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, STDOUT_SYSLOG_BRIDGE_SOCKET, sizeof(sa.un.sun_path));
 192
 193         if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + sizeof(STDOUT_SYSLOG_BRIDGE_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         int err;
 720         char **e = NULL;
 721         bool close_session = false;
 722         pid_t pam_pid = 0, parent_pid;
 723
 724         assert(name);
 725         assert(user);
 726         assert(pam_env);
 727
 728         /* We set up PAM in the parent process, then fork. The child
 729          * will then stay around until killed via PR_GET_PDEATHSIG or
 730          * systemd via the cgroup logic. It will then remove the PAM
 731          * session again. The parent process will exec() the actual
 732          * daemon. We do things this way to ensure that the main PID
 733          * of the daemon is the one we initially fork()ed. */
 734
 735         if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
 736                 handle = NULL;
 737                 goto fail;
 738         }
 739
 740         if (tty)
 741                 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
 742                         goto fail;
 743
 744         if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
 745                 goto fail;
 746
 747         if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
 748                 goto fail;
 749
 750         close_session = true;
 751
 752         if ((!(e = pam_getenvlist(handle)))) {
 753                 pam_code = PAM_BUF_ERR;
 754                 goto fail;
 755         }
 756
 757         /* Block SIGTERM, so that we know that it won't get lost in
 758          * the child */
 759         if (sigemptyset(&ss) < 0 ||
 760             sigaddset(&ss, SIGTERM) < 0 ||
 761             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 762                 goto fail;
 763
 764         parent_pid = getpid();
 765
 766         if ((pam_pid = fork()) < 0)
 767                 goto fail;
 768
 769         if (pam_pid == 0) {
 770                 int sig;
 771                 int r = EXIT_PAM;
 772
 773                 /* The child's job is to reset the PAM session on
 774                  * termination */
 775
 776                 /* This string must fit in 10 chars (i.e. the length
 777                  * of "/sbin/init") */
 778                 rename_process("sd(PAM)");
 779
 780                 /* Make sure we don't keep open the passed fds in this
 781                 child. We assume that otherwise only those fds are
 782                 open here that have been opened by PAM. */
 783                 close_many(fds, n_fds);
 784
 785                 /* Wait until our parent died. This will most likely
 786                  * not work since the kernel does not allow
 787                  * unprivileged parents kill their privileged children
 788                  * this way. We rely on the control groups kill logic
 789                  * to do the rest for us. */
 790                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 791                         goto child_finish;
 792
 793                 /* Check if our parent process might already have
 794                  * died? */
 795                 if (getppid() == parent_pid) {
 796                         for (;;) {
 797                                 if (sigwait(&ss, &sig) < 0) {
 798                                         if (errno == EINTR)
 799                                                 continue;
 800
 801                                         goto child_finish;
 802                                 }
 803
 804                                 assert(sig == SIGTERM);
 805                                 break;
 806                         }
 807                 }
 808
 809                 /* If our parent died we'll end the session */
 810                 if (getppid() != parent_pid)
 811                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
 812                                 goto child_finish;
 813
 814                 r = 0;
 815
 816         child_finish:
 817                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 818                 _exit(r);
 819         }
 820
 821         /* If the child was forked off successfully it will do all the
 822          * cleanups, so forget about the handle here. */
 823         handle = NULL;
 824
 825         /* Unblock SIGTERM again in the parent */
 826         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 827                 goto fail;
 828
 829         /* We close the log explicitly here, since the PAM modules
 830          * might have opened it, but we don't want this fd around. */
 831         closelog();
 832
 833         *pam_env = e;
 834         e = NULL;
 835
 836         return 0;
 837
 838 fail:
 839         if (pam_code != PAM_SUCCESS)
 840                 err = -EPERM;  /* PAM errors do not map to errno */
 841         else
 842                 err = -errno;
 843
 844         if (handle) {
 845                 if (close_session)
 846                         pam_code = pam_close_session(handle, PAM_DATA_SILENT);
 847
 848                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 849         }
 850
 851         strv_free(e);
 852
 853         closelog();
 854
 855         if (pam_pid > 1) {
 856                 kill(pam_pid, SIGTERM);
 857                 kill(pam_pid, SIGCONT);
 858         }
 859
 860         return err;
 861 }
 862 #endif
 863
 864 static int do_capability_bounding_set_drop(uint64_t drop) {
 865         unsigned long i;
 866         cap_t old_cap = NULL, new_cap = NULL;
 867         cap_flag_value_t fv;
 868         int r;
 869
 870         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 871          * in the effective set (yes, the kernel drops that when
 872          * executing init!), so get it back temporarily so that we can
 873          * call PR_CAPBSET_DROP. */
 874
 875         old_cap = cap_get_proc();
 876         if (!old_cap)
 877                 return -errno;
 878
 879         if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
 880                 r = -errno;
 881                 goto finish;
 882         }
 883
 884         if (fv != CAP_SET) {
 885                 static const cap_value_t v = CAP_SETPCAP;
 886
