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
  60 /* This assumes there is a 'tty' group */
  61 #define TTY_MODE 0620
  62
  63 static int shift_fds(int fds[], unsigned n_fds) {
  64         int start, restart_from;
  65
  66         if (n_fds <= 0)
  67                 return 0;
  68
  69         /* Modifies the fds array! (sorts it) */
  70
  71         assert(fds);
  72
  73         start = 0;
  74         for (;;) {
  75                 int i;
  76
  77                 restart_from = -1;
  78
  79                 for (i = start; i < (int) n_fds; i++) {
  80                         int nfd;
  81
  82                         /* Already at right index? */
  83                         if (fds[i] == i+3)
  84                                 continue;
  85
  86                         if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
  87                                 return -errno;
  88
  89                         close_nointr_nofail(fds[i]);
  90                         fds[i] = nfd;
  91
  92                         /* Hmm, the fd we wanted isn't free? Then
  93                          * let's remember that and try again from here*/
  94                         if (nfd != i+3 && restart_from < 0)
  95                                 restart_from = i;
  96                 }
  97
  98                 if (restart_from < 0)
  99                         break;
 100
 101                 start = restart_from;
 102         }
 103
 104         return 0;
 105 }
 106
 107 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
 108         unsigned i;
 109         int r;
 110
 111         if (n_fds <= 0)
 112                 return 0;
 113
 114         assert(fds);
 115
 116         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
 117
 118         for (i = 0; i < n_fds; i++) {
 119
 120                 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
 121                         return r;
 122
 123                 /* We unconditionally drop FD_CLOEXEC from the fds,
 124                  * since after all we want to pass these fds to our
 125                  * children */
 126
 127                 if ((r = fd_cloexec(fds[i], false)) < 0)
 128                         return r;
 129         }
 130
 131         return 0;
 132 }
 133
 134 static const char *tty_path(const ExecContext *context) {
 135         assert(context);
 136
 137         if (context->tty_path)
 138                 return context->tty_path;
 139
 140         return "/dev/console";
 141 }
 142
 143 void exec_context_tty_reset(const ExecContext *context) {
 144         assert(context);
 145
 146         if (context->tty_vhangup)
 147                 terminal_vhangup(tty_path(context));
 148
 149         if (context->tty_reset)
 150                 reset_terminal(tty_path(context));
 151
 152         if (context->tty_vt_disallocate && context->tty_path)
 153                 vt_disallocate(context->tty_path);
 154 }
 155
 156 static int open_null_as(int flags, int nfd) {
 157         int fd, r;
 158
 159         assert(nfd >= 0);
 160
 161         if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
 162                 return -errno;
 163
 164         if (fd != nfd) {
 165                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 166                 close_nointr_nofail(fd);
 167         } else
 168                 r = nfd;
 169
 170         return r;
 171 }
 172
 173 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
 174         int fd, r;
 175         union {
 176                 struct sockaddr sa;
 177                 struct sockaddr_un un;
 178         } sa;
 179
 180         assert(context);
 181         assert(output < _EXEC_OUTPUT_MAX);
 182         assert(ident);
 183         assert(nfd >= 0);
 184
 185         if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
 186                 return -errno;
 187
 188         zero(sa);
 189         sa.sa.sa_family = AF_UNIX;
 190         strncpy(sa.un.sun_path, LOGGER_SOCKET, sizeof(sa.un.sun_path));
 191
 192         if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + sizeof(LOGGER_SOCKET) - 1) < 0) {
 193                 close_nointr_nofail(fd);
 194                 return -errno;
 195         }
 196
 197         if (shutdown(fd, SHUT_RD) < 0) {
 198                 close_nointr_nofail(fd);
 199                 return -errno;
 200         }
 201
 202         /* We speak a very simple protocol between log server
 203          * and client: one line for the log destination (kmsg
 204          * or syslog), followed by the priority field,
 205          * followed by the process name. Since we replaced
 206          * stdin/stderr we simple use stdio to write to
 207          * it. Note that we use stderr, to minimize buffer
 208          * flushing issues. */
 209
 210         dprintf(fd,
 211                 "%s\n"
 212                 "%i\n"
 213                 "%s\n"
 214                 "%i\n",
 215                 output == EXEC_OUTPUT_KMSG ?             "kmsg" :
 216                 output == EXEC_OUTPUT_KMSG_AND_CONSOLE ? "kmsg+console" :
 217                 output == EXEC_OUTPUT_SYSLOG ?           "syslog" :
 218                                                          "syslog+console",
 219                 context->syslog_priority,
 220                 context->syslog_identifier ? context->syslog_identifier : ident,
 221                 context->syslog_level_prefix);
