close_nointr_nofail(fd);
 219         } else
 220                 r = nfd;
 221
 222         return r;
 223 }
 224 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
 225         int fd, r;
 226
 227         assert(path);
 228         assert(nfd >= 0);
 229
 230         if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
 231                 return fd;
 232
 233         if (fd != nfd) {
 234                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 235                 close_nointr_nofail(fd);
 236         } else
 237                 r = nfd;
 238
 239         return r;
 240 }
 241
 242 static bool is_terminal_input(ExecInput i) {
 243         return
 244                 i == EXEC_INPUT_TTY ||
 245                 i == EXEC_INPUT_TTY_FORCE ||
 246                 i == EXEC_INPUT_TTY_FAIL;
 247 }
 248
 249 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
 250
 251         if (is_terminal_input(std_input) && !apply_tty_stdin)
 252                 return EXEC_INPUT_NULL;
 253
 254         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 255                 return EXEC_INPUT_NULL;
 256
 257         return std_input;
 258 }
 259
 260 static int fixup_output(ExecOutput std_output, int socket_fd) {
 261
 262         if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 263                 return EXEC_OUTPUT_INHERIT;
 264
 265         return std_output;
 266 }
 267
 268 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
 269         ExecInput i;
 270
 271         assert(context);
 272
 273         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 274
 275         switch (i) {
 276
 277         case EXEC_INPUT_NULL:
 278                 return open_null_as(O_RDONLY, STDIN_FILENO);
 279
 280         case EXEC_INPUT_TTY:
 281         case EXEC_INPUT_TTY_FORCE:
 282         case EXEC_INPUT_TTY_FAIL: {
 283                 int fd, r;
 284
 285                 if ((fd = acquire_terminal(
 286                                      tty_path(context),
 287                                      i == EXEC_INPUT_TTY_FAIL,
 288                                      i == EXEC_INPUT_TTY_FORCE,
 289                                      false)) < 0)
 290                         return fd;
 291
 292                 if (fd != STDIN_FILENO) {
 293                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 294                         close_nointr_nofail(fd);
 295                 } else
 296                         r = STDIN_FILENO;
 297
 298                 return r;
 299         }
 300
 301         case EXEC_INPUT_SOCKET:
 302                 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 303
 304         default:
 305                 assert_not_reached("Unknown input type");
 306         }
 307 }
 308
 309 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 310         ExecOutput o;
 311         ExecInput i;
 312
 313         assert(context);
 314         assert(ident);
 315
 316         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 317         o = fixup_output(context->std_output, socket_fd);
 318
 319         /* This expects the input is already set up */
 320
 321         switch (o) {
 322
 323         case EXEC_OUTPUT_INHERIT:
 324
 325                 /* If input got downgraded, inherit the original value */
 326                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 327                         return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 328
 329                 /* If the input is connected to anything that's not a /dev/null, inherit that... */
 330                 if (i != EXEC_INPUT_NULL)
 331                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 332
 333                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 334                 if (getppid() != 1)
 335                         return STDOUT_FILENO;
 336
 337                 /* We need to open /dev/null here anew, to get the
 338                  * right access mode. So we fall through */
 339
 340         case EXEC_OUTPUT_NULL:
 341                 return open_null_as(O_WRONLY, STDOUT_FILENO);
 342
 343         case EXEC_OUTPUT_TTY:
 344                 if (is_terminal_input(i))
 345                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 346
 347                 /* We don't reset the terminal if this is just about output */
 348                 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 349
 350         case EXEC_OUTPUT_SYSLOG:
 351         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 352         case EXEC_OUTPUT_KMSG:
 353         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 354         case EXEC_OUTPUT_JOURNAL:
 355         case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
 356                 return connect_logger_as(context, o, ident, STDOUT_FILENO);
 357
 358         case EXEC_OUTPUT_SOCKET:
 359                 assert(socket_fd >= 0);
 360                 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 361
 362         default:
 363                 assert_not_reached("Unknown output type");
 364         }
 365 }
 366
 367 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 368         ExecOutput o, e;
 369         ExecInput i;
 370
 371         assert(context);
 372         assert(ident);
 373
 374         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 375         o = fixup_output(context->std_output, socket_fd);
 376         e = fixup_output(context->std_error, socket_fd);
 377
 378         /* This expects the input and output are already set up */
 379
 380         /* Don't change the stderr file descriptor if we inherit all
 381          * the way and are not on a tty */
 382         if (e == EXEC_OUTPUT_INHERIT &&
 383             o == EXEC_OUTPUT_INHERIT &&
 384             i == EXEC_INPUT_NULL &&
 385             !is_terminal_input(context->std_input) &&
 386             getppid () != 1)
 387                 return STDERR_FILENO;
 388
 389         /* Duplicate from stdout if possible */
 390         if (e == o || e == EXEC_OUTPUT_INHERIT)
 391                 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 392
 393         switch (e) {
 394
 395         case EXEC_OUTPUT_NULL:
 396                 return open_null_as(O_WRONLY, STDERR_FILENO);
 397
 398         case EXEC_OUTPUT_TTY:
 399                 if (is_terminal_input(i))
 400                         return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 401
 402                 /* We don't reset the terminal if this is just about output */
 403                 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
 404
 405         case EXEC_OUTPUT_SYSLOG:
 406         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 407         case EXEC_OUTPUT_KMSG:
 408         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 409         case EXEC_OUTPUT_JOURNAL:
 410         case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
 411                 return connect_logger_as(context, e, ident, STDERR_FILENO);
 412
 413         case EXEC_OUTPUT_SOCKET:
 414                 assert(socket_fd >= 0);
 415                 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 416
 417         default:
 418                 assert_not_reached("Unknown error type");
 419         }
 420 }
 421
 422 static int chown_terminal(int fd, uid_t uid) {
 423         struct stat st;
 424
 425         assert(fd >= 0);
 426
 427         /* This might fail. What matters are the results. */
 428         (void) fchown(fd, uid, -1);
 429         (void) fchmod(fd, TTY_MODE);
 430
 431         if (fstat(fd, &st) < 0)
 432                 return -errno;
 433
 434         if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
 435                 return -EPERM;
 436
 437         return 0;
 438 }
 439
 440 static int setup_confirm_stdio(const ExecContext *context,
 441                                int *_saved_stdin,
 442                                int *_saved_stdout) {
 443         int fd = -1, saved_stdin, saved_stdout = -1, r;
 444
 445         assert(context);
 446         assert(_saved_stdin);
 447         assert(_saved_stdout);
 448
 449         /* This returns positive EXIT_xxx return values instead of
 450          * negative errno style values! */
 451
 452         if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
 453                 return EXIT_STDIN;
 454
 455         if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
 456                 r = EXIT_STDOUT;
 457                 goto fail;
 458         }
 459
 460         if ((fd = acquire_terminal(
 461                              tty_path(context),