 887                 new_cap = cap_dup(old_cap);
 888                 if (!new_cap) {
 889                         r = -errno;
 890                         goto finish;
 891                 }
 892
 893                 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
 894                         r = -errno;
 895                         goto finish;
 896                 }
 897
 898                 if (cap_set_proc(new_cap) < 0) {
 899                         r = -errno;
 900                         goto finish;
 901                 }
 902         }
 903
 904         for (i = 0; i <= cap_last_cap(); i++)
 905                 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
 906                         if (prctl(PR_CAPBSET_DROP, i) < 0) {
 907                                 r = -errno;
 908                                 goto finish;
 909                         }
 910                 }
 911
 912         r = 0;
 913
 914 finish:
 915         if (new_cap)
 916                 cap_free(new_cap);
 917
 918         if (old_cap) {
 919                 cap_set_proc(old_cap);
 920                 cap_free(old_cap);
 921         }
 922
 923         return r;
 924 }
 925
 926 int exec_spawn(ExecCommand *command,
 927                char **argv,
 928                const ExecContext *context,
 929                int fds[], unsigned n_fds,
 930                char **environment,
 931                bool apply_permissions,
 932                bool apply_chroot,
 933                bool apply_tty_stdin,
 934                bool confirm_spawn,
 935                CGroupBonding *cgroup_bondings,
 936                CGroupAttribute *cgroup_attributes,
 937                pid_t *ret) {
 938
 939         pid_t pid;
 940         int r;
 941         char *line;
 942         int socket_fd;
 943         char **files_env = NULL;
 944
 945         assert(command);
 946         assert(context);
 947         assert(ret);
 948         assert(fds || n_fds <= 0);
 949
 950         if (context->std_input == EXEC_INPUT_SOCKET ||
 951             context->std_output == EXEC_OUTPUT_SOCKET ||
 952             context->std_error == EXEC_OUTPUT_SOCKET) {
 953
 954                 if (n_fds != 1)
 955                         return -EINVAL;
 956
 957                 socket_fd = fds[0];
 958
 959                 fds = NULL;
 960                 n_fds = 0;
 961         } else
 962                 socket_fd = -1;
 963
 964         if ((r = exec_context_load_environment(context, &files_env)) < 0) {
 965                 log_error("Failed to load environment files: %s", strerror(-r));
 966                 return r;
 967         }
 968
 969         if (!argv)
 970                 argv = command->argv;
 971
 972         if (!(line = exec_command_line(argv))) {
 973                 r = -ENOMEM;
 974                 goto fail_parent;
 975         }
 976
 977         log_debug("About to execute: %s", line);
 978         free(line);
 979
 980         r = cgroup_bonding_realize_list(cgroup_bondings);
 981         if (r < 0)
 982                 goto fail_parent;
 983
 984         cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
 985
 986         if ((pid = fork()) < 0) {
 987                 r = -errno;
 988                 goto fail_parent;
 989         }
 990
 991         if (pid == 0) {
 992                 int i, err;
 993                 sigset_t ss;
 994                 const char *username = NULL, *home = NULL;
 995                 uid_t uid = (uid_t) -1;
 996                 gid_t gid = (gid_t) -1;
 997                 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
 998                 unsigned n_env = 0;
 999                 int saved_stdout = -1, saved_stdin = -1;
1000                 bool keep_stdout = false, keep_stdin = false;
1001
1002                 /* child */
1003
1004                 /* This string must fit in 10 chars (i.e. the length
1005                  * of "/sbin/init") */
1006                 rename_process("sd(EXEC)");
1007
1008                 /* We reset exactly these signals, since they are the
1009                  * only ones we set to SIG_IGN in the main daemon. All
1010                  * others we leave untouched because we set them to
1011                  * SIG_DFL or a valid handler initially, both of which
1012                  * will be demoted to SIG_DFL. */
1013                 default_signals(SIGNALS_CRASH_HANDLER,
1014                                 SIGNALS_IGNORE, -1);
1015
1016                 if (sigemptyset(&ss) < 0 ||
1017                     sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1018                         err = -errno;
1019                         r = EXIT_SIGNAL_MASK;
1020                         goto fail_child;
1021                 }
1022
1023                 /* Close sockets very early to make sure we don't
1024                  * block init reexecution because it cannot bind its
1025                  * sockets */
1026                 log_forget_fds();
1027                 err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1028                                            socket_fd >= 0 ? 1 : n_fds);
1029                 if (err < 0) {
1030                         r = EXIT_FDS;
1031                         goto fail_child;
1032                 }
1033
1034                 if (!context->same_pgrp)
1035                         if (setsid() < 0) {
1036                                 err = -errno;
1037                                 r = EXIT_SETSID;
1038                                 goto fail_child;
1039                         }
1040
1041                 if (context->tcpwrap_name) {
1042                         if (socket_fd >= 0)
1043                                 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1044                                         err = -EACCES;
1045                                         r = EXIT_TCPWRAP;
1046                                         goto fail_child;
1047                                 }
1048
1049                         for (i = 0; i < (int) n_fds; i++) {
1050                                 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1051                                         err = -EACCES;
1052                                         r = EXIT_TCPWRAP;
1053                                         goto fail_child;
1054                                 }
1055                         }
1056                 }