 222
 223         if (fd != nfd) {
 224                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 225                 close_nointr_nofail(fd);
 226         } else
 227                 r = nfd;
 228
 229         return r;
 230 }
 231 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
 232         int fd, r;
 233
 234         assert(path);
 235         assert(nfd >= 0);
 236
 237         if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
 238                 return fd;
 239
 240         if (fd != nfd) {
 241                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 242                 close_nointr_nofail(fd);
 243         } else
 244                 r = nfd;
 245
 246         return r;
 247 }
 248
 249 static bool is_terminal_input(ExecInput i) {
 250         return
 251                 i == EXEC_INPUT_TTY ||
 252                 i == EXEC_INPUT_TTY_FORCE ||
 253                 i == EXEC_INPUT_TTY_FAIL;
 254 }
 255
 256 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
 257
 258         if (is_terminal_input(std_input) && !apply_tty_stdin)
 259                 return EXEC_INPUT_NULL;
 260
 261         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 262                 return EXEC_INPUT_NULL;
 263
 264         return std_input;
 265 }
 266
 267 static int fixup_output(ExecOutput std_output, int socket_fd) {
 268
 269         if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 270                 return EXEC_OUTPUT_INHERIT;
 271
 272         return std_output;
 273 }
 274
 275 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
 276         ExecInput i;
 277
 278         assert(context);
 279
 280         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 281
 282         switch (i) {
 283
 284         case EXEC_INPUT_NULL:
 285                 return open_null_as(O_RDONLY, STDIN_FILENO);
 286
 287         case EXEC_INPUT_TTY:
 288         case EXEC_INPUT_TTY_FORCE:
 289         case EXEC_INPUT_TTY_FAIL: {
 290                 int fd, r;
 291
 292                 if ((fd = acquire_terminal(
 293                                      tty_path(context),
 294                                      i == EXEC_INPUT_TTY_FAIL,
 295                                      i == EXEC_INPUT_TTY_FORCE,
 296                                      false)) < 0)
 297                         return fd;
 298
 299                 if (fd != STDIN_FILENO) {
 300                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 301                         close_nointr_nofail(fd);
 302                 } else
 303                         r = STDIN_FILENO;
 304
 305                 return r;
 306         }
 307
 308         case EXEC_INPUT_SOCKET:
 309                 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 310
 311         default:
 312                 assert_not_reached("Unknown input type");
 313         }
 314 }
 315
 316 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 317         ExecOutput o;
 318         ExecInput i;
 319
 320         assert(context);
 321         assert(ident);
 322
 323         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 324         o = fixup_output(context->std_output, socket_fd);
 325
 326         /* This expects the input is already set up */
 327
 328         switch (o) {
 329
 330         case EXEC_OUTPUT_INHERIT:
 331
 332                 /* If input got downgraded, inherit the original value */
 333                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 334                         return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 335
 336                 /* If the input is connected to anything that's not a /dev/null, inherit that... */
 337                 if (i != EXEC_INPUT_NULL)
 338                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 339
 340                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 341                 if (getppid() != 1)
 342                         return STDOUT_FILENO;
 343
 344                 /* We need to open /dev/null here anew, to get the
 345                  * right access mode. So we fall through */
 346
 347         case EXEC_OUTPUT_NULL:
 348                 return open_null_as(O_WRONLY, STDOUT_FILENO);
 349
 350         case EXEC_OUTPUT_TTY:
 351                 if (is_terminal_input(i))
 352                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 353
 354                 /* We don't reset the terminal if this is just about output */
 355                 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 356
 357         case EXEC_OUTPUT_SYSLOG:
 358         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 359         case EXEC_OUTPUT_KMSG:
 360         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 361                 return connect_logger_as(context, o, ident, STDOUT_FILENO);
 362
 363         case EXEC_OUTPUT_SOCKET:
 364                 assert(socket_fd >= 0);
 365                 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 366
 367         default:
 368                 assert_not_reached("Unknown output type");
 369         }
 370 }
 371
 372 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 373         ExecOutput o, e;
 374         ExecInput i;
 375
 376         assert(context);
 377         assert(ident);
 378
 379         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 380         o = fixup_output(context->std_output, socket_fd);
