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