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"), to look pretty in /bin/ps */
 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 static void rename_process_from_path(const char *path) {
 923         char process_name[11];
 924         const char *p;
 925         size_t l;
 926
 927         /* This resulting string must fit in 10 chars (i.e. the length
 928          * of "/sbin/init") to look pretty in /bin/ps */
 929
 930         p = file_name_from_path(path);
 931         if (isempty(p)) {
 932                 rename_process("(...)");
 933                 return;
 934         }
 935
 936         l = strlen(p);
 937         if (l > 8) {
 938                 /* The end of the process name is usually more
 939                  * interesting, since the first bit might just be
 940                  * "systemd-" */
 941                 p = p + l - 8;
 942                 l = 8;
 943         }
 944
 945         process_name[0] = '(';
 946         memcpy(process_name+1, p, l);
 947         process_name[1+l] = ')';
 948         process_name[1+l+1] = 0;
 949
 950         rename_process(process_name);
 951 }
 952
 953 int exec_spawn(ExecCommand *command,
 954                char **argv,
 955                const ExecContext *context,
 956                int fds[], unsigned n_fds,
 957                char **environment,
 958                bool apply_permissions,
 959                bool apply_chroot,
 960                bool apply_tty_stdin,
 961                bool confirm_spawn,
 962                CGroupBonding *cgroup_bondings,
 963                CGroupAttribute *cgroup_attributes,
 964                pid_t *ret) {
 965
 966         pid_t pid;
 967         int r;
 968         char *line;
 969         int socket_fd;
 970         char **files_env = NULL;
 971
 972         assert(command);
 973         assert(context);
 974         assert(ret);
 975         assert(fds || n_fds <= 0);
 976
 977         if (context->std_input == EXEC_INPUT_SOCKET ||
 978             context->std_output == EXEC_OUTPUT_SOCKET ||
 979             context->std_error == EXEC_OUTPUT_SOCKET) {
 980
 981                 if (n_fds != 1)
 982                         return -EINVAL;
 983
 984                 socket_fd = fds[0];
 985
 986                 fds = NULL;
 987                 n_fds = 0;
 988         } else
 989                 socket_fd = -1;
 990
 991         if ((r = exec_context_load_environment(context, &files_env)) < 0) {
 992                 log_error("Failed to load environment files: %s", strerror(-r));
 993                 return r;
 994         }
 995
 996         if (!argv)
 997                 argv = command->argv;
 998
 999         if (!(line = exec_command_line(argv))) {
1000                 r = -ENOMEM;
1001                 goto fail_parent;
1002         }
1003
1004         log_debug("About to execute: %s", line);
1005         free(line);
1006
1007         r = cgroup_bonding_realize_list(cgroup_bondings);
1008         if (r < 0)
1009                 goto fail_parent;
1010
1011         cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
1012
1013         if ((pid = fork()) < 0) {
1014                 r = -errno;
1015                 goto fail_parent;
1016         }
1017
1018         if (pid == 0) {
1019                 int i, err;
1020                 sigset_t ss;
1021                 const char *username = NULL, *home = NULL;
1022                 uid_t uid = (uid_t) -1;
1023                 gid_t gid = (gid_t) -1;
1024                 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1025                 unsigned n_env = 0;
1026                 int saved_stdout = -1, saved_stdin = -1;
1027                 bool keep_stdout = false, keep_stdin = false, set_access = false;
1028
1029                 /* child */
1030
1031                 rename_process_from_path(command->path);
1032
1033                 /* We reset exactly these signals, since they are the
1034                  * only ones we set to SIG_IGN in the main daemon. All
1035                  * others we leave untouched because we set them to
1036                  * SIG_DFL or a valid handler initially, both of which
1037                  * will be demoted to SIG_DFL. */
1038                 default_signals(SIGNALS_CRASH_HANDLER,
1039                                 SIGNALS_IGNORE, -1);
1040
1041                 if (context->ignore_sigpipe)
1042                         ignore_signals(SIGPIPE, -1);
1043
1044                 assert_se(sigemptyset(&ss) == 0);
1045                 if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1046                         err = -errno;
1047                         r = EXIT_SIGNAL_MASK;
1048                         goto fail_child;
1049                 }
1050
1051                 /* Close sockets very early to make sure we don't
1052                  * block init reexecution because it cannot bind its
1053                  * sockets */
1054                 log_forget_fds();
1055                 err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1056                                            socket_fd >= 0 ? 1 : n_fds);
1057                 if (err < 0) {
1058                         r = EXIT_FDS;
1059                         goto fail_child;
1060                 }
1061
1062                 if (!context->same_pgrp)
1063                         if (setsid() < 0) {
1064                                 err = -errno;
1065                                 r = EXIT_SETSID;
1066                                 goto fail_child;
1067                         }
1068
1069                 if (context->tcpwrap_name) {
1070                         if (socket_fd >= 0)