 462                              context->std_input == EXEC_INPUT_TTY_FAIL,
 463                              context->std_input == EXEC_INPUT_TTY_FORCE,
 464                              false)) < 0) {
 465                 r = EXIT_STDIN;
 466                 goto fail;
 467         }
 468
 469         if (chown_terminal(fd, getuid()) < 0) {
 470                 r = EXIT_STDIN;
 471                 goto fail;
 472         }
 473
 474         if (dup2(fd, STDIN_FILENO) < 0) {
 475                 r = EXIT_STDIN;
 476                 goto fail;
 477         }
 478
 479         if (dup2(fd, STDOUT_FILENO) < 0) {
 480                 r = EXIT_STDOUT;
 481                 goto fail;
 482         }
 483
 484         if (fd >= 2)
 485                 close_nointr_nofail(fd);
 486
 487         *_saved_stdin = saved_stdin;
 488         *_saved_stdout = saved_stdout;
 489
 490         return 0;
 491
 492 fail:
 493         if (saved_stdout >= 0)
 494                 close_nointr_nofail(saved_stdout);
 495
 496         if (saved_stdin >= 0)
 497                 close_nointr_nofail(saved_stdin);
 498
 499         if (fd >= 0)
 500                 close_nointr_nofail(fd);
 501
 502         return r;
 503 }
 504
 505 static int restore_confirm_stdio(const ExecContext *context,
 506                                  int *saved_stdin,
 507                                  int *saved_stdout,
 508                                  bool *keep_stdin,
 509                                  bool *keep_stdout) {
 510
 511         assert(context);
 512         assert(saved_stdin);
 513         assert(*saved_stdin >= 0);
 514         assert(saved_stdout);
 515         assert(*saved_stdout >= 0);
 516
 517         /* This returns positive EXIT_xxx return values instead of
 518          * negative errno style values! */
 519
 520         if (is_terminal_input(context->std_input)) {
 521
 522                 /* The service wants terminal input. */
 523
 524                 *keep_stdin = true;
 525                 *keep_stdout =
 526                         context->std_output == EXEC_OUTPUT_INHERIT ||
 527                         context->std_output == EXEC_OUTPUT_TTY;
 528
 529         } else {
 530                 /* If the service doesn't want a controlling terminal,
 531                  * then we need to get rid entirely of what we have
 532                  * already. */
 533
 534                 if (release_terminal() < 0)
 535                         return EXIT_STDIN;
 536
 537                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 538                         return EXIT_STDIN;
 539
 540                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 541                         return EXIT_STDOUT;
 542
 543                 *keep_stdout = *keep_stdin = false;
 544         }
 545
 546         return 0;
 547 }
 548
 549 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
 550         bool keep_groups = false;
 551         int r;
 552
 553         assert(context);
 554
 555         /* Lookup and set GID and supplementary group list. Here too
 556          * we avoid NSS lookups for gid=0. */
 557
 558         if (context->group || username) {
 559
 560                 if (context->group) {
 561                         const char *g = context->group;
 562
 563                         if ((r = get_group_creds(&g, &gid)) < 0)
 564                                 return r;
 565                 }
 566
 567                 /* First step, initialize groups from /etc/groups */
 568                 if (username && gid != 0) {
 569                         if (initgroups(username, gid) < 0)
 570                                 return -errno;
 571
 572                         keep_groups = true;
 573                 }
 574
 575                 /* Second step, set our gids */
 576                 if (setresgid(gid, gid, gid) < 0)
 577                         return -errno;
 578         }
 579
 580         if (context->supplementary_groups) {
 581                 int ngroups_max, k;
 582                 gid_t *gids;
 583                 char **i;
 584
 585                 /* Final step, initialize any manually set supplementary groups */
 586                 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
 587
 588                 if (!(gids = new(gid_t, ngroups_max)))
 589                         return -ENOMEM;
 590
 591                 if (keep_groups) {
 592                         if ((k = getgroups(ngroups_max, gids)) < 0) {
 593                                 free(gids);
 594                                 return -errno;
 595                         }
 596                 } else
 597                         k = 0;
 598
 599                 STRV_FOREACH(i, context->supplementary_groups) {
 600                         const char *g;
 601
 602                         if (k >= ngroups_max) {
 603                                 free(gids);
 604                                 return -E2BIG;
 605                         }
 606
 607                         g = *i;
 608                         r = get_group_creds(&g, gids+k);
 609                         if (r < 0) {
 610                                 free(gids);
 611                                 return r;
 612                         }
 613
 614                         k++;
 615                 }
 616
 617                 if (setgroups(k, gids) < 0) {
 618                         free(gids);
 619                         return -errno;
 620                 }
 621
 622                 free(gids);
 623         }
 624
 625         return 0;
 626 }
 627
 628 static int enforce_user(const ExecContext *context, uid_t uid) {
 629         int r;
 630         assert(context);
 631
 632         /* Sets (but doesn't lookup) the uid and make sure we keep the
 633          * capabilities while doing so. */
 634
 635         if (context->capabilities) {
 636                 cap_t d;
 637                 static const cap_value_t bits[] = {
 638                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
 639                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
 640                 };
 641
 642                 /* First step: If we need to keep capabilities but
 643                  * drop privileges we need to make sure we keep our
 644                  * caps, whiel we drop privileges. */
 645                 if (uid != 0) {
 646                         int sb = context->secure_bits|SECURE_KEEP_CAPS;
 647
 648                         if (prctl(PR_GET_SECUREBITS) != sb)
 649                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 650                                         return -errno;
 651                 }
 652
 653                 /* Second step: set the capabilities. This will reduce
 654                  * the capabilities to the minimum we need. */
 655
 656                 if (!(d = cap_dup(context->capabilities)))
 657                         return -errno;
 658
 659                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
 660                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
 661                         r = -errno;
 662                         cap_free(d);
 663                         return r;
 664                 }
 665
 666                 if (cap_set_proc(d) < 0) {
 667                         r = -errno;
 668                         cap_free(d);
 669                         return r;
 670                 }
 671
 672                 cap_free(d);
 673         }
 674
 675         /* Third step: actually set the uids */
 676         if (setresuid(uid, uid, uid) < 0)
 677                 return -errno;
 678
 679         /* At this point we should have all necessary capabilities but
 680            are otherwise a normal user. However, the caps might got
 681            corrupted due to the setresuid() so we need clean them up
 682            later. This is done outside of this call. */
 683
 684         return 0;
 685 }
 686
 687 #ifdef HAVE_PAM
 688
 689 static int null_conv(
 690                 int num_msg,
 691                 const struct pam_message **msg,
 692                 struct pam_response **resp,
 693                 void *appdata_ptr) {
 694
 695         /* We don't support conversations */
 696
 697         return PAM_CONV_ERR;
 698 }
 699
 700 static int setup_pam(
 701                 const char *name,
 702                 const char *user,
 703                 const char *tty,
 704                 char ***pam_env,
 705                 int fds[], unsigned n_fds) {
 706
 707         static const struct pam_conv conv = {
 708                 .conv = null_conv,
 709                 .appdata_ptr = NULL