 381         e = fixup_output(context->std_error, socket_fd);
 382
 383         /* This expects the input and output are already set up */
 384
 385         /* Don't change the stderr file descriptor if we inherit all
 386          * the way and are not on a tty */
 387         if (e == EXEC_OUTPUT_INHERIT &&
 388             o == EXEC_OUTPUT_INHERIT &&
 389             i == EXEC_INPUT_NULL &&
 390             !is_terminal_input(context->std_input) &&
 391             getppid () != 1)
 392                 return STDERR_FILENO;
 393
 394         /* Duplicate from stdout if possible */
 395         if (e == o || e == EXEC_OUTPUT_INHERIT)
 396                 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 397
 398         switch (e) {
 399
 400         case EXEC_OUTPUT_NULL:
 401                 return open_null_as(O_WRONLY, STDERR_FILENO);
 402
 403         case EXEC_OUTPUT_TTY:
 404                 if (is_terminal_input(i))
 405                         return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 406
 407                 /* We don't reset the terminal if this is just about output */
 408                 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
 409
 410         case EXEC_OUTPUT_SYSLOG:
 411         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 412         case EXEC_OUTPUT_KMSG:
 413         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 414                 return connect_logger_as(context, e, ident, STDERR_FILENO);
 415
 416         case EXEC_OUTPUT_SOCKET:
 417                 assert(socket_fd >= 0);
 418                 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 419
 420         default:
 421                 assert_not_reached("Unknown error type");
 422         }
 423 }
 424
 425 static int chown_terminal(int fd, uid_t uid) {
 426         struct stat st;
 427
 428         assert(fd >= 0);
 429
 430         /* This might fail. What matters are the results. */
 431         (void) fchown(fd, uid, -1);
 432         (void) fchmod(fd, TTY_MODE);
 433
 434         if (fstat(fd, &st) < 0)
 435                 return -errno;
 436
 437         if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
 438                 return -EPERM;
 439
 440         return 0;
 441 }
 442
 443 static int setup_confirm_stdio(const ExecContext *context,
 444                                int *_saved_stdin,
 445                                int *_saved_stdout) {
 446         int fd = -1, saved_stdin, saved_stdout = -1, r;
 447
 448         assert(context);
 449         assert(_saved_stdin);
 450         assert(_saved_stdout);
 451
 452         /* This returns positive EXIT_xxx return values instead of
 453          * negative errno style values! */
 454
 455         if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
 456                 return EXIT_STDIN;
 457
 458         if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
 459                 r = EXIT_STDOUT;
 460                 goto fail;
 461         }
 462
 463         if ((fd = acquire_terminal(
 464                              tty_path(context),
 465                              context->std_input == EXEC_INPUT_TTY_FAIL,
 466                              context->std_input == EXEC_INPUT_TTY_FORCE,
 467                              false)) < 0) {
 468                 r = EXIT_STDIN;
 469                 goto fail;
 470         }
 471
 472         if (chown_terminal(fd, getuid()) < 0) {
 473                 r = EXIT_STDIN;
 474                 goto fail;
 475         }
 476
 477         if (dup2(fd, STDIN_FILENO) < 0) {
 478                 r = EXIT_STDIN;
 479                 goto fail;
 480         }
 481
 482         if (dup2(fd, STDOUT_FILENO) < 0) {
 483                 r = EXIT_STDOUT;
 484                 goto fail;
 485         }
 486
 487         if (fd >= 2)
 488                 close_nointr_nofail(fd);
 489
 490         *_saved_stdin = saved_stdin;
 491         *_saved_stdout = saved_stdout;
 492
 493         return 0;
 494
 495 fail:
 496         if (saved_stdout >= 0)
 497                 close_nointr_nofail(saved_stdout);
 498
 499         if (saved_stdin >= 0)
 500                 close_nointr_nofail(saved_stdin);
 501
 502         if (fd >= 0)
 503                 close_nointr_nofail(fd);
 504
 505         return r;
 506 }
 507
 508 static int restore_confirm_stdio(const ExecContext *context,
 509                                  int *saved_stdin,
 510                                  int *saved_stdout,
 511                                  bool *keep_stdin,
 512                                  bool *keep_stdout) {
 513
 514         assert(context);
 515         assert(saved_stdin);
 516         assert(*saved_stdin >= 0);
 517         assert(saved_stdout);
 518         assert(*saved_stdout >= 0);
 519
 520         /* This returns positive EXIT_xxx return values instead of
 521          * negative errno style values! */
 522
 523         if (is_terminal_input(context->std_input)) {
 524
 525                 /* The service wants terminal input. */
 526
 527                 *keep_stdin = true;
 528                 *keep_stdout =
 529                         context->std_output == EXEC_OUTPUT_INHERIT ||
 530                         context->std_output == EXEC_OUTPUT_TTY;
 531
 532         } else {
 533                 /* If the service doesn't want a controlling terminal,