1057
1058                 exec_context_tty_reset(context);
1059
1060                 /* We skip the confirmation step if we shall not apply the TTY */
1061                 if (confirm_spawn &&
1062                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1063                         char response;
1064
1065                         /* Set up terminal for the question */
1066                         if ((r = setup_confirm_stdio(context,
1067                                                      &saved_stdin, &saved_stdout))) {
1068                                 err = -errno;
1069                                 goto fail_child;
1070                         }
1071
1072                         /* Now ask the question. */
1073                         if (!(line = exec_command_line(argv))) {
1074                                 err = -ENOMEM;
1075                                 r = EXIT_MEMORY;
1076                                 goto fail_child;
1077                         }
1078
1079                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1080                         free(line);
1081
1082                         if (r < 0 || response == 'n') {
1083                                 err = -ECANCELED;
1084                                 r = EXIT_CONFIRM;
1085                                 goto fail_child;
1086                         } else if (response == 's') {
1087                                 err = r = 0;
1088                                 goto fail_child;
1089                         }
1090
1091                         /* Release terminal for the question */
1092                         if ((r = restore_confirm_stdio(context,
1093                                                        &saved_stdin, &saved_stdout,
1094                                                        &keep_stdin, &keep_stdout))) {
1095                                 err = -errno;
1096                                 goto fail_child;
1097                         }
1098                 }
1099
1100                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1101                  * must sure to drop O_NONBLOCK */
1102                 if (socket_fd >= 0)
1103                         fd_nonblock(socket_fd, false);
1104
1105                 if (!keep_stdin) {
1106                         err = setup_input(context, socket_fd, apply_tty_stdin);
1107                         if (err < 0) {
1108                                 r = EXIT_STDIN;
1109                                 goto fail_child;
1110                         }
1111                 }
1112
1113                 if (!keep_stdout) {
1114                         err = setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1115                         if (err < 0) {
1116                                 r = EXIT_STDOUT;
1117                                 goto fail_child;
1118                         }
1119                 }
1120
1121                 err = setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1122                 if (err < 0) {
1123                         r = EXIT_STDERR;
1124                         goto fail_child;
1125                 }
1126
1127                 if (cgroup_bondings) {
1128                         err = cgroup_bonding_install_list(cgroup_bondings, 0);
1129                         if (err < 0) {
1130                                 r = EXIT_CGROUP;
1131                                 goto fail_child;
1132                         }
1133                 }
1134
1135                 if (context->oom_score_adjust_set) {
1136                         char t[16];
1137
1138                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1139                         char_array_0(t);
1140
1141                         if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1142                                 /* Compatibility with Linux <= 2.6.35 */
1143
1144                                 int adj;
1145
1146                                 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1147                                 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1148
1149                                 snprintf(t, sizeof(t), "%i", adj);
1150                                 char_array_0(t);
1151
1152                                 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1153                                     && errno != EACCES) {
1154                                         err = -errno;
1155                                         r = EXIT_OOM_ADJUST;
1156                                         goto fail_child;
1157                                 }
1158                         }
1159                 }
1160
1161                 if (context->nice_set)
1162                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1163                                 err = -errno;
1164                                 r = EXIT_NICE;
1165                                 goto fail_child;
1166                         }
1167
1168                 if (context->cpu_sched_set) {
1169                         struct sched_param param;
1170
1171                         zero(param);
1172                         param.sched_priority = context->cpu_sched_priority;
1173
1174                         if (sched_setscheduler(0, context->cpu_sched_policy |
1175                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
1176                                 err = -errno;
1177                                 r = EXIT_SETSCHEDULER;
1178                                 goto fail_child;
1179                         }
1180                 }
1181
1182                 if (context->cpuset)
1183                         if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1184                                 err = -errno;
1185                                 r = EXIT_CPUAFFINITY;
1186                                 goto fail_child;
1187                         }
1188
1189                 if (context->ioprio_set)
1190                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1191                                 err = -errno;
1192                                 r = EXIT_IOPRIO;
1193                                 goto fail_child;
1194                         }
1195
1196                 if (context->timer_slack_nsec_set)