 710         };
 711
 712         pam_handle_t *handle = NULL;
 713         sigset_t ss, old_ss;
 714         int pam_code = PAM_SUCCESS;
 715         int err;
 716         char **e = NULL;
 717         bool close_session = false;
 718         pid_t pam_pid = 0, parent_pid;
 719
 720         assert(name);
 721         assert(user);
 722         assert(pam_env);
 723
 724         /* We set up PAM in the parent process, then fork. The child
 725          * will then stay around until killed via PR_GET_PDEATHSIG or
 726          * systemd via the cgroup logic. It will then remove the PAM
 727          * session again. The parent process will exec() the actual
 728          * daemon. We do things this way to ensure that the main PID
 729          * of the daemon is the one we initially fork()ed. */
 730
 731         if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
 732                 handle = NULL;
 733                 goto fail;
 734         }
 735
 736         if (tty)
 737                 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
 738                         goto fail;
 739
 740         if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
 741                 goto fail;
 742
 743         if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
 744                 goto fail;
 745
 746         close_session = true;
 747
 748         if ((!(e = pam_getenvlist(handle)))) {
 749                 pam_code = PAM_BUF_ERR;
 750                 goto fail;
 751         }
 752
 753         /* Block SIGTERM, so that we know that it won't get lost in
 754          * the child */
 755         if (sigemptyset(&ss) < 0 ||
 756             sigaddset(&ss, SIGTERM) < 0 ||
 757             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 758                 goto fail;
 759
 760         parent_pid = getpid();
 761
 762         if ((pam_pid = fork()) < 0)
 763                 goto fail;
 764
 765         if (pam_pid == 0) {
 766                 int sig;
 767                 int r = EXIT_PAM;
 768
 769                 /* The child's job is to reset the PAM session on
 770                  * termination */
 771
 772                 /* This string must fit in 10 chars (i.e. the length
 773                  * of "/sbin/init") */
 774                 rename_process("sd(PAM)");
 775
 776                 /* Make sure we don't keep open the passed fds in this
 777                 child. We assume that otherwise only those fds are
 778                 open here that have been opened by PAM. */
 779                 close_many(fds, n_fds);
 780
 781                 /* Wait until our parent died. This will most likely
 782                  * not work since the kernel does not allow
 783                  * unprivileged parents kill their privileged children
 784                  * this way. We rely on the control groups kill logic
 785                  * to do the rest for us. */
 786                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 787                         goto child_finish;
 788
 789                 /* Check if our parent process might already have
 790                  * died? */
 791                 if (getppid() == parent_pid) {
 792                         for (;;) {
 793                                 if (sigwait(&ss, &sig) < 0) {
 794                                         if (errno == EINTR)
 795                                                 continue;
 796
 797                                         goto child_finish;
 798                                 }
 799
 800                                 assert(sig == SIGTERM);
 801                                 break;
 802                         }
 803                 }
 804
 805                 /* If our parent died we'll end the session */
 806                 if (getppid() != parent_pid)
 807                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
 808                                 goto child_finish;
 809
 810                 r = 0;
 811
 812         child_finish:
 813                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 814                 _exit(r);
 815         }
 816
 817         /* If the child was forked off successfully it will do all the
 818          * cleanups, so forget about the handle here. */
 819         handle = NULL;
 820
 821         /* Unblock SIGTERM again in the parent */
 822         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 823                 goto fail;
 824
 825         /* We close the log explicitly here, since the PAM modules
 826          * might have opened it, but we don't want this fd around. */
 827         closelog();
 828
 829         *pam_env = e;
 830         e = NULL;
 831
 832         return 0;
 833
 834 fail:
 835         if (pam_code != PAM_SUCCESS)
 836                 err = -EPERM;  /* PAM errors do not map to errno */
 837         else
 838                 err = -errno;
 839
 840         if (handle) {
 841                 if (close_session)
 842                         pam_code = pam_close_session(handle, PAM_DATA_SILENT);
 843
 844                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 845         }
 846
 847         strv_free(e);
 848
 849         closelog();
 850
 851         if (pam_pid > 1) {
 852                 kill(pam_pid, SIGTERM);
 853                 kill(pam_pid, SIGCONT);
 854         }
 855
 856         return err;
 857 }
 858 #endif
 859
 860 static int do_capability_bounding_set_drop(uint64_t drop) {
 861         unsigned long i;
 862         cap_t old_cap = NULL, new_cap = NULL;
 863         cap_flag_value_t fv;
 864         int r;
 865
 866         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 867          * in the effective set (yes, the kernel drops that when
 868          * executing init!), so get it back temporarily so that we can
 869          * call PR_CAPBSET_DROP. */
 870
 871         old_cap = cap_get_proc();
 872         if (!old_cap)
 873                 return -errno;
 874
 875         if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
 876                 r = -errno;
 877                 goto finish;
 878         }
 879
 880         if (fv != CAP_SET) {
 881                 static const cap_value_t v = CAP_SETPCAP;
 882
 883                 new_cap = cap_dup(old_cap);
 884                 if (!new_cap) {
 885                         r = -errno;
 886                         goto finish;
 887                 }
 888
 889                 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
 890                         r = -errno;
 891                         goto finish;
 892                 }
 893
 894                 if (cap_set_proc(new_cap) < 0) {
 895                         r = -errno;
 896                         goto finish;
 897                 }
 898         }
 899
 900         for (i = 0; i <= cap_last_cap(); i++)
 901                 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
 902                         if (prctl(PR_CAPBSET_DROP, i) < 0) {
 903                                 r = -errno;
 904                                 goto finish;
 905                         }
 906                 }
 907
 908         r = 0;
 909
 910 finish:
 911         if (new_cap)
 912                 cap_free(new_cap);
 913
 914         if (old_cap) {
 915                 cap_set_proc(old_cap);
 916                 cap_free(old_cap);
 917         }
 918
 919         return r;
 920 }
 921
 922 int exec_spawn(ExecCommand *command,
 923                char **argv,
 924                const ExecContext *context,
 925                int fds[], unsigned n_fds,
 926                char **environment,
 927                bool apply_permissions,
 928                bool apply_chroot,
 929                bool apply_tty_stdin,
 930                bool confirm_spawn,
 931                CGroupBonding *cgroup_bondings,
 932                CGroupAttribute *cgroup_attributes,
 933                pid_t *ret) {
 934
 935         pid_t pid;
 936         int r;
 937         char *line;
 938         int socket_fd;
 939         char **files_env = NULL;
 940
 941         assert(command);
 942         assert(context);
 943         assert(ret);
 944         assert(fds || n_fds <= 0);
 945
 946         if (context->std_input == EXEC_INPUT_SOCKET ||
 947             context->std_output == EXEC_OUTPUT_SOCKET ||
 948             context->std_error == EXEC_OUTPUT_SOCKET) {
 949
 950                 if (n_fds != 1)
 951                         return -EINVAL;
 952
 953                 socket_fd = fds[0];
 954
 955                 fds = NULL;
 956                 n_fds = 0;
 957         } else