 534                  * then we need to get rid entirely of what we have
 535                  * already. */
 536
 537                 if (release_terminal() < 0)
 538                         return EXIT_STDIN;
 539
 540                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 541                         return EXIT_STDIN;
 542
 543                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 544                         return EXIT_STDOUT;
 545
 546                 *keep_stdout = *keep_stdin = false;
 547         }
 548
 549         return 0;
 550 }
 551
 552 static int get_group_creds(const char *groupname, gid_t *gid) {
 553         struct group *g;
 554         unsigned long lu;
 555
 556         assert(groupname);
 557         assert(gid);
 558
 559         /* We enforce some special rules for gid=0: in order to avoid
 560          * NSS lookups for root we hardcode its data. */
 561
 562         if (streq(groupname, "root") || streq(groupname, "0")) {
 563                 *gid = 0;
 564                 return 0;
 565         }
 566
 567         if (safe_atolu(groupname, &lu) >= 0) {
 568                 errno = 0;
 569                 g = getgrgid((gid_t) lu);
 570         } else {
 571                 errno = 0;
 572                 g = getgrnam(groupname);
 573         }
 574
 575         if (!g)
 576                 return errno != 0 ? -errno : -ESRCH;
 577
 578         *gid = g->gr_gid;
 579         return 0;
 580 }
 581
 582 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
 583         bool keep_groups = false;
 584         int r;
 585
 586         assert(context);
 587
 588         /* Lookup and set GID and supplementary group list. Here too
 589          * we avoid NSS lookups for gid=0. */
 590
 591         if (context->group || username) {
 592
 593                 if (context->group)
 594                         if ((r = get_group_creds(context->group, &gid)) < 0)
 595                                 return r;
 596
 597                 /* First step, initialize groups from /etc/groups */
 598                 if (username && gid != 0) {
 599                         if (initgroups(username, gid) < 0)
 600                                 return -errno;
 601
 602                         keep_groups = true;
 603                 }
 604
 605                 /* Second step, set our gids */
 606                 if (setresgid(gid, gid, gid) < 0)
 607                         return -errno;
 608         }
 609
 610         if (context->supplementary_groups) {
 611                 int ngroups_max, k;
 612                 gid_t *gids;
 613                 char **i;
 614
 615                 /* Final step, initialize any manually set supplementary groups */
 616                 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
 617
 618                 if (!(gids = new(gid_t, ngroups_max)))
 619                         return -ENOMEM;
 620
 621                 if (keep_groups) {
 622                         if ((k = getgroups(ngroups_max, gids)) < 0) {
 623                                 free(gids);
 624                                 return -errno;
 625                         }
 626                 } else
 627                         k = 0;
 628
 629                 STRV_FOREACH(i, context->supplementary_groups) {
 630
 631                         if (k >= ngroups_max) {
 632                                 free(gids);
 633                                 return -E2BIG;
 634                         }
 635
 636                         if ((r = get_group_creds(*i, gids+k)) < 0) {
 637                                 free(gids);
 638                                 return r;
 639                         }
 640
 641                         k++;
 642                 }
 643
 644                 if (setgroups(k, gids) < 0) {
 645                         free(gids);
 646                         return -errno;
 647                 }
 648
 649                 free(gids);
 650         }
 651
 652         return 0;
 653 }
 654
 655 static int enforce_user(const ExecContext *context, uid_t uid) {
 656         int r;
 657         assert(context);
 658
 659         /* Sets (but doesn't lookup) the uid and make sure we keep the
 660          * capabilities while doing so. */
 661
 662         if (context->capabilities) {
 663                 cap_t d;
 664                 static const cap_value_t bits[] = {
 665                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
 666                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
 667                 };
 668
 669                 /* First step: If we need to keep capabilities but
 670                  * drop privileges we need to make sure we keep our
 671                  * caps, whiel we drop privileges. */
 672                 if (uid != 0) {
 673                         int sb = context->secure_bits|SECURE_KEEP_CAPS;
 674
 675                         if (prctl(PR_GET_SECUREBITS) != sb)
 676                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 677                                         return -errno;
 678                 }
 679
 680                 /* Second step: set the capabilities. This will reduce
 681                  * the capabilities to the minimum we need. */
 682
 683                 if (!(d = cap_dup(context->capabilities)))
 684                         return -errno;