1071                                 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1072                                         err = -EACCES;
1073                                         r = EXIT_TCPWRAP;
1074                                         goto fail_child;
1075                                 }
1076
1077                         for (i = 0; i < (int) n_fds; i++) {
1078                                 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1079                                         err = -EACCES;
1080                                         r = EXIT_TCPWRAP;
1081                                         goto fail_child;
1082                                 }
1083                         }
1084                 }
1085
1086                 exec_context_tty_reset(context);
1087
1088                 /* We skip the confirmation step if we shall not apply the TTY */
1089                 if (confirm_spawn &&
1090                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1091                         char response;
1092
1093                         /* Set up terminal for the question */
1094                         if ((r = setup_confirm_stdio(context,
1095                                                      &saved_stdin, &saved_stdout))) {
1096                                 err = -errno;
1097                                 goto fail_child;
1098                         }
1099
1100                         /* Now ask the question. */
1101                         if (!(line = exec_command_line(argv))) {
1102                                 err = -ENOMEM;
1103                                 r = EXIT_MEMORY;
1104                                 goto fail_child;
1105                         }
1106
1107                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1108                         free(line);
1109
1110                         if (r < 0 || response == 'n') {
1111                                 err = -ECANCELED;
1112                                 r = EXIT_CONFIRM;
1113                                 goto fail_child;
1114                         } else if (response == 's') {
1115                                 err = r = 0;
1116                                 goto fail_child;
1117                         }
1118
1119                         /* Release terminal for the question */
1120                         if ((r = restore_confirm_stdio(context,
1121                                                        &saved_stdin, &saved_stdout,
1122                                                        &keep_stdin, &keep_stdout))) {
1123                                 err = -errno;
1124                                 goto fail_child;
1125                         }
1126                 }
1127
1128                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1129                  * must sure to drop O_NONBLOCK */
1130                 if (socket_fd >= 0)
1131                         fd_nonblock(socket_fd, false);
1132
1133                 if (!keep_stdin) {
1134                         err = setup_input(context, socket_fd, apply_tty_stdin);
1135                         if (err < 0) {
1136                                 r = EXIT_STDIN;
1137                                 goto fail_child;
1138                         }
1139                 }
1140
1141                 if (!keep_stdout) {
1142                         err = setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1143                         if (err < 0) {
1144                                 r = EXIT_STDOUT;
1145                                 goto fail_child;
1146                         }
1147                 }
1148
1149                 err = setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1150                 if (err < 0) {
1151                         r = EXIT_STDERR;
1152                         goto fail_child;
1153                 }
1154
1155                 if (cgroup_bondings) {
1156                         err = cgroup_bonding_install_list(cgroup_bondings, 0);
1157                         if (err < 0) {
1158                                 r = EXIT_CGROUP;
1159                                 goto fail_child;
1160                         }
1161                 }
1162
1163                 if (context->oom_score_adjust_set) {
1164                         char t[16];
1165
1166                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1167                         char_array_0(t);
1168
1169                         if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1170                                 /* Compatibility with Linux <= 2.6.35 */
1171
1172                                 int adj;
1173
1174                                 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1175                                 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1176
1177                                 snprintf(t, sizeof(t), "%i", adj);
1178                                 char_array_0(t);
1179
1180                                 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1181                                     && errno != EACCES) {
1182                                         err = -errno;
1183                                         r = EXIT_OOM_ADJUST;
1184                                         goto fail_child;
1185                                 }
1186                         }
1187                 }
1188
1189                 if (context->nice_set)
1190                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1191                                 err = -errno;
1192                                 r = EXIT_NICE;
1193                                 goto fail_child;
1194                         }
1195