 958                 socket_fd = -1;
 959
 960         if ((r = exec_context_load_environment(context, &files_env)) < 0) {
 961                 log_error("Failed to load environment files: %s", strerror(-r));
 962                 return r;
 963         }
 964
 965         if (!argv)
 966                 argv = command->argv;
 967
 968         if (!(line = exec_command_line(argv))) {
 969                 r = -ENOMEM;
 970                 goto fail_parent;
 971         }
 972
 973         log_debug("About to execute: %s", line);
 974         free(line);
 975
 976         r = cgroup_bonding_realize_list(cgroup_bondings);
 977         if (r < 0)
 978                 goto fail_parent;
 979
 980         cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
 981
 982         if ((pid = fork()) < 0) {
 983                 r = -errno;
 984                 goto fail_parent;
 985         }
 986
 987         if (pid == 0) {
 988                 int i, err;
 989                 sigset_t ss;
 990                 const char *username = NULL, *home = NULL;
 991                 uid_t uid = (uid_t) -1;
 992                 gid_t gid = (gid_t) -1;
 993                 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
 994                 unsigned n_env = 0;
 995                 int saved_stdout = -1, saved_stdin = -1;
 996                 bool keep_stdout = false, keep_stdin = false, set_access = false;
 997
 998                 /* child */
 999
1000                 /* This string must fit in 10 chars (i.e. the length
1001                  * of "/sbin/init") */
1002                 rename_process("sd(EXEC)");
1003
1004                 /* We reset exactly these signals, since they are the
1005                  * only ones we set to SIG_IGN in the main daemon. All
1006                  * others we leave untouched because we set them to
1007                  * SIG_DFL or a valid handler initially, both of which
1008                  * will be demoted to SIG_DFL. */
1009                 default_signals(SIGNALS_CRASH_HANDLER,
1010                                 SIGNALS_IGNORE, -1);
1011
1012                 if (sigemptyset(&ss) < 0 ||
1013                     sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1014                         err = -errno;
1015                         r = EXIT_SIGNAL_MASK;
1016                         goto fail_child;
1017                 }
1018
1019                 /* Close sockets very early to make sure we don't
1020                  * block init reexecution because it cannot bind its
1021                  * sockets */
1022                 log_forget_fds();
1023                 err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1024                                            socket_fd >= 0 ? 1 : n_fds);
1025                 if (err < 0) {
1026                         r = EXIT_FDS;
1027                         goto fail_child;
1028                 }
1029
1030                 if (!context->same_pgrp)
1031                         if (setsid() < 0) {
1032                                 err = -errno;
1033                                 r = EXIT_SETSID;
1034                                 goto fail_child;
1035                         }
1036
1037                 if (context->tcpwrap_name) {
1038                         if (socket_fd >= 0)
1039                                 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1040                                         err = -EACCES;
1041                                         r = EXIT_TCPWRAP;
1042                                         goto fail_child;
1043                                 }
1044
1045                         for (i = 0; i < (int) n_fds; i++) {
1046                                 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1047                                         err = -EACCES;
1048                                         r = EXIT_TCPWRAP;
1049                                         goto fail_child;
1050                                 }
1051                         }
1052                 }
1053
1054                 exec_context_tty_reset(context);
1055
1056                 /* We skip the confirmation step if we shall not apply the TTY */
1057                 if (confirm_spawn &&
1058                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1059                         char response;
1060
1061                         /* Set up terminal for the question */
1062                         if ((r = setup_confirm_stdio(context,
1063                                                      &saved_stdin, &saved_stdout))) {
1064                                 err = -errno;
1065                                 goto fail_child;
1066                         }
1067
1068                         /* Now ask the question. */
1069                         if (!(line = exec_command_line(argv))) {
1070                                 err = -ENOMEM;
1071                                 r = EXIT_MEMORY;
1072                                 goto fail_child;
1073                         }
1074
1075                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1076                         free(line);
1077
1078                         if (r < 0 || response == 'n') {
1079                                 err = -ECANCELED;
1080                                 r = EXIT_CONFIRM;
1081                                 goto fail_child;
1082                         } else if (response == 's') {
1083                                 err = r = 0;
1084                                 goto fail_child;
1085                         }
1086
1087                         /* Release terminal for the question */
1088                         if ((r = restore_confirm_stdio(context,
1089                                                        &saved_stdin, &saved_stdout,
1090                                                        &keep_stdin, &keep_stdout))) {
1091                                 err = -errno;
1092                                 goto fail_child;
1093                         }
1094                 }
1095
1096                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1097                  * must sure to drop O_NONBLOCK */
1098                 if (socket_fd >= 0)
1099                         fd_nonblock(socket_fd, false);
1100
1101                 if (!keep_stdin) {
1102                         err = setup_input(context, socket_fd, apply_tty_stdin);
1103                         if (err < 0) {
1104                                 r = EXIT_STDIN;
1105                                 goto fail_child;
1106                         }
1107                 }
1108
1109                 if (!keep_stdout) {
1110                         err = setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1111                         if (err < 0) {
1112                                 r = EXIT_STDOUT;
1113                                 goto fail_child;
1114                         }
1115                 }
1116
1117                 err = setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1118                 if (err < 0) {
1119                         r = EXIT_STDERR;
1120                         goto fail_child;
1121                 }
1122
1123                 if (cgroup_bondings) {
1124                         err = cgroup_bonding_install_list(cgroup_bondings, 0);
1125                         if (err < 0) {
1126                                 r = EXIT_CGROUP;
1127                                 goto fail_child;
1128                         }
1129                 }
1130
1131                 if (context->oom_score_adjust_set) {
1132                         char t[16];
1133
1134                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1135                         char_array_0(t);
1136
1137                         if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1138                                 /* Compatibility with Linux <= 2.6.35 */
1139
1140                                 int adj;
1141
1142                                 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1143                                 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1144
1145                                 snprintf(t, sizeof(t), "%i", adj);
1146                                 char_array_0(t);
1147
1148                                 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1149                                     && errno != EACCES) {
1150                                         err = -errno;
1151                                         r = EXIT_OOM_ADJUST;
1152                                         goto fail_child;