 685
 686                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
 687                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
 688                         r = -errno;
 689                         cap_free(d);
 690                         return r;
 691                 }
 692
 693                 if (cap_set_proc(d) < 0) {
 694                         r = -errno;
 695                         cap_free(d);
 696                         return r;
 697                 }
 698
 699                 cap_free(d);
 700         }
 701
 702         /* Third step: actually set the uids */
 703         if (setresuid(uid, uid, uid) < 0)
 704                 return -errno;
 705
 706         /* At this point we should have all necessary capabilities but
 707            are otherwise a normal user. However, the caps might got
 708            corrupted due to the setresuid() so we need clean them up
 709            later. This is done outside of this call. */
 710
 711         return 0;
 712 }
 713
 714 #ifdef HAVE_PAM
 715
 716 static int null_conv(
 717                 int num_msg,
 718                 const struct pam_message **msg,
 719                 struct pam_response **resp,
 720                 void *appdata_ptr) {
 721
 722         /* We don't support conversations */
 723
 724         return PAM_CONV_ERR;
 725 }
 726
 727 static int setup_pam(
 728                 const char *name,
 729                 const char *user,
 730                 const char *tty,
 731                 char ***pam_env,
 732                 int fds[], unsigned n_fds) {
 733
 734         static const struct pam_conv conv = {
 735                 .conv = null_conv,
 736                 .appdata_ptr = NULL
 737         };
 738
 739         pam_handle_t *handle = NULL;
 740         sigset_t ss, old_ss;
 741         int pam_code = PAM_SUCCESS;
 742         char **e = NULL;
 743         bool close_session = false;
 744         pid_t pam_pid = 0, parent_pid;
 745
 746         assert(name);
 747         assert(user);
 748         assert(pam_env);
 749
 750         /* We set up PAM in the parent process, then fork. The child
 751          * will then stay around until killed via PR_GET_PDEATHSIG or
 752          * systemd via the cgroup logic. It will then remove the PAM
 753          * session again. The parent process will exec() the actual
 754          * daemon. We do things this way to ensure that the main PID
 755          * of the daemon is the one we initially fork()ed. */
 756
 757         if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
 758                 handle = NULL;
 759                 goto fail;
 760         }
 761
 762         if (tty)
 763                 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
 764                         goto fail;
 765
 766         if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
 767                 goto fail;
 768
 769         if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
 770                 goto fail;
 771
 772         close_session = true;
 773
 774         if ((!(e = pam_getenvlist(handle)))) {
 775                 pam_code = PAM_BUF_ERR;
 776                 goto fail;
 777         }
 778
 779         /* Block SIGTERM, so that we know that it won't get lost in
 780          * the child */
 781         if (sigemptyset(&ss) < 0 ||
 782             sigaddset(&ss, SIGTERM) < 0 ||
 783             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 784                 goto fail;
 785
 786         parent_pid = getpid();
 787
 788         if ((pam_pid = fork()) < 0)
 789                 goto fail;
 790
 791         if (pam_pid == 0) {
 792                 int sig;
 793                 int r = EXIT_PAM;
 794
 795                 /* The child's job is to reset the PAM session on
 796                  * termination */
 797
 798                 /* This string must fit in 10 chars (i.e. the length
 799                  * of "/sbin/init") */
 800                 rename_process("sd(PAM)");
 801
 802                 /* Make sure we don't keep open the passed fds in this
 803                 child. We assume that otherwise only those fds are
 804                 open here that have been opened by PAM. */
 805                 close_many(fds, n_fds);
 806
 807                 /* Wait until our parent died. This will most likely
 808                  * not work since the kernel does not allow
 809                  * unprivileged parents kill their privileged children
 810                  * this way. We rely on the control groups kill logic
 811                  * to do the rest for us. */
 812                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 813                         goto child_finish;
 814
 815                 /* Check if our parent process might already have
 816                  * died? */
 817                 if (getppid() == parent_pid) {
 818                         for (;;) {
 819                                 if (sigwait(&ss, &sig) < 0) {
 820                                         if (errno == EINTR)
 821                                                 continue;
 822
 823                                         goto child_finish;
 824                                 }
 825
 826                                 assert(sig == SIGTERM);
 827                                 break;
 828                         }