1196                 if (context->cpu_sched_set) {
1197                         struct sched_param param;
1198
1199                         zero(param);
1200                         param.sched_priority = context->cpu_sched_priority;
1201
1202                         if (sched_setscheduler(0, context->cpu_sched_policy |
1203                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
1204                                 err = -errno;
1205                                 r = EXIT_SETSCHEDULER;
1206                                 goto fail_child;
1207                         }
1208                 }
1209
1210                 if (context->cpuset)
1211                         if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1212                                 err = -errno;
1213                                 r = EXIT_CPUAFFINITY;
1214                                 goto fail_child;
1215                         }
1216
1217                 if (context->ioprio_set)
1218                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1219                                 err = -errno;
1220                                 r = EXIT_IOPRIO;
1221                                 goto fail_child;
1222                         }
1223
1224                 if (context->timer_slack_nsec_set)
1225                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1226                                 err = -errno;
1227                                 r = EXIT_TIMERSLACK;
1228                                 goto fail_child;
1229                         }
1230
1231                 if (context->utmp_id)
1232                         utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1233
1234                 if (context->user) {
1235                         username = context->user;
1236                         err = get_user_creds(&username, &uid, &gid, &home);
1237                         if (err < 0) {
1238                                 r = EXIT_USER;
1239                                 goto fail_child;
1240                         }
1241
1242                         if (is_terminal_input(context->std_input)) {
1243                                 err = chown_terminal(STDIN_FILENO, uid);
1244                                 if (err < 0) {
1245                                         r = EXIT_STDIN;
1246                                         goto fail_child;
1247                                 }
1248                         }
1249
1250                         if (cgroup_bondings && context->control_group_modify) {
1251                                 err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
1252                                 if (err >= 0)
1253                                         err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid, context->control_group_persistent);
1254                                 if (err < 0) {
1255                                         r = EXIT_CGROUP;
1256                                         goto fail_child;
1257                                 }
1258
1259                                 set_access = true;
1260                         }
1261                 }
1262
1263                 if (cgroup_bondings && !set_access && context->control_group_persistent >= 0)  {
1264                         err = cgroup_bonding_set_task_access_list(cgroup_bondings, (mode_t) -1, (uid_t) -1, (uid_t) -1, context->control_group_persistent);
1265                         if (err < 0) {
1266                                 r = EXIT_CGROUP;
1267                                 goto fail_child;
1268                         }
1269                 }
1270
1271                 if (apply_permissions) {
1272                         err = enforce_groups(context, username, gid);
1273                         if (err < 0) {
1274                                 r = EXIT_GROUP;
1275                                 goto fail_child;
1276                         }
1277                 }
1278
1279                 umask(context->umask);
1280
1281 #ifdef HAVE_PAM
1282                 if (context->pam_name && username) {
1283                         err = setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds);
1284                         if (err < 0) {
1285                                 r = EXIT_PAM;
1286                                 goto fail_child;
1287                         }
1288                 }
1289 #endif
1290                 if (context->private_network) {
1291                         if (unshare(CLONE_NEWNET) < 0) {
1292                                 err = -errno;
1293                                 r = EXIT_NETWORK;
1294                                 goto fail_child;
1295                         }
1296
1297                         loopback_setup();
1298                 }
1299
1300                 if (strv_length(context->read_write_dirs) > 0 ||
1301                     strv_length(context->read_only_dirs) > 0 ||
1302                     strv_length(context->inaccessible_dirs) > 0 ||
1303                     context->mount_flags != MS_SHARED ||
1304                     context->private_tmp) {
1305                         err = setup_namespace(context->read_write_dirs,
1306                                               context->read_only_dirs,
1307                                               context->inaccessible_dirs,
1308                                               context->private_tmp,
1309                                               context->mount_flags);
1310                         if (err < 0) {
1311                                 r = EXIT_NAMESPACE;
1312                                 goto fail_child;
1313                         }
1314                 }
1315