1153                                 }
1154                         }
1155                 }
1156
1157                 if (context->nice_set)
1158                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1159                                 err = -errno;
1160                                 r = EXIT_NICE;
1161                                 goto fail_child;
1162                         }
1163
1164                 if (context->cpu_sched_set) {
1165                         struct sched_param param;
1166
1167                         zero(param);
1168                         param.sched_priority = context->cpu_sched_priority;
1169
1170                         if (sched_setscheduler(0, context->cpu_sched_policy |
1171                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
1172                                 err = -errno;
1173                                 r = EXIT_SETSCHEDULER;
1174                                 goto fail_child;
1175                         }
1176                 }
1177
1178                 if (context->cpuset)
1179                         if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1180                                 err = -errno;
1181                                 r = EXIT_CPUAFFINITY;
1182                                 goto fail_child;
1183                         }
1184
1185                 if (context->ioprio_set)
1186                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1187                                 err = -errno;
1188                                 r = EXIT_IOPRIO;
1189                                 goto fail_child;
1190                         }
1191
1192                 if (context->timer_slack_nsec_set)
1193                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1194                                 err = -errno;
1195                                 r = EXIT_TIMERSLACK;
1196                                 goto fail_child;
1197                         }
1198
1199                 if (context->utmp_id)
1200                         utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1201
1202                 if (context->user) {
1203                         username = context->user;
1204                         err = get_user_creds(&username, &uid, &gid, &home);
1205                         if (err < 0) {
1206                                 r = EXIT_USER;
1207                                 goto fail_child;
1208                         }
1209
1210                         if (is_terminal_input(context->std_input)) {
1211                                 err = chown_terminal(STDIN_FILENO, uid);
1212                                 if (err < 0) {
1213                                         r = EXIT_STDIN;
1214                                         goto fail_child;
1215                                 }
1216                         }
1217
1218                         if (cgroup_bondings && context->control_group_modify) {
1219                                 err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
1220                                 if (err >= 0)
1221                                         err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid, context->control_group_persistent);
1222                                 if (err < 0) {
1223                                         r = EXIT_CGROUP;
1224                                         goto fail_child;
1225                                 }
1226
1227                                 set_access = true;
1228                         }
1229                 }
1230
1231                 if (cgroup_bondings && !set_access && context->control_group_persistent >= 0)  {
1232                         err = cgroup_bonding_set_task_access_list(cgroup_bondings, (mode_t) -1, (uid_t) -1, (uid_t) -1, context->control_group_persistent);
1233                         if (err < 0) {
1234                                 r = EXIT_CGROUP;
1235                                 goto fail_child;
1236                         }
1237                 }
1238
1239                 if (apply_permissions) {
1240                         err = enforce_groups(context, username, gid);
1241                         if (err < 0) {
1242                                 r = EXIT_GROUP;
1243                                 goto fail_child;
1244                         }
1245                 }
1246
1247                 umask(context->umask);
1248
1249 #ifdef HAVE_PAM
1250                 if (context->pam_name && username) {
1251                         err = setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds);
1252                         if (err < 0) {
1253                                 r = EXIT_PAM;
1254                                 goto fail_child;
1255                         }
1256                 }
1257 #endif
1258                 if (context->private_network) {
1259                         if (unshare(CLONE_NEWNET) < 0) {
1260                                 err = -errno;
1261                                 r = EXIT_NETWORK;
1262                                 goto fail_child;
1263                         }
1264
1265                         loopback_setup();
1266                 }
1267
1268                 if (strv_length(context->read_write_dirs) > 0 ||
1269                     strv_length(context->read_only_dirs) > 0 ||
1270                     strv_length(context->inaccessible_dirs) > 0 ||
1271                     context->mount_flags != MS_SHARED ||
1272                     context->private_tmp) {
1273                         err = setup_namespace(context->read_write_dirs,
1274                                               context->read_only_dirs,
1275                                               context->inaccessible_dirs,
1276                                               context->private_tmp,
1277                                               context->mount_flags);
1278                         if (err < 0) {
1279                                 r = EXIT_NAMESPACE;
1280                                 goto fail_child;
1281                         }
1282                 }
1283
1284                 if (apply_chroot) {
1285                         if (context->root_directory)
1286                                 if (chroot(context->root_directory) < 0) {
1287                                         err = -errno;
1288                                         r = EXIT_CHROOT;
1289                                         goto fail_child;
1290                                 }
1291
1292                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1293                                 err = -errno;
1294                                 r = EXIT_CHDIR;
1295                                 goto fail_child;
1296                         }
1297                 } else {
1298
1299                         char *d;
1300
1301                         if (asprintf(&d, "%s/%s",
1302                                      context->root_directory ? context->root_directory : "",
1303                                      context->working_directory ? context->working_directory : "") < 0) {
1304                                 err = -ENOMEM;
1305                                 r = EXIT_MEMORY;
1306                                 goto fail_child;
1307                         }
1308
1309                         if (chdir(d) < 0) {
1310                                 err = -errno;
1311                                 free(d);
1312                                 r = EXIT_CHDIR;
1313                                 goto fail_child;
1314                         }
1315
1316                         free(d);
1317                 }
1318
1319                 /* We repeat the fd closing here, to make sure that
1320                  * nothing is leaked from the PAM modules */
1321                 err = close_all_fds(fds, n_fds);
1322                 if (err >= 0)
1323                         err = shift_fds(fds, n_fds);
1324                 if (err >= 0)
1325                         err = flags_fds(fds, n_fds, context->non_blocking);
1326                 if (err < 0) {
1327                         r = EXIT_FDS;
1328                         goto fail_child;
1329                 }
1330
1331                 if (apply_permissions) {
1332
1333                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1334                                 if (!context->rlimit[i])
1335                                         continue;
1336
1337                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1338                                         err = -errno;