 829                 }
 830
 831                 /* If our parent died we'll end the session */
 832                 if (getppid() != parent_pid)
 833                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
 834                                 goto child_finish;
 835
 836                 r = 0;
 837
 838         child_finish:
 839                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 840                 _exit(r);
 841         }
 842
 843         /* If the child was forked off successfully it will do all the
 844          * cleanups, so forget about the handle here. */
 845         handle = NULL;
 846
 847         /* Unblock SIGTERM again in the parent */
 848         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 849                 goto fail;
 850
 851         /* We close the log explicitly here, since the PAM modules
 852          * might have opened it, but we don't want this fd around. */
 853         closelog();
 854
 855         *pam_env = e;
 856         e = NULL;
 857
 858         return 0;
 859
 860 fail:
 861         if (handle) {
 862                 if (close_session)
 863                         pam_code = pam_close_session(handle, PAM_DATA_SILENT);
 864
 865                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 866         }
 867
 868         strv_free(e);
 869
 870         closelog();
 871
 872         if (pam_pid > 1) {
 873                 kill(pam_pid, SIGTERM);
 874                 kill(pam_pid, SIGCONT);
 875         }
 876
 877         return EXIT_PAM;
 878 }
 879 #endif
 880
 881 static int do_capability_bounding_set_drop(uint64_t drop) {
 882         unsigned long i;
 883         cap_t old_cap = NULL, new_cap = NULL;
 884         cap_flag_value_t fv;
 885         int r;
 886
 887         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 888          * in the effective set (yes, the kernel drops that when
 889          * executing init!), so get it back temporarily so that we can
 890          * call PR_CAPBSET_DROP. */
 891
 892         old_cap = cap_get_proc();
 893         if (!old_cap)
 894                 return -errno;
 895
 896         if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
 897                 r = -errno;
 898                 goto finish;
 899         }
 900
 901         if (fv != CAP_SET) {
 902                 static const cap_value_t v = CAP_SETPCAP;
 903
 904                 new_cap = cap_dup(old_cap);
 905                 if (!new_cap) {
 906                         r = -errno;
 907                         goto finish;
 908                 }
 909
 910                 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
 911                         r = -errno;
 912                         goto finish;
 913                 }
 914
 915                 if (cap_set_proc(new_cap) < 0) {
 916                         r = -errno;
 917                         goto finish;
 918                 }
 919         }
 920
 921         for (i = 0; i <= MAX(63LU, (unsigned long) CAP_LAST_CAP); i++)
 922                 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
 923                         if (prctl(PR_CAPBSET_DROP, i) < 0) {
 924                                 if (errno == EINVAL)
 925                                         break;
 926
 927                                 r = -errno;
 928                                 goto finish;
 929                         }
 930                 }
 931
 932         r = 0;
 933
 934 finish:
 935         if (new_cap)
 936                 cap_free(new_cap);
 937
 938         if (old_cap) {
 939                 cap_set_proc(old_cap);
 940                 cap_free(old_cap);
 941         }
 942
 943         return r;
 944 }
 945
 946 int exec_spawn(ExecCommand *command,
 947                char **argv,
 948                const ExecContext *context,
 949                int fds[], unsigned n_fds,
 950                char **environment,
 951                bool apply_permissions,
 952                bool apply_chroot,
 953                bool apply_tty_stdin,
 954                bool confirm_spawn,
 955                CGroupBonding *cgroup_bondings,
 956                pid_t *ret) {
 957
 958         pid_t pid;
 959         int r;
 960         char *line;
 961         int socket_fd;
 962         char **files_env = NULL;
 963
 964         assert(command);
 965         assert(context);
 966         assert(ret);
 967         assert(fds || n_fds <= 0);
 968
 969         if (context->std_input == EXEC_INPUT_SOCKET ||
 970             context->std_output == EXEC_OUTPUT_SOCKET ||
 971             context->std_error == EXEC_OUTPUT_SOCKET) {
 972
 973                 if (n_fds != 1)
 974                         return -EINVAL;
 975
 976                 socket_fd = fds[0];
 977
 978                 fds = NULL;
 979                 n_fds = 0;
 980         } else
 981                 socket_fd = -1;
 982
 983         if ((r = exec_context_load_environment(context, &files_env)) < 0) {
 984                 log_error("Failed to load environment files: %s", strerror(-r));
 985                 return r;
 986         }
 987
 988         if (!argv)
 989                 argv = command->argv;
 990
 991         if (!(line = exec_command_line(argv))) {
 992                 r = -ENOMEM;
 993                 goto fail_parent;
 994         }
 995
 996         log_debug("About to execute: %s", line);
 997         free(line);