1197                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1198                                 err = -errno;
1199                                 r = EXIT_TIMERSLACK;
1200                                 goto fail_child;
1201                         }
1202
1203                 if (context->utmp_id)
1204                         utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1205
1206                 if (context->user) {
1207                         username = context->user;
1208                         err = get_user_creds(&username, &uid, &gid, &home);
1209                         if (err < 0) {
1210                                 r = EXIT_USER;
1211                                 goto fail_child;
1212                         }
1213
1214                         if (is_terminal_input(context->std_input)) {
1215                                 err = chown_terminal(STDIN_FILENO, uid);
1216                                 if (err < 0) {
1217                                         r = EXIT_STDIN;
1218                                         goto fail_child;
1219                                 }
1220                         }
1221
1222                         if (cgroup_bondings && context->control_group_modify) {
1223                                 err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
1224                                 if (err >= 0)
1225                                         err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid);
1226                                 if (err < 0) {
1227                                         r = EXIT_CGROUP;
1228                                         goto fail_child;
1229                                 }
1230                         }
1231                 }
1232
1233                 if (apply_permissions) {
1234                         err = enforce_groups(context, username, gid);
1235                         if (err < 0) {
1236                                 r = EXIT_GROUP;
1237                                 goto fail_child;
1238                         }
1239                 }
1240
1241                 umask(context->umask);
1242
1243 #ifdef HAVE_PAM
1244                 if (context->pam_name && username) {
1245                         err = setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds);
1246                         if (err < 0) {
1247                                 r = EXIT_PAM;
1248                                 goto fail_child;
1249                         }
1250                 }
1251 #endif
1252                 if (context->private_network) {
1253                         if (unshare(CLONE_NEWNET) < 0) {
1254                                 err = -errno;
1255                                 r = EXIT_NETWORK;
1256                                 goto fail_child;
1257                         }
1258
1259                         loopback_setup();
1260                 }
1261
1262                 if (strv_length(context->read_write_dirs) > 0 ||
1263                     strv_length(context->read_only_dirs) > 0 ||
1264                     strv_length(context->inaccessible_dirs) > 0 ||
1265                     context->mount_flags != MS_SHARED ||
1266                     context->private_tmp) {
1267                         err = setup_namespace(context->read_write_dirs,
1268                                               context->read_only_dirs,
1269                                               context->inaccessible_dirs,
1270                                               context->private_tmp,
1271                                               context->mount_flags);
1272                         if (err < 0) {
1273                                 r = EXIT_NAMESPACE;
1274                                 goto fail_child;
1275                         }
1276                 }
1277
1278                 if (apply_chroot) {
1279                         if (context->root_directory)
1280                                 if (chroot(context->root_directory) < 0) {
1281                                         err = -errno;
1282                                         r = EXIT_CHROOT;
1283                                         goto fail_child;
1284                                 }
1285
1286                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1287                                 err = -errno;
1288                                 r = EXIT_CHDIR;
1289                                 goto fail_child;
1290                         }
1291                 } else {
1292
1293                         char *d;
1294
1295                         if (asprintf(&d, "%s/%s",
1296                                      context->root_directory ? context->root_directory : "",
1297                                      context->working_directory ? context->working_directory : "") < 0) {
1298                                 err = -ENOMEM;
1299                                 r = EXIT_MEMORY;
1300                                 goto fail_child;
1301                         }
1302
1303                         if (chdir(d) < 0) {
1304                                 err = -errno;
1305                                 free(d);
1306                                 r = EXIT_CHDIR;
1307                                 goto fail_child;
1308                         }
1309
1310                         free(d);
1311                 }
1312
1313                 /* We repeat the fd closing here, to make sure that
1314                  * nothing is leaked from the PAM modules */
1315                 err = close_all_fds(fds, n_fds);
1316                 if (err >= 0)
1317                         err = shift_fds(fds, n_fds);
1318                 if (err >= 0)
1319                         err = flags_fds(fds, n_fds, context->non_blocking);
1320                 if (err < 0) {
1321                         r = EXIT_FDS;
1322                         goto fail_child;
1323                 }
1324
1325                 if (apply_permissions) {
1326
1327                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1328                                 if (!context->rlimit[i])
1329                                         continue;
1330
1331                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1332                                         err = -errno;
1333                                         r = EXIT_LIMITS;
1334                                         goto fail_child;