1339                                         r = EXIT_LIMITS;
1340                                         goto fail_child;
1341                                 }
1342                         }
1343
1344                         if (context->capability_bounding_set_drop) {
1345                                 err = do_capability_bounding_set_drop(context->capability_bounding_set_drop);
1346                                 if (err < 0) {
1347                                         r = EXIT_CAPABILITIES;
1348                                         goto fail_child;
1349                                 }
1350                         }
1351
1352                         if (context->user) {
1353                                 err = enforce_user(context, uid);
1354                                 if (err < 0) {
1355                                         r = EXIT_USER;
1356                                         goto fail_child;
1357                                 }
1358                         }
1359
1360                         /* PR_GET_SECUREBITS is not privileged, while
1361                          * PR_SET_SECUREBITS is. So to suppress
1362                          * potential EPERMs we'll try not to call
1363                          * PR_SET_SECUREBITS unless necessary. */
1364                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1365                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1366                                         err = -errno;
1367                                         r = EXIT_SECUREBITS;
1368                                         goto fail_child;
1369                                 }
1370
1371                         if (context->capabilities)
1372                                 if (cap_set_proc(context->capabilities) < 0) {
1373                                         err = -errno;
1374                                         r = EXIT_CAPABILITIES;
1375                                         goto fail_child;
1376                                 }
1377                 }
1378
1379                 if (!(our_env = new0(char*, 7))) {
1380                         err = -ENOMEM;
1381                         r = EXIT_MEMORY;
1382                         goto fail_child;
1383                 }
1384
1385                 if (n_fds > 0)
1386                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1387                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1388                                 err = -ENOMEM;
1389                                 r = EXIT_MEMORY;
1390                                 goto fail_child;
1391                         }
1392
1393                 if (home)
1394                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1395                                 err = -ENOMEM;
1396                                 r = EXIT_MEMORY;
1397                                 goto fail_child;
1398                         }
1399
1400                 if (username)
1401                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1402                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1403                                 err = -ENOMEM;
1404                                 r = EXIT_MEMORY;
1405                                 goto fail_child;
1406                         }
1407
1408                 if (is_terminal_input(context->std_input) ||
1409                     context->std_output == EXEC_OUTPUT_TTY ||
1410                     context->std_error == EXEC_OUTPUT_TTY)
1411                         if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1412                                 err = -ENOMEM;
1413                                 r = EXIT_MEMORY;
1414                                 goto fail_child;
1415                         }
1416
1417                 assert(n_env <= 7);
1418
1419                 if (!(final_env = strv_env_merge(
1420                                       5,
1421                                       environment,
1422                                       our_env,
1423                                       context->environment,
1424                                       files_env,
1425                                       pam_env,
1426                                       NULL))) {
1427                         err = -ENOMEM;
1428                         r = EXIT_MEMORY;
1429                         goto fail_child;
1430                 }
1431
1432                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1433                         err = -ENOMEM;
1434                         r = EXIT_MEMORY;
1435                         goto fail_child;
1436                 }
1437
1438                 final_env = strv_env_clean(final_env);
1439
1440                 execve(command->path, final_argv, final_env);
1441                 err = -errno;
1442                 r = EXIT_EXEC;
1443
1444         fail_child:
1445                 if (r != 0) {
1446                         log_open();
1447                         log_warning("Failed at step %s spawning %s: %s",
1448                                     exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1449                                     command->path, strerror(-err));
1450                 }
1451
1452                 strv_free(our_env);
1453                 strv_free(final_env);
1454                 strv_free(pam_env);
1455                 strv_free(files_env);
1456                 strv_free(final_argv);
1457
1458                 if (saved_stdin >= 0)
1459                         close_nointr_nofail(saved_stdin);
1460
1461                 if (saved_stdout >= 0)
1462                         close_nointr_nofail(saved_stdout);
1463
1464                 _exit(r);
1465         }
1466
1467         strv_free(files_env);
1468
1469         /* We add the new process to the cgroup both in the child (so
1470          * that we can be sure that no user code is ever executed
1471          * outside of the cgroup) and in the parent (so that we can be
1472          * sure that when we kill the cgroup the process will be
1473          * killed too). */
1474         if (cgroup_bondings)
1475                 cgroup_bonding_install_list(cgroup_bondings, pid);
1476
1477         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1478
1479         exec_status_start(&command->exec_status, pid);
1480
1481         *ret = pid;
1482         return 0;
1483
1484 fail_parent:
1485         strv_free(files_env);
1486
1487         return r;
1488 }
1489
1490 void exec_context_init(ExecContext *c) {
1491         assert(c);
1492
1493         c->umask = 0022;
1494         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1495         c->cpu_sched_policy = SCHED_OTHER;
1496         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1497         c->syslog_level_prefix = true;
1498         c->mount_flags = MS_SHARED;
1499         c->kill_signal = SIGTERM;
1500         c->send_sigkill = true;
1501         c->control_group_persistent = -1;
1502 }
1503
1504 void exec_context_done(ExecContext *c) {
1505         unsigned l;
1506
1507         assert(c);
1508
1509         strv_free(c->environment);
1510         c->environment = NULL;
1511
1512         strv_free(c->environment_files);
1513         c->environment_files = NULL;
1514
1515         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1516                 free(c->rlimit[l]);
1517                 c->rlimit[l] = NULL;
1518         }
1519
1520         free(c->working_directory);
1521         c->working_directory = NULL;
1522         free(c->root_directory);
1523         c->root_directory = NULL;
1524
1525         free(c->tty_path);
1526         c->tty_path = NULL;
1527
1528         free(c->tcpwrap_name);
1529         c->tcpwrap_name = NULL;
1530
1531         free(c->syslog_identifier);
1532         c->syslog_identifier = NULL;
1533
1534         free(c->user);
1535         c->user = NULL;
1536
1537         free(c->group);
1538         c->group = NULL;
1539
1540         strv_free(c->supplementary_groups);
1541         c->supplementary_groups = NULL;
1542
1543         free(c->pam_name);
1544         c->pam_name = NULL;
1545
1546         if (c->capabilities) {
1547                 cap_free(c->capabilities);
1548                 c->capabilities = NULL;
1549         }
1550
1551         strv_free(c->read_only_dirs);
1552         c->read_only_dirs = NULL;
1553
1554         strv_free(c->read_write_dirs);
1555         c->read_write_dirs = NULL;
1556
1557         strv_free(c->inaccessible_dirs);
1558         c->inaccessible_dirs = NULL;
1559
1560         if (c->cpuset)
1561                 CPU_FREE(c->cpuset);
1562