 998
 999         if (cgroup_bondings)
1000                 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
1001                         goto fail_parent;
1002
1003         if ((pid = fork()) < 0) {
1004                 r = -errno;
1005                 goto fail_parent;
1006         }
1007
1008         if (pid == 0) {
1009                 int i;
1010                 sigset_t ss;
1011                 const char *username = NULL, *home = NULL;
1012                 uid_t uid = (uid_t) -1;
1013                 gid_t gid = (gid_t) -1;
1014                 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1015                 unsigned n_env = 0;
1016                 int saved_stdout = -1, saved_stdin = -1;
1017                 bool keep_stdout = false, keep_stdin = false;
1018
1019                 /* child */
1020
1021                 /* This string must fit in 10 chars (i.e. the length
1022                  * of "/sbin/init") */
1023                 rename_process("sd(EXEC)");
1024
1025                 /* We reset exactly these signals, since they are the
1026                  * only ones we set to SIG_IGN in the main daemon. All
1027                  * others we leave untouched because we set them to
1028                  * SIG_DFL or a valid handler initially, both of which
1029                  * will be demoted to SIG_DFL. */
1030                 default_signals(SIGNALS_CRASH_HANDLER,
1031                                 SIGNALS_IGNORE, -1);
1032
1033                 if (sigemptyset(&ss) < 0 ||
1034                     sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1035                         r = EXIT_SIGNAL_MASK;
1036                         goto fail_child;
1037                 }
1038
1039                 /* Close sockets very early to make sure we don't
1040                  * block init reexecution because it cannot bind its
1041                  * sockets */
1042                 if (close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1043                                   socket_fd >= 0 ? 1 : n_fds) < 0) {
1044                         r = EXIT_FDS;
1045                         goto fail_child;
1046                 }
1047
1048                 if (!context->same_pgrp)
1049                         if (setsid() < 0) {
1050                                 r = EXIT_SETSID;
1051                                 goto fail_child;
1052                         }
1053
1054                 if (context->tcpwrap_name) {
1055                         if (socket_fd >= 0)
1056                                 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1057                                         r = EXIT_TCPWRAP;
1058                                         goto fail_child;
1059                                 }
1060
1061                         for (i = 0; i < (int) n_fds; i++) {
1062                                 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1063                                         r = EXIT_TCPWRAP;
1064                                         goto fail_child;
1065                                 }
1066                         }
1067                 }
1068
1069                 exec_context_tty_reset(context);
1070
1071                 /* We skip the confirmation step if we shall not apply the TTY */
1072                 if (confirm_spawn &&
1073                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1074                         char response;
1075
1076                         /* Set up terminal for the question */
1077                         if ((r = setup_confirm_stdio(context,
1078                                                      &saved_stdin, &saved_stdout)))
1079                                 goto fail_child;
1080
1081                         /* Now ask the question. */
1082                         if (!(line = exec_command_line(argv))) {
1083                                 r = EXIT_MEMORY;
1084                                 goto fail_child;
1085                         }
1086
1087                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1088                         free(line);
1089
1090                         if (r < 0 || response == 'n') {
1091                                 r = EXIT_CONFIRM;
1092                                 goto fail_child;
1093                         } else if (response == 's') {
1094                                 r = 0;
1095                                 goto fail_child;
1096                         }
1097
1098                         /* Release terminal for the question */
1099                         if ((r = restore_confirm_stdio(context,
1100                                                        &saved_stdin, &saved_stdout,
1101                                                        &keep_stdin, &keep_stdout)))
1102                                 goto fail_child;
1103                 }
1104
1105                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1106                  * must sure to drop O_NONBLOCK */
1107                 if (socket_fd >= 0)
1108                         fd_nonblock(socket_fd, false);
1109
1110                 if (!keep_stdin)
1111                         if (setup_input(context, socket_fd, apply_tty_stdin) < 0) {
1112                                 r = EXIT_STDIN;
1113                                 goto fail_child;
1114                         }