1316                 if (apply_chroot) {
1317                         if (context->root_directory)
1318                                 if (chroot(context->root_directory) < 0) {
1319                                         err = -errno;
1320                                         r = EXIT_CHROOT;
1321                                         goto fail_child;
1322                                 }
1323
1324                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1325                                 err = -errno;
1326                                 r = EXIT_CHDIR;
1327                                 goto fail_child;
1328                         }
1329                 } else {
1330
1331                         char *d;
1332
1333                         if (asprintf(&d, "%s/%s",
1334                                      context->root_directory ? context->root_directory : "",
1335                                      context->working_directory ? context->working_directory : "") < 0) {
1336                                 err = -ENOMEM;
1337                                 r = EXIT_MEMORY;
1338                                 goto fail_child;
1339                         }
1340
1341                         if (chdir(d) < 0) {
1342                                 err = -errno;
1343                                 free(d);
1344                                 r = EXIT_CHDIR;
1345                                 goto fail_child;
1346                         }
1347
1348                         free(d);
1349                 }
1350
1351                 /* We repeat the fd closing here, to make sure that
1352                  * nothing is leaked from the PAM modules */
1353                 err = close_all_fds(fds, n_fds);
1354                 if (err >= 0)
1355                         err = shift_fds(fds, n_fds);
1356                 if (err >= 0)
1357                         err = flags_fds(fds, n_fds, context->non_blocking);
1358                 if (err < 0) {
1359                         r = EXIT_FDS;
1360                         goto fail_child;
1361                 }
1362
1363                 if (apply_permissions) {
1364
1365                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1366                                 if (!context->rlimit[i])
1367                                         continue;
1368
1369                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1370                                         err = -errno;
1371                                         r = EXIT_LIMITS;
1372                                         goto fail_child;
1373                                 }
1374                         }
1375
1376                         if (context->capability_bounding_set_drop) {
1377                                 err = do_capability_bounding_set_drop(context->capability_bounding_set_drop);
1378                                 if (err < 0) {
1379                                         r = EXIT_CAPABILITIES;
1380                                         goto fail_child;
1381                                 }
1382                         }
1383
1384                         if (context->user) {
1385                                 err = enforce_user(context, uid);
1386                                 if (err < 0) {
1387                                         r = EXIT_USER;
1388                                         goto fail_child;
1389                                 }
1390                         }
1391
1392                         /* PR_GET_SECUREBITS is not privileged, while
1393                          * PR_SET_SECUREBITS is. So to suppress
1394                          * potential EPERMs we'll try not to call
1395                          * PR_SET_SECUREBITS unless necessary. */
1396                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1397                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1398                                         err = -errno;
1399                                         r = EXIT_SECUREBITS;
1400                                         goto fail_child;
1401                                 }
1402
1403                         if (context->capabilities)
1404                                 if (cap_set_proc(context->capabilities) < 0) {
1405                                         err = -errno;
1406                                         r = EXIT_CAPABILITIES;
1407                                         goto fail_child;
1408                                 }
1409                 }
1410
1411                 if (!(our_env = new0(char*, 7))) {
1412                         err = -ENOMEM;
1413                         r = EXIT_MEMORY;
1414                         goto fail_child;
1415                 }
1416
1417                 if (n_fds > 0)
1418                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1419                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1420                                 err = -ENOMEM;
1421                                 r = EXIT_MEMORY;
1422                                 goto fail_child;
1423                         }
1424
1425                 if (home)
1426                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1427                                 err = -ENOMEM;
1428                                 r = EXIT_MEMORY;
1429                                 goto fail_child;
1430                         }
1431
1432                 if (username)
1433                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1434                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1435                                 err = -ENOMEM;
1436                                 r = EXIT_MEMORY;