1563         free(c->utmp_id);
1564         c->utmp_id = NULL;
1565 }
1566
1567 void exec_command_done(ExecCommand *c) {
1568         assert(c);
1569
1570         free(c->path);
1571         c->path = NULL;
1572
1573         strv_free(c->argv);
1574         c->argv = NULL;
1575 }
1576
1577 void exec_command_done_array(ExecCommand *c, unsigned n) {
1578         unsigned i;
1579
1580         for (i = 0; i < n; i++)
1581                 exec_command_done(c+i);
1582 }
1583
1584 void exec_command_free_list(ExecCommand *c) {
1585         ExecCommand *i;
1586
1587         while ((i = c)) {
1588                 LIST_REMOVE(ExecCommand, command, c, i);
1589                 exec_command_done(i);
1590                 free(i);
1591         }
1592 }
1593
1594 void exec_command_free_array(ExecCommand **c, unsigned n) {
1595         unsigned i;
1596
1597         for (i = 0; i < n; i++) {
1598                 exec_command_free_list(c[i]);
1599                 c[i] = NULL;
1600         }
1601 }
1602
1603 int exec_context_load_environment(const ExecContext *c, char ***l) {
1604         char **i, **r = NULL;
1605
1606         assert(c);
1607         assert(l);
1608
1609         STRV_FOREACH(i, c->environment_files) {
1610                 char *fn;
1611                 int k;
1612                 bool ignore = false;
1613                 char **p;
1614
1615                 fn = *i;
1616
1617                 if (fn[0] == '-') {
1618                         ignore = true;
1619                         fn ++;
1620                 }
1621
1622                 if (!path_is_absolute(fn)) {
1623
1624                         if (ignore)
1625                                 continue;
1626
1627                         strv_free(r);
1628                         return -EINVAL;
1629                 }
1630
1631                 if ((k = load_env_file(fn, &p)) < 0) {
1632
1633                         if (ignore)
1634                                 continue;
1635
1636                         strv_free(r);
1637                         return k;
1638                 }
1639
1640                 if (r == NULL)
1641                         r = p;
1642                 else {
1643                         char **m;
1644
1645                         m = strv_env_merge(2, r, p);
1646                         strv_free(r);
1647                         strv_free(p);
1648
1649                         if (!m)
1650                                 return -ENOMEM;
1651
1652                         r = m;
1653                 }
1654         }
1655
1656         *l = r;
1657
1658         return 0;
1659 }
1660
1661 static void strv_fprintf(FILE *f, char **l) {
1662         char **g;
1663
1664         assert(f);
1665
1666         STRV_FOREACH(g, l)
1667                 fprintf(f, " %s", *g);
1668 }
1669
1670 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1671         char ** e;
1672         unsigned i;
1673
1674         assert(c);
1675         assert(f);
1676
1677         if (!prefix)
1678                 prefix = "";
1679
1680         fprintf(f,
1681                 "%sUMask: %04o\n"
1682                 "%sWorkingDirectory: %s\n"
1683                 "%sRootDirectory: %s\n"
1684                 "%sNonBlocking: %s\n"
1685                 "%sPrivateTmp: %s\n"
1686                 "%sControlGroupModify: %s\n"
1687                 "%sControlGroupPersistent: %s\n"
1688                 "%sPrivateNetwork: %s\n",
1689                 prefix, c->umask,
1690                 prefix, c->working_directory ? c->working_directory : "/",
1691                 prefix, c->root_directory ? c->root_directory : "/",
1692                 prefix, yes_no(c->non_blocking),
1693                 prefix, yes_no(c->private_tmp),
1694                 prefix, yes_no(c->control_group_modify),
1695                 prefix, yes_no(c->control_group_persistent),
1696                 prefix, yes_no(c->private_network));
1697
1698         STRV_FOREACH(e, c->environment)
1699                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1700
1701         STRV_FOREACH(e, c->environment_files)
1702                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1703
1704         if (c->tcpwrap_name)
1705                 fprintf(f,
1706                         "%sTCPWrapName: %s\n",
1707                         prefix, c->tcpwrap_name);
1708
1709         if (c->nice_set)
1710                 fprintf(f,
1711                         "%sNice: %i\n",
1712                         prefix, c->nice);
1713
1714         if (c->oom_score_adjust_set)
1715                 fprintf(f,
1716                         "%sOOMScoreAdjust: %i\n",
1717                         prefix, c->oom_score_adjust);
1718
1719         for (i = 0; i < RLIM_NLIMITS; i++)
1720                 if (c->rlimit[i])
1721                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1722
1723         if (c->ioprio_set)
1724                 fprintf(f,
1725                         "%sIOSchedulingClass: %s\n"
1726                         "%sIOPriority: %i\n",
1727                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1728                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1729
1730         if (c->cpu_sched_set)
1731                 fprintf(f,
1732                         "%sCPUSchedulingPolicy: %s\n"
1733                         "%sCPUSchedulingPriority: %i\n"
1734                         "%sCPUSchedulingResetOnFork: %s\n",
1735                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1736                         prefix, c->cpu_sched_priority,
1737                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1738
1739         if (c->cpuset) {
1740                 fprintf(f, "%sCPUAffinity:", prefix);
1741                 for (i = 0; i < c->cpuset_ncpus; i++)
1742                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1743                                 fprintf(f, " %i", i);
1744                 fputs("\n", f);
1745         }
1746
1747         if (c->timer_slack_nsec_set)
1748                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1749
1750         fprintf(f,
1751                 "%sStandardInput: %s\n"
1752                 "%sStandardOutput: %s\n"
1753                 "%sStandardError: %s\n",
1754                 prefix, exec_input_to_string(c->std_input),
1755                 prefix, exec_output_to_string(c->std_output),
1756                 prefix, exec_output_to_string(c->std_error));
1757
1758         if (c->tty_path)
1759                 fprintf(f,
1760                         "%sTTYPath: %s\n"
1761                         "%sTTYReset: %s\n"
1762                         "%sTTYVHangup: %s\n"
1763                         "%sTTYVTDisallocate: %s\n",
1764                         prefix, c->tty_path,
1765                         prefix, yes_no(c->tty_reset),
1766                         prefix, yes_no(c->tty_vhangup),
1767                         prefix, yes_no(c->tty_vt_disallocate));
1768
1769         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL ||
1770             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
1771             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL ||
1772             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE)
1773                 fprintf(f,
1774                         "%sSyslogFacility: %s\n"
1775                         "%sSyslogLevel: %s\n",
1776                         prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1777                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1778
1779         if (c->capabilities) {
1780                 char *t;
1781                 if ((t = cap_to_text(c->capabilities, NULL))) {
1782                         fprintf(f, "%sCapabilities: %s\n",
1783                                 prefix, t);
1784                         cap_free(t);
1785                 }
1786         }
1787
1788         if (c->secure_bits)
1789                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1790                         prefix,
1791                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1792                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1793                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1794                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1795                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1796                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1797
1798         if (c->capability_bounding_set_drop) {
1799                 unsigned long l;
1800                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1801