1335                                 }
1336                         }
1337
1338                         if (context->capability_bounding_set_drop) {
1339                                 err = do_capability_bounding_set_drop(context->capability_bounding_set_drop);
1340                                 if (err < 0) {
1341                                         r = EXIT_CAPABILITIES;
1342                                         goto fail_child;
1343                                 }
1344                         }
1345
1346                         if (context->user) {
1347                                 err = enforce_user(context, uid);
1348                                 if (err < 0) {
1349                                         r = EXIT_USER;
1350                                         goto fail_child;
1351                                 }
1352                         }
1353
1354                         /* PR_GET_SECUREBITS is not privileged, while
1355                          * PR_SET_SECUREBITS is. So to suppress
1356                          * potential EPERMs we'll try not to call
1357                          * PR_SET_SECUREBITS unless necessary. */
1358                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1359                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1360                                         err = -errno;
1361                                         r = EXIT_SECUREBITS;
1362                                         goto fail_child;
1363                                 }
1364
1365                         if (context->capabilities)
1366                                 if (cap_set_proc(context->capabilities) < 0) {
1367                                         err = -errno;
1368                                         r = EXIT_CAPABILITIES;
1369                                         goto fail_child;
1370                                 }
1371                 }
1372
1373                 if (!(our_env = new0(char*, 7))) {
1374                         err = -ENOMEM;
1375                         r = EXIT_MEMORY;
1376                         goto fail_child;
1377                 }
1378
1379                 if (n_fds > 0)
1380                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1381                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1382                                 err = -ENOMEM;
1383                                 r = EXIT_MEMORY;
1384                                 goto fail_child;
1385                         }
1386
1387                 if (home)
1388                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1389                                 err = -ENOMEM;
1390                                 r = EXIT_MEMORY;
1391                                 goto fail_child;
1392                         }
1393
1394                 if (username)
1395                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1396                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1397                                 err = -ENOMEM;
1398                                 r = EXIT_MEMORY;
1399                                 goto fail_child;
1400                         }
1401
1402                 if (is_terminal_input(context->std_input) ||
1403                     context->std_output == EXEC_OUTPUT_TTY ||
1404                     context->std_error == EXEC_OUTPUT_TTY)
1405                         if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1406                                 err = -ENOMEM;
1407                                 r = EXIT_MEMORY;
1408                                 goto fail_child;
1409                         }
1410
1411                 assert(n_env <= 7);
1412
1413                 if (!(final_env = strv_env_merge(
1414                                       5,
1415                                       environment,
1416                                       our_env,
1417                                       context->environment,
1418                                       files_env,
1419                                       pam_env,
1420                                       NULL))) {
1421                         err = -ENOMEM;
1422                         r = EXIT_MEMORY;
1423                         goto fail_child;
1424                 }
1425
1426                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1427                         err = -ENOMEM;
1428                         r = EXIT_MEMORY;
1429                         goto fail_child;
1430                 }
1431
1432                 final_env = strv_env_clean(final_env);
1433
1434                 execve(command->path, final_argv, final_env);
1435                 err = -errno;
1436                 r = EXIT_EXEC;
1437
1438         fail_child:
1439                 if (r != 0) {
1440                         log_open();
1441                         log_warning("Failed at step %s spawning %s: %s",
1442                                     exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1443                                     command->path, strerror(-err));
1444                 }
1445
1446                 strv_free(our_env);
1447                 strv_free(final_env);
1448                 strv_free(pam_env);
1449                 strv_free(files_env);
1450                 strv_free(final_argv);
1451
1452                 if (saved_stdin >= 0)
1453                         close_nointr_nofail(saved_stdin);
1454
1455                 if (saved_stdout >= 0)
1456                         close_nointr_nofail(saved_stdout);
1457
1458                 _exit(r);
1459         }
1460
1461         strv_free(files_env);
1462
1463         /* We add the new process to the cgroup both in the child (so
1464          * that we can be sure that no user code is ever executed
1465          * outside of the cgroup) and in the parent (so that we can be
1466          * sure that when we kill the cgroup the process will be
1467          * killed too). */
1468         if (cgroup_bondings)
1469                 cgroup_bonding_install_list(cgroup_bondings, pid);
1470
1471         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1472
1473         exec_status_start(&command->exec_status, pid);
1474
1475         *ret = pid;
1476         return 0;
1477
1478 fail_parent:
1479         strv_free(files_env);
1480