1437                                 goto fail_child;
1438                         }
1439
1440                 if (is_terminal_input(context->std_input) ||
1441                     context->std_output == EXEC_OUTPUT_TTY ||
1442                     context->std_error == EXEC_OUTPUT_TTY)
1443                         if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1444                                 err = -ENOMEM;
1445                                 r = EXIT_MEMORY;
1446                                 goto fail_child;
1447                         }
1448
1449                 assert(n_env <= 7);
1450
1451                 if (!(final_env = strv_env_merge(
1452                                       5,
1453                                       environment,
1454                                       our_env,
1455                                       context->environment,
1456                                       files_env,
1457                                       pam_env,
1458                                       NULL))) {
1459                         err = -ENOMEM;
1460                         r = EXIT_MEMORY;
1461                         goto fail_child;
1462                 }
1463
1464                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1465                         err = -ENOMEM;
1466                         r = EXIT_MEMORY;
1467                         goto fail_child;
1468                 }
1469
1470                 final_env = strv_env_clean(final_env);
1471
1472                 execve(command->path, final_argv, final_env);
1473                 err = -errno;
1474                 r = EXIT_EXEC;
1475
1476         fail_child:
1477                 if (r != 0) {
1478                         log_open();
1479                         log_warning("Failed at step %s spawning %s: %s",
1480                                     exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1481                                     command->path, strerror(-err));
1482                 }
1483
1484                 strv_free(our_env);
1485                 strv_free(final_env);
1486                 strv_free(pam_env);
1487                 strv_free(files_env);
1488                 strv_free(final_argv);
1489
1490                 if (saved_stdin >= 0)
1491                         close_nointr_nofail(saved_stdin);
1492
1493                 if (saved_stdout >= 0)
1494                         close_nointr_nofail(saved_stdout);
1495
1496                 _exit(r);
1497         }
1498
1499         strv_free(files_env);
1500
1501         /* We add the new process to the cgroup both in the child (so
1502          * that we can be sure that no user code is ever executed
1503          * outside of the cgroup) and in the parent (so that we can be
1504          * sure that when we kill the cgroup the process will be
1505          * killed too). */
1506         if (cgroup_bondings)
1507                 cgroup_bonding_install_list(cgroup_bondings, pid);
1508
1509         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1510
1511         exec_status_start(&command->exec_status, pid);
1512
1513         *ret = pid;
1514         return 0;
1515
1516 fail_parent:
1517         strv_free(files_env);
1518
1519         return r;
1520 }
1521
1522 void exec_context_init(ExecContext *c) {
1523         assert(c);
1524
1525         c->umask = 0022;
1526         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1527         c->cpu_sched_policy = SCHED_OTHER;
1528         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1529         c->syslog_level_prefix = true;
1530         c->mount_flags = MS_SHARED;
1531         c->kill_signal = SIGTERM;
1532         c->send_sigkill = true;
1533         c->control_group_persistent = -1;
1534         c->ignore_sigpipe = true;
1535 }
1536
1537 void exec_context_done(ExecContext *c) {
1538         unsigned l;
1539
1540         assert(c);
1541
1542         strv_free(c->environment);
1543         c->environment = NULL;
1544
1545         strv_free(c->environment_files);
1546         c->environment_files = NULL;
1547
1548         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1549                 free(c->rlimit[l]);
1550                 c->rlimit[l] = NULL;
1551         }
1552
1553         free(c->working_directory);
1554         c->working_directory = NULL;
1555         free(c->root_directory);
1556         c->root_directory = NULL;
1557
1558         free(c->tty_path);
1559         c->tty_path = NULL;
1560
1561         free(c->tcpwrap_name);
1562         c->tcpwrap_name = NULL;
1563
1564         free(c->syslog_identifier);
1565         c->syslog_identifier = NULL;
1566
1567         free(c->user);
1568         c->user = NULL;
1569
1570         free(c->group);
1571         c->group = NULL;
1572
1573         strv_free(c->supplementary_groups);
1574         c->supplementary_groups = NULL;
1575
1576         free(c->pam_name);
1577         c->pam_name = NULL;
1578
1579         if (c->capabilities) {
1580                 cap_free(c->capabilities);
1581                 c->capabilities = NULL;
1582         }
1583
1584         strv_free(c->read_only_dirs);
1585         c->read_only_dirs = NULL;
1586
1587         strv_free(c->read_write_dirs);
1588         c->read_write_dirs = NULL;
1589
1590         strv_free(c->inaccessible_dirs);
1591         c->inaccessible_dirs = NULL;
1592
1593         if (c->cpuset)
1594                 CPU_FREE(c->cpuset);
1595
1596         free(c->utmp_id);
1597         c->utmp_id = NULL;
1598 }
1599
1600 void exec_command_done(ExecCommand *c) {
1601         assert(c);
1602
1603         free(c->path);
1604         c->path = NULL;
1605
1606         strv_free(c->argv);