1802                 for (l = 0; l <= cap_last_cap(); l++)
1803                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1804                                 char *t;
1805
1806                                 if ((t = cap_to_name(l))) {
1807                                         fprintf(f, " %s", t);
1808                                         cap_free(t);
1809                                 }
1810                         }
1811
1812                 fputs("\n", f);
1813         }
1814
1815         if (c->user)
1816                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1817         if (c->group)
1818                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1819
1820         if (strv_length(c->supplementary_groups) > 0) {
1821                 fprintf(f, "%sSupplementaryGroups:", prefix);
1822                 strv_fprintf(f, c->supplementary_groups);
1823                 fputs("\n", f);
1824         }
1825
1826         if (c->pam_name)
1827                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1828
1829         if (strv_length(c->read_write_dirs) > 0) {
1830                 fprintf(f, "%sReadWriteDirs:", prefix);
1831                 strv_fprintf(f, c->read_write_dirs);
1832                 fputs("\n", f);
1833         }
1834
1835         if (strv_length(c->read_only_dirs) > 0) {
1836                 fprintf(f, "%sReadOnlyDirs:", prefix);
1837                 strv_fprintf(f, c->read_only_dirs);
1838                 fputs("\n", f);
1839         }
1840
1841         if (strv_length(c->inaccessible_dirs) > 0) {
1842                 fprintf(f, "%sInaccessibleDirs:", prefix);
1843                 strv_fprintf(f, c->inaccessible_dirs);
1844                 fputs("\n", f);
1845         }
1846
1847         fprintf(f,
1848                 "%sKillMode: %s\n"
1849                 "%sKillSignal: SIG%s\n"
1850                 "%sSendSIGKILL: %s\n",
1851                 prefix, kill_mode_to_string(c->kill_mode),
1852                 prefix, signal_to_string(c->kill_signal),
1853                 prefix, yes_no(c->send_sigkill));
1854
1855         if (c->utmp_id)
1856                 fprintf(f,
1857                         "%sUtmpIdentifier: %s\n",
1858                         prefix, c->utmp_id);
1859 }
1860
1861 void exec_status_start(ExecStatus *s, pid_t pid) {
1862         assert(s);
1863
1864         zero(*s);
1865         s->pid = pid;
1866         dual_timestamp_get(&s->start_timestamp);
1867 }
1868
1869 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1870         assert(s);
1871
1872         if (s->pid && s->pid != pid)
1873                 zero(*s);
1874
1875         s->pid = pid;
1876         dual_timestamp_get(&s->exit_timestamp);
1877
1878         s->code = code;
1879         s->status = status;
1880
1881         if (context) {
1882                 if (context->utmp_id)
1883                         utmp_put_dead_process(context->utmp_id, pid, code, status);
1884
1885                 exec_context_tty_reset(context);
1886         }
1887 }
1888
1889 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1890         char buf[FORMAT_TIMESTAMP_MAX];
1891
1892         assert(s);
1893         assert(f);
1894
1895         if (!prefix)
1896                 prefix = "";
1897
1898         if (s->pid <= 0)
1899                 return;
1900
1901         fprintf(f,
1902                 "%sPID: %lu\n",
1903                 prefix, (unsigned long) s->pid);
1904
1905         if (s->start_timestamp.realtime > 0)
1906                 fprintf(f,
1907                         "%sStart Timestamp: %s\n",
1908                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1909
1910         if (s->exit_timestamp.realtime > 0)
1911                 fprintf(f,
1912                         "%sExit Timestamp: %s\n"
1913                         "%sExit Code: %s\n"
1914                         "%sExit Status: %i\n",
1915                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1916                         prefix, sigchld_code_to_string(s->code),
1917                         prefix, s->status);
1918 }
1919
1920 char *exec_command_line(char **argv) {
1921         size_t k;
1922         char *n, *p, **a;
1923         bool first = true;
1924
1925         assert(argv);
1926
1927         k = 1;
1928         STRV_FOREACH(a, argv)
1929                 k += strlen(*a)+3;
1930
1931         if (!(n = new(char, k)))
1932                 return NULL;
1933
1934         p = n;
1935         STRV_FOREACH(a, argv) {
1936
1937                 if (!first)
1938                         *(p++) = ' ';
1939                 else
1940                         first = false;
1941
1942                 if (strpbrk(*a, WHITESPACE)) {
1943                         *(p++) = '\'';
1944                         p = stpcpy(p, *a);
1945                         *(p++) = '\'';
1946                 } else
1947                         p = stpcpy(p, *a);
1948
1949         }
1950
1951         *p = 0;
1952
1953         /* FIXME: this doesn't really handle arguments that have
1954          * spaces and ticks in them */
1955
1956         return n;
1957 }
1958
1959 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1960         char *p2;
1961         const char *prefix2;
1962
1963         char *cmd;
1964
1965         assert(c);
1966         assert(f);
1967
1968         if (!prefix)
1969                 prefix = "";
1970         p2 = strappend(prefix, "\t");
1971         prefix2 = p2 ? p2 : prefix;
1972
1973         cmd = exec_command_line(c->argv);
1974
1975         fprintf(f,
1976                 "%sCommand Line: %s\n",
1977                 prefix, cmd ? cmd : strerror(ENOMEM));
1978
1979         free(cmd);
1980
1981         exec_status_dump(&c->exec_status, f, prefix2);
1982
1983         free(p2);
1984 }
1985
1986 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1987         assert(f);
1988
1989         if (!prefix)
1990                 prefix = "";
1991
1992         LIST_FOREACH(command, c, c)
1993                 exec_command_dump(c, f, prefix);
1994 }
1995
1996 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1997         ExecCommand *end;
1998
1999         assert(l);
2000         assert(e);
2001
2002         if (*l) {
2003                 /* It's kind of important, that we keep the order here */
2004                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
2005                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
2006         } else
2007               *l = e;
2008 }
2009
2010 int exec_command_set(ExecCommand *c, const char *path, ...) {
2011         va_list ap;
2012         char **l, *p;
2013
2014         assert(c);
2015         assert(path);
2016
2017         va_start(ap, path);
2018         l = strv_new_ap(path, ap);
2019         va_end(ap);
2020
2021         if (!l)
2022                 return -ENOMEM;
2023
2024         if (!(p = strdup(path))) {
2025                 strv_free(l);
2026                 return -ENOMEM;
2027         }
2028
2029         free(c->path);
2030         c->path = p;
2031
2032         strv_free(c->argv);
2033         c->argv = l;
2034
2035         return 0;
2036 }
2037
2038 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2039         [EXEC_INPUT_NULL] = "null",
2040         [EXEC_INPUT_TTY] = "tty",
2041         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2042         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2043         [EXEC_INPUT_SOCKET] = "socket"
2044 };
2045
2046 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2047
2048 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2049         [EXEC_OUTPUT_INHERIT] = "inherit",
2050         [EXEC_OUTPUT_NULL] = "null",
2051         [EXEC_OUTPUT_TTY] = "tty",
2052         [EXEC_OUTPUT_SYSLOG] = "syslog",
2053         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2054         [EXEC_OUTPUT_KMSG] = "kmsg",
2055         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2056         [EXEC_OUTPUT_JOURNAL] = "journal",
2057         [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2058         [EXEC_OUTPUT_SOCKET] = "socket"
2059 };
2060
2061 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2062
2063 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2064         [KILL_CONTROL_GROUP] = "control-group",
2065         [KILL_PROCESS] = "process",
2066         [KILL_NONE] = "none"
2067 };
2068
2069 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2070
2071 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2072         [KILL_MAIN] = "main",
2073         [KILL_CONTROL] = "control",
2074         [KILL_ALL] = "all"
2075 };
2076
2077 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);