1481         return r;
1482 }
1483
1484 void exec_context_init(ExecContext *c) {
1485         assert(c);
1486
1487         c->umask = 0022;
1488         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1489         c->cpu_sched_policy = SCHED_OTHER;
1490         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1491         c->syslog_level_prefix = true;
1492         c->mount_flags = MS_SHARED;
1493         c->kill_signal = SIGTERM;
1494         c->send_sigkill = true;
1495 }
1496
1497 void exec_context_done(ExecContext *c) {
1498         unsigned l;
1499
1500         assert(c);
1501
1502         strv_free(c->environment);
1503         c->environment = NULL;
1504
1505         strv_free(c->environment_files);
1506         c->environment_files = NULL;
1507
1508         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1509                 free(c->rlimit[l]);
1510                 c->rlimit[l] = NULL;
1511         }
1512
1513         free(c->working_directory);
1514         c->working_directory = NULL;
1515         free(c->root_directory);
1516         c->root_directory = NULL;
1517
1518         free(c->tty_path);
1519         c->tty_path = NULL;
1520
1521         free(c->tcpwrap_name);
1522         c->tcpwrap_name = NULL;
1523
1524         free(c->syslog_identifier);
1525         c->syslog_identifier = NULL;
1526
1527         free(c->user);
1528         c->user = NULL;
1529
1530         free(c->group);
1531         c->group = NULL;
1532
1533         strv_free(c->supplementary_groups);
1534         c->supplementary_groups = NULL;
1535
1536         free(c->pam_name);
1537         c->pam_name = NULL;
1538
1539         if (c->capabilities) {
1540                 cap_free(c->capabilities);
1541                 c->capabilities = NULL;
1542         }
1543
1544         strv_free(c->read_only_dirs);
1545         c->read_only_dirs = NULL;
1546
1547         strv_free(c->read_write_dirs);
1548         c->read_write_dirs = NULL;
1549
1550         strv_free(c->inaccessible_dirs);
1551         c->inaccessible_dirs = NULL;
1552
1553         if (c->cpuset)
1554                 CPU_FREE(c->cpuset);
1555
1556         free(c->utmp_id);
1557         c->utmp_id = NULL;
1558 }
1559
1560 void exec_command_done(ExecCommand *c) {
1561         assert(c);
1562
1563         free(c->path);
1564         c->path = NULL;
1565
1566         strv_free(c->argv);
1567         c->argv = NULL;
1568 }
1569
1570 void exec_command_done_array(ExecCommand *c, unsigned n) {
1571         unsigned i;
1572
1573         for (i = 0; i < n; i++)
1574                 exec_command_done(c+i);
1575 }
1576
1577 void exec_command_free_list(ExecCommand *c) {
1578         ExecCommand *i;
1579
1580         while ((i = c)) {
1581                 LIST_REMOVE(ExecCommand, command, c, i);
1582                 exec_command_done(i);
1583                 free(i);
1584         }
1585 }
1586
1587 void exec_command_free_array(ExecCommand **c, unsigned n) {
1588         unsigned i;
1589
1590         for (i = 0; i < n; i++) {
1591                 exec_command_free_list(c[i]);
1592                 c[i] = NULL;
1593         }
1594 }
1595
1596 int exec_context_load_environment(const ExecContext *c, char ***l) {
1597         char **i, **r = NULL;
1598
1599         assert(c);
1600         assert(l);
1601
1602         STRV_FOREACH(i, c->environment_files) {
1603                 char *fn;
1604                 int k;
1605                 bool ignore = false;
1606                 char **p;
1607
1608                 fn = *i;
1609
1610                 if (fn[0] == '-') {
1611                         ignore = true;
1612                         fn ++;
1613                 }
1614
1615                 if (!path_is_absolute(fn)) {
1616
1617                         if (ignore)
1618                                 continue;
1619
1620                         strv_free(r);
1621                         return -EINVAL;
1622                 }
1623
1624                 if ((k = load_env_file(fn, &p)) < 0) {
1625
1626                         if (ignore)
1627                                 continue;
1628
1629                         strv_free(r);
1630                         return k;
1631                 }
1632
1633                 if (r == NULL)
1634                         r = p;
1635                 else {
1636                         char **m;
1637
1638                         m = strv_env_merge(2, r, p);
1639                         strv_free(r);
1640                         strv_free(p);
1641
1642                         if (!m)
1643                                 return -ENOMEM;
1644
1645                         r = m;
1646                 }
1647         }
1648
1649         *l = r;
1650
1651         return 0;
1652 }
1653
1654 static void strv_fprintf(FILE *f, char **l) {
1655         char **g;
1656
1657         assert(f);
1658
1659         STRV_FOREACH(g, l)
1660                 fprintf(f, " %s", *g);
1661 }
1662
1663 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1664         char ** e;
1665         unsigned i;
1666
1667         assert(c);
1668         assert(f);
1669
1670         if (!prefix)
1671                 prefix = "";
1672
1673         fprintf(f,
1674                 "%sUMask: %04o\n"
1675                 "%sWorkingDirectory: %s\n"
1676                 "%sRootDirectory: %s\n"
1677                 "%sNonBlocking: %s\n"
1678                 "%sPrivateTmp: %s\n"
1679                 "%sControlGroupModify: %s\n"
1680                 "%sPrivateNetwork: %s\n",
1681                 prefix, c->umask,
1682                 prefix, c->working_directory ? c->working_directory : "/",
1683                 prefix, c->root_directory ? c->root_directory : "/",
1684                 prefix, yes_no(c->non_blocking),
1685                 prefix, yes_no(c->private_tmp),
1686                 prefix, yes_no(c->control_group_modify),
1687                 prefix, yes_no(c->private_network));
1688
1689         STRV_FOREACH(e, c->environment)
1690                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1691
1692         STRV_FOREACH(e, c->environment_files)
1693                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1694
1695         if (c->tcpwrap_name)
1696                 fprintf(f,
1697                         "%sTCPWrapName: %s\n",
1698                         prefix, c->tcpwrap_name);
1699
1700         if (c->nice_set)
1701                 fprintf(f,