1607         c->argv = NULL;
1608 }
1609
1610 void exec_command_done_array(ExecCommand *c, unsigned n) {
1611         unsigned i;
1612
1613         for (i = 0; i < n; i++)
1614                 exec_command_done(c+i);
1615 }
1616
1617 void exec_command_free_list(ExecCommand *c) {
1618         ExecCommand *i;
1619
1620         while ((i = c)) {
1621                 LIST_REMOVE(ExecCommand, command, c, i);
1622                 exec_command_done(i);
1623                 free(i);
1624         }
1625 }
1626
1627 void exec_command_free_array(ExecCommand **c, unsigned n) {
1628         unsigned i;
1629
1630         for (i = 0; i < n; i++) {
1631                 exec_command_free_list(c[i]);
1632                 c[i] = NULL;
1633         }
1634 }
1635
1636 int exec_context_load_environment(const ExecContext *c, char ***l) {
1637         char **i, **r = NULL;
1638
1639         assert(c);
1640         assert(l);
1641
1642         STRV_FOREACH(i, c->environment_files) {
1643                 char *fn;
1644                 int k;
1645                 bool ignore = false;
1646                 char **p;
1647
1648                 fn = *i;
1649
1650                 if (fn[0] == '-') {
1651                         ignore = true;
1652                         fn ++;
1653                 }
1654
1655                 if (!path_is_absolute(fn)) {
1656
1657                         if (ignore)
1658                                 continue;
1659
1660                         strv_free(r);
1661                         return -EINVAL;
1662                 }
1663
1664                 if ((k = load_env_file(fn, &p)) < 0) {
1665
1666                         if (ignore)
1667                                 continue;
1668
1669                         strv_free(r);
1670                         return k;
1671                 }
1672
1673                 if (r == NULL)
1674                         r = p;
1675                 else {
1676                         char **m;
1677
1678                         m = strv_env_merge(2, r, p);
1679                         strv_free(r);
1680                         strv_free(p);
1681
1682                         if (!m)
1683                                 return -ENOMEM;
1684
1685                         r = m;
1686                 }
1687         }
1688
1689         *l = r;
1690
1691         return 0;
1692 }
1693
1694 static void strv_fprintf(FILE *f, char **l) {
1695         char **g;
1696
1697         assert(f);
1698
1699         STRV_FOREACH(g, l)
1700                 fprintf(f, " %s", *g);
1701 }
1702
1703 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1704         char ** e;
1705         unsigned i;
1706
1707         assert(c);
1708         assert(f);
1709
1710         if (!prefix)
1711                 prefix = "";
1712
1713         fprintf(f,
1714                 "%sUMask: %04o\n"
1715                 "%sWorkingDirectory: %s\n"
1716                 "%sRootDirectory: %s\n"
1717                 "%sNonBlocking: %s\n"
1718                 "%sPrivateTmp: %s\n"
1719                 "%sControlGroupModify: %s\n"
1720                 "%sControlGroupPersistent: %s\n"
1721                 "%sPrivateNetwork: %s\n",
1722                 prefix, c->umask,
1723                 prefix, c->working_directory ? c->working_directory : "/",
1724                 prefix, c->root_directory ? c->root_directory : "/",
1725                 prefix, yes_no(c->non_blocking),
1726                 prefix, yes_no(c->private_tmp),
1727                 prefix, yes_no(c->control_group_modify),
1728                 prefix, yes_no(c->control_group_persistent),
1729                 prefix, yes_no(c->private_network));
1730
1731         STRV_FOREACH(e, c->environment)
1732                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1733
1734         STRV_FOREACH(e, c->environment_files)
1735                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1736
1737         if (c->tcpwrap_name)
1738                 fprintf(f,
1739                         "%sTCPWrapName: %s\n",
1740                         prefix, c->tcpwrap_name);
1741
1742         if (c->nice_set)
1743                 fprintf(f,
1744                         "%sNice: %i\n",
1745                         prefix, c->nice);
1746
1747         if (c->oom_score_adjust_set)
1748                 fprintf(f,
1749                         "%sOOMScoreAdjust: %i\n",
1750                         prefix, c->oom_score_adjust);
1751
1752         for (i = 0; i < RLIM_NLIMITS; i++)
1753                 if (c->rlimit[i])
1754                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1755
1756         if (c->ioprio_set)
1757                 fprintf(f,
1758                         "%sIOSchedulingClass: %s\n"
1759                         "%sIOPriority: %i\n",
1760                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1761                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1762
1763         if (c->cpu_sched_set)
1764                 fprintf(f,
1765                         "%sCPUSchedulingPolicy: %s\n"
1766                         "%sCPUSchedulingPriority: %i\n"
1767                         "%sCPUSchedulingResetOnFork: %s\n",
1768                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1769                         prefix, c->cpu_sched_priority,
1770                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1771
1772         if (c->cpuset) {
1773                 fprintf(f, "%sCPUAffinity:", prefix);
1774                 for (i = 0; i < c->cpuset_ncpus; i++)
1775                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1776                                 fprintf(f, " %i", i);
1777                 fputs("\n", f);
1778         }
1779
