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