1702                         "%sNice: %i\n",
1703                         prefix, c->nice);
1704
1705         if (c->oom_score_adjust_set)
1706                 fprintf(f,
1707                         "%sOOMScoreAdjust: %i\n",
1708                         prefix, c->oom_score_adjust);
1709
1710         for (i = 0; i < RLIM_NLIMITS; i++)
1711                 if (c->rlimit[i])
1712                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1713
1714         if (c->ioprio_set)
1715                 fprintf(f,
1716                         "%sIOSchedulingClass: %s\n"
1717                         "%sIOPriority: %i\n",
1718                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1719                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1720
1721         if (c->cpu_sched_set)
1722                 fprintf(f,
1723                         "%sCPUSchedulingPolicy: %s\n"
1724                         "%sCPUSchedulingPriority: %i\n"
1725                         "%sCPUSchedulingResetOnFork: %s\n",
1726                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1727                         prefix, c->cpu_sched_priority,
1728                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1729
1730         if (c->cpuset) {
1731                 fprintf(f, "%sCPUAffinity:", prefix);
1732                 for (i = 0; i < c->cpuset_ncpus; i++)
1733                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1734                                 fprintf(f, " %i", i);
1735                 fputs("\n", f);
1736         }
1737
1738         if (c->timer_slack_nsec_set)
1739                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1740
1741         fprintf(f,
1742                 "%sStandardInput: %s\n"
1743                 "%sStandardOutput: %s\n"
1744                 "%sStandardError: %s\n",
1745                 prefix, exec_input_to_string(c->std_input),
1746                 prefix, exec_output_to_string(c->std_output),
1747                 prefix, exec_output_to_string(c->std_error));
1748
1749         if (c->tty_path)
1750                 fprintf(f,
1751                         "%sTTYPath: %s\n"
1752                         "%sTTYReset: %s\n"
1753                         "%sTTYVHangup: %s\n"
1754                         "%sTTYVTDisallocate: %s\n",
1755                         prefix, c->tty_path,
1756                         prefix, yes_no(c->tty_reset),
1757                         prefix, yes_no(c->tty_vhangup),
1758                         prefix, yes_no(c->tty_vt_disallocate));
1759
1760         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1761             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1762             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1763             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1764                 fprintf(f,
1765                         "%sSyslogFacility: %s\n"
1766                         "%sSyslogLevel: %s\n",
1767                         prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1768                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1769
1770         if (c->capabilities) {
1771                 char *t;
1772                 if ((t = cap_to_text(c->capabilities, NULL))) {
1773                         fprintf(f, "%sCapabilities: %s\n",
1774                                 prefix, t);
1775                         cap_free(t);
1776                 }
1777         }
1778
1779         if (c->secure_bits)
1780                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1781                         prefix,
1782                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1783                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1784                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1785                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1786                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1787                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1788
1789         if (c->capability_bounding_set_drop) {
1790                 unsigned long l;
1791                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1792
1793                 for (l = 0; l <= cap_last_cap(); l++)
1794                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1795                                 char *t;
1796
1797                                 if ((t = cap_to_name(l))) {
1798                                         fprintf(f, " %s", t);
1799                                         cap_free(t);
1800                                 }
1801                         }
1802
1803                 fputs("\n", f);
1804         }
1805
1806         if (c->user)
1807                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1808         if (c->group)
1809                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1810
1811         if (strv_length(c->supplementary_groups) > 0) {
1812                 fprintf(f, "%sSupplementaryGroups:", prefix);
1813                 strv_fprintf(f, c->supplementary_groups);
1814                 fputs("\n", f);
1815         }
1816
1817         if (c->pam_name)
1818                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1819
1820         if (strv_length(c->read_write_dirs) > 0) {
1821                 fprintf(f, "%sReadWriteDirs:", prefix);
1822                 strv_fprintf(f, c->read_write_dirs);
1823                 fputs("\n", f);
1824         }
1825
1826         if (strv_length(c->read_only_dirs) > 0) {
1827                 fprintf(f, "%sReadOnlyDirs:", prefix);
1828                 strv_fprintf(f, c->read_only_dirs);
1829                 fputs("\n", f);
1830         }
1831
1832         if (strv_length(c->inaccessible_dirs) > 0) {
1833                 fprintf(f, "%sInaccessibleDirs:", prefix);
1834                 strv_fprintf(f, c->inaccessible_dirs);
1835                 fputs("\n", f);
1836         }
1837
1838         fprintf(f,
1839                 "%sKillMode: %s\n"
1840                 "%sKillSignal: SIG%s\n"
1841                 "%sSendSIGKILL: %s\n",
1842                 prefix, kill_mode_to_string(c->kill_mode),
1843                 prefix, signal_to_string(c->kill_signal),
1844                 prefix, yes_no(c->send_sigkill));
1845
1846         if (c->utmp_id)
1847                 fprintf(f,