1780         if (c->timer_slack_nsec_set)
1781                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1782
1783         fprintf(f,
1784                 "%sStandardInput: %s\n"
1785                 "%sStandardOutput: %s\n"
1786                 "%sStandardError: %s\n",
1787                 prefix, exec_input_to_string(c->std_input),
1788                 prefix, exec_output_to_string(c->std_output),
1789                 prefix, exec_output_to_string(c->std_error));
1790
1791         if (c->tty_path)
1792                 fprintf(f,
1793                         "%sTTYPath: %s\n"
1794                         "%sTTYReset: %s\n"
1795                         "%sTTYVHangup: %s\n"
1796                         "%sTTYVTDisallocate: %s\n",
1797                         prefix, c->tty_path,
1798                         prefix, yes_no(c->tty_reset),
1799                         prefix, yes_no(c->tty_vhangup),
1800                         prefix, yes_no(c->tty_vt_disallocate));
1801
1802         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL ||
1803             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
1804             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL ||
1805             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE)
1806                 fprintf(f,
1807                         "%sSyslogFacility: %s\n"
1808                         "%sSyslogLevel: %s\n",
1809                         prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1810                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1811
1812         if (c->capabilities) {
1813                 char *t;
1814                 if ((t = cap_to_text(c->capabilities, NULL))) {
1815                         fprintf(f, "%sCapabilities: %s\n",
1816                                 prefix, t);
1817                         cap_free(t);
1818                 }
1819         }
1820
1821         if (c->secure_bits)
1822                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1823                         prefix,
1824                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1825                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1826                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1827                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1828                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1829                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1830
1831         if (c->capability_bounding_set_drop) {
1832                 unsigned long l;
1833                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1834
1835                 for (l = 0; l <= cap_last_cap(); l++)
1836                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1837                                 char *t;
1838
1839                                 if ((t = cap_to_name(l))) {
1840                                         fprintf(f, " %s", t);
1841                                         cap_free(t);
1842                                 }
1843                         }
1844
1845                 fputs("\n", f);
1846         }
1847
1848         if (c->user)
1849                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1850         if (c->group)
1851                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1852
1853         if (strv_length(c->supplementary_groups) > 0) {
1854                 fprintf(f, "%sSupplementaryGroups:", prefix);
1855                 strv_fprintf(f, c->supplementary_groups);
1856                 fputs("\n", f);
1857         }
1858
1859         if (c->pam_name)
1860                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1861
1862         if (strv_length(c->read_write_dirs) > 0) {
1863                 fprintf(f, "%sReadWriteDirs:", prefix);
1864                 strv_fprintf(f, c->read_write_dirs);
1865                 fputs("\n", f);
1866         }
1867
1868         if (strv_length(c->read_only_dirs) > 0) {
1869                 fprintf(f, "%sReadOnlyDirs:", prefix);
1870                 strv_fprintf(f, c->read_only_dirs);
1871                 fputs("\n", f);
1872         }
1873
1874         if (strv_length(c->inaccessible_dirs) > 0) {
1875                 fprintf(f, "%sInaccessibleDirs:", prefix);
1876                 strv_fprintf(f, c->inaccessible_dirs);
1877                 fputs("\n", f);
1878         }
1879
1880         fprintf(f,
1881                 "%sKillMode: %s\n"
1882                 "%sKillSignal: SIG%s\n"
1883                 "%sSendSIGKILL: %s\n"
1884                 "%sIgnoreSIGPIPE: %s\n",
1885                 prefix, kill_mode_to_string(c->kill_mode),
1886                 prefix, signal_to_string(c->kill_signal),
1887                 prefix, yes_no(c->send_sigkill),
1888                 prefix, yes_no(c->ignore_sigpipe));
1889
1890         if (c->utmp_id)
1891                 fprintf(f,
1892                         "%sUtmpIdentifier: %s\n",
1893                         prefix, c->utmp_id);
1894 }
1895
1896 void exec_status_start(ExecStatus *s, pid_t pid) {
1897         assert(s);
1898
1899         zero(*s);
1900         s->pid = pid;
1901         dual_timestamp_get(&s->start_timestamp);
1902 }
1903
1904 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1905         assert(s);
1906
1907         if (s->pid && s->pid != pid)
1908                 zero(*s);
1909
1910         s->pid = pid;
1911         dual_timestamp_get(&s->exit_timestamp);
1912
1913         s->code = code;
1914         s->status = status;
1915
1916         if (context) {
1917                 if (context->utmp_id)
1918                         utmp_put_dead_process(context->utmp_id, pid, code, status);