1848                         "%sUtmpIdentifier: %s\n",
1849                         prefix, c->utmp_id);
1850 }
1851
1852 void exec_status_start(ExecStatus *s, pid_t pid) {
1853         assert(s);
1854
1855         zero(*s);
1856         s->pid = pid;
1857         dual_timestamp_get(&s->start_timestamp);
1858 }
1859
1860 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1861         assert(s);
1862
1863         if (s->pid && s->pid != pid)
1864                 zero(*s);
1865
1866         s->pid = pid;
1867         dual_timestamp_get(&s->exit_timestamp);
1868
1869         s->code = code;
1870         s->status = status;
1871
1872         if (context) {
1873                 if (context->utmp_id)
1874                         utmp_put_dead_process(context->utmp_id, pid, code, status);
1875
1876                 exec_context_tty_reset(context);
1877         }
1878 }
1879
1880 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1881         char buf[FORMAT_TIMESTAMP_MAX];
1882
1883         assert(s);
1884         assert(f);
1885
1886         if (!prefix)
1887                 prefix = "";
1888
1889         if (s->pid <= 0)
1890                 return;
1891
1892         fprintf(f,
1893                 "%sPID: %lu\n",
1894                 prefix, (unsigned long) s->pid);
1895
1896         if (s->start_timestamp.realtime > 0)
1897                 fprintf(f,
1898                         "%sStart Timestamp: %s\n",
1899                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1900
1901         if (s->exit_timestamp.realtime > 0)
1902                 fprintf(f,
1903                         "%sExit Timestamp: %s\n"
1904                         "%sExit Code: %s\n"
1905                         "%sExit Status: %i\n",
1906                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1907                         prefix, sigchld_code_to_string(s->code),
1908                         prefix, s->status);
1909 }
1910
1911 char *exec_command_line(char **argv) {
1912         size_t k;
1913         char *n, *p, **a;
1914         bool first = true;
1915
1916         assert(argv);
1917
1918         k = 1;
1919         STRV_FOREACH(a, argv)
1920                 k += strlen(*a)+3;
1921
1922         if (!(n = new(char, k)))
1923                 return NULL;
1924
1925         p = n;
1926         STRV_FOREACH(a, argv) {
1927
1928                 if (!first)
1929                         *(p++) = ' ';
1930                 else
1931                         first = false;
1932
1933                 if (strpbrk(*a, WHITESPACE)) {
1934                         *(p++) = '\'';
1935                         p = stpcpy(p, *a);
1936                         *(p++) = '\'';
1937                 } else
1938                         p = stpcpy(p, *a);
1939
1940         }
1941
1942         *p = 0;
1943
1944         /* FIXME: this doesn't really handle arguments that have
1945          * spaces and ticks in them */
1946
1947         return n;
1948 }
1949
1950 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1951         char *p2;
1952         const char *prefix2;
1953
1954         char *cmd;
1955
1956         assert(c);
1957         assert(f);
1958
1959         if (!prefix)
1960                 prefix = "";
1961         p2 = strappend(prefix, "\t");
1962         prefix2 = p2 ? p2 : prefix;
1963
1964         cmd = exec_command_line(c->argv);
1965
1966         fprintf(f,
1967                 "%sCommand Line: %s\n",
1968                 prefix, cmd ? cmd : strerror(ENOMEM));
1969
1970         free(cmd);
1971
1972         exec_status_dump(&c->exec_status, f, prefix2);
1973
1974         free(p2);
1975 }
1976
1977 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1978         assert(f);
1979
1980         if (!prefix)
1981                 prefix = "";
1982
1983         LIST_FOREACH(command, c, c)
1984                 exec_command_dump(c, f, prefix);
1985 }
1986
1987 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1988         ExecCommand *end;
1989
1990         assert(l);
1991         assert(e);
1992
1993         if (*l) {
1994                 /* It's kind of important, that we keep the order here */
1995                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1996                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1997         } else
1998               *l = e;
1999 }
2000
2001 int exec_command_set(ExecCommand *c, const char *path, ...) {
2002         va_list ap;
2003         char **l, *p;
2004
2005         assert(c);
2006         assert(path);
2007
2008         va_start(ap, path);
2009         l = strv_new_ap(path, ap);
2010         va_end(ap);
2011
2012         if (!l)
2013                 return -ENOMEM;
2014
2015         if (!(p = strdup(path))) {
2016                 strv_free(l);
2017                 return -ENOMEM;
2018         }
2019
2020         free(c->path);
2021         c->path = p;
2022
2023         strv_free(c->argv);
2024         c->argv = l;
2025
2026         return 0;
2027 }
2028
2029 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2030         [EXEC_INPUT_NULL] = "null",
2031         [EXEC_INPUT_TTY] = "tty",
2032         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2033         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2034         [EXEC_INPUT_SOCKET] = "socket"
2035 };
2036
2037 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2038
2039 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2040         [EXEC_OUTPUT_INHERIT] = "inherit",
2041         [EXEC_OUTPUT_NULL] = "null",
2042         [EXEC_OUTPUT_TTY] = "tty",
2043         [EXEC_OUTPUT_SYSLOG] = "syslog",
2044         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2045         [EXEC_OUTPUT_KMSG] = "kmsg",
2046         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2047         [EXEC_OUTPUT_SOCKET] = "socket"
2048 };
2049
2050 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2051
2052 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2053         [KILL_CONTROL_GROUP] = "control-group",
2054         [KILL_PROCESS] = "process",
2055         [KILL_NONE] = "none"
2056 };
2057
2058 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2059
2060 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2061         [KILL_MAIN] = "main",
2062         [KILL_CONTROL] = "control",
2063         [KILL_ALL] = "all"
2064 };
2065
2066 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);