1919
1920                 exec_context_tty_reset(context);
1921         }
1922 }
1923
1924 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1925         char buf[FORMAT_TIMESTAMP_MAX];
1926
1927         assert(s);
1928         assert(f);
1929
1930         if (!prefix)
1931                 prefix = "";
1932
1933         if (s->pid <= 0)
1934                 return;
1935
1936         fprintf(f,
1937                 "%sPID: %lu\n",
1938                 prefix, (unsigned long) s->pid);
1939
1940         if (s->start_timestamp.realtime > 0)
1941                 fprintf(f,
1942                         "%sStart Timestamp: %s\n",
1943                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1944
1945         if (s->exit_timestamp.realtime > 0)
1946                 fprintf(f,
1947                         "%sExit Timestamp: %s\n"
1948                         "%sExit Code: %s\n"
1949                         "%sExit Status: %i\n",
1950                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1951                         prefix, sigchld_code_to_string(s->code),
1952                         prefix, s->status);
1953 }
1954
1955 char *exec_command_line(char **argv) {
1956         size_t k;
1957         char *n, *p, **a;
1958         bool first = true;
1959
1960         assert(argv);
1961
1962         k = 1;
1963         STRV_FOREACH(a, argv)
1964                 k += strlen(*a)+3;
1965
1966         if (!(n = new(char, k)))
1967                 return NULL;
1968
1969         p = n;
1970         STRV_FOREACH(a, argv) {
1971
1972                 if (!first)
1973                         *(p++) = ' ';
1974                 else
1975                         first = false;
1976
1977                 if (strpbrk(*a, WHITESPACE)) {
1978                         *(p++) = '\'';
1979                         p = stpcpy(p, *a);
1980                         *(p++) = '\'';
1981                 } else
1982                         p = stpcpy(p, *a);
1983
1984         }
1985
1986         *p = 0;
1987
1988         /* FIXME: this doesn't really handle arguments that have
1989          * spaces and ticks in them */
1990
1991         return n;
1992 }
1993
1994 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1995         char *p2;
1996         const char *prefix2;
1997
1998         char *cmd;
1999
2000         assert(c);
2001         assert(f);
2002
2003         if (!prefix)
2004                 prefix = "";
2005         p2 = strappend(prefix, "\t");
2006         prefix2 = p2 ? p2 : prefix;
2007
2008         cmd = exec_command_line(c->argv);
2009
2010         fprintf(f,
2011                 "%sCommand Line: %s\n",
2012                 prefix, cmd ? cmd : strerror(ENOMEM));
2013
2014         free(cmd);
2015
2016         exec_status_dump(&c->exec_status, f, prefix2);
2017
2018         free(p2);
2019 }
2020
2021 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2022         assert(f);
2023
2024         if (!prefix)
2025                 prefix = "";
2026
2027         LIST_FOREACH(command, c, c)
2028                 exec_command_dump(c, f, prefix);
2029 }
2030
2031 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2032         ExecCommand *end;
2033
2034         assert(l);
2035         assert(e);
2036
2037         if (*l) {
2038                 /* It's kind of important, that we keep the order here */
2039                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
2040                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
2041         } else
2042               *l = e;
2043 }
2044
2045 int exec_command_set(ExecCommand *c, const char *path, ...) {
2046         va_list ap;
2047         char **l, *p;
2048
2049         assert(c);
2050         assert(path);
2051
2052         va_start(ap, path);
2053         l = strv_new_ap(path, ap);
2054         va_end(ap);
2055
2056         if (!l)
2057                 return -ENOMEM;
2058
2059         if (!(p = strdup(path))) {
2060                 strv_free(l);
2061                 return -ENOMEM;
2062         }
2063
2064         free(c->path);
2065         c->path = p;
2066
2067         strv_free(c->argv);
2068         c->argv = l;
2069
2070         return 0;
2071 }
2072
2073 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2074         [EXEC_INPUT_NULL] = "null",
2075         [EXEC_INPUT_TTY] = "tty",
2076         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2077         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2078         [EXEC_INPUT_SOCKET] = "socket"
2079 };
2080
2081 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2082
2083 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2084         [EXEC_OUTPUT_INHERIT] = "inherit",
2085         [EXEC_OUTPUT_NULL] = "null",
2086         [EXEC_OUTPUT_TTY] = "tty",
2087         [EXEC_OUTPUT_SYSLOG] = "syslog",
2088         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2089         [EXEC_OUTPUT_KMSG] = "kmsg",
2090         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2091         [EXEC_OUTPUT_JOURNAL] = "journal",
2092         [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2093         [EXEC_OUTPUT_SOCKET] = "socket"
2094 };
2095
2096 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2097
2098 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2099         [KILL_CONTROL_GROUP] = "control-group",
2100         [KILL_PROCESS] = "process",
2101         [KILL_NONE] = "none"
2102 };
2103
2104 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2105
2106 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2107         [KILL_MAIN] = "main",
2108         [KILL_CONTROL] = "control",
2109         [KILL_ALL] = "all"
2110 };
2111
2112 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);