1115
1116                 if (!keep_stdout)
1117                         if (setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1118                                 r = EXIT_STDOUT;
1119                                 goto fail_child;
1120                         }
1121
1122                 if (setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1123                         r = EXIT_STDERR;
1124                         goto fail_child;
1125                 }
1126
1127                 if (cgroup_bondings)
1128                         if (cgroup_bonding_install_list(cgroup_bondings, 0) < 0) {
1129                                 r = EXIT_CGROUP;
1130                                 goto fail_child;
1131                         }
1132
1133                 if (context->oom_score_adjust_set) {
1134                         char t[16];
1135
1136                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1137                         char_array_0(t);
1138
1139                         if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1140                                 /* Compatibility with Linux <= 2.6.35 */
1141
1142                                 int adj;
1143
1144                                 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1145                                 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1146
1147                                 snprintf(t, sizeof(t), "%i", adj);
1148                                 char_array_0(t);
1149
1150                                 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1151                                     && errno != EACCES) {
1152                                         r = EXIT_OOM_ADJUST;
1153                                         goto fail_child;
1154                                 }
1155                         }
1156                 }
1157
1158                 if (context->nice_set)
1159                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1160                                 r = EXIT_NICE;
1161                                 goto fail_child;
1162                         }
1163
1164                 if (context->cpu_sched_set) {
1165                         struct sched_param param;
1166
1167                         zero(param);
1168                         param.sched_priority = context->cpu_sched_priority;
1169
1170                         if (sched_setscheduler(0, context->cpu_sched_policy |
1171                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
1172                                 r = EXIT_SETSCHEDULER;
1173                                 goto fail_child;
1174                         }
1175                 }
1176
1177                 if (context->cpuset)
1178                         if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1179                                 r = EXIT_CPUAFFINITY;
1180                                 goto fail_child;
1181                         }
1182
1183                 if (context->ioprio_set)
1184                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1185                                 r = EXIT_IOPRIO;
1186                                 goto fail_child;
1187                         }
1188
1189                 if (context->timer_slack_nsec_set)
1190                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1191                                 r = EXIT_TIMERSLACK;
1192                                 goto fail_child;
1193                         }
1194
1195                 if (context->utmp_id)
1196                         utmp_put_init_process(0, context->utmp_id, getpid(), getsid(0), context->tty_path);
1197
1198                 if (context->user) {
1199                         username = context->user;
1200                         if (get_user_creds(&username, &uid, &gid, &home) < 0) {
1201                                 r = EXIT_USER;
1202                                 goto fail_child;
1203                         }
1204
1205                         if (is_terminal_input(context->std_input))
1206                                 if (chown_terminal(STDIN_FILENO, uid) < 0) {
1207                                         r = EXIT_STDIN;
1208                                         goto fail_child;
1209                                 }
1210
1211                         if (cgroup_bondings && context->control_group_modify)
1212                                 if (cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid) < 0 ||
1213                                     cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid) < 0) {
1214                                         r = EXIT_CGROUP;
1215                                         goto fail_child;
1216                                 }
1217                 }
1218
1219                 if (apply_permissions)
1220                         if (enforce_groups(context, username, uid) < 0) {
1221                                 r = EXIT_GROUP;
1222                                 goto fail_child;
1223                         }
1224
1225                 umask(context->umask);
1226
1227 #ifdef HAVE_PAM
1228                 if (context->pam_name && username) {
1229                         if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) != 0) {
1230                                 r = EXIT_PAM;
1231                                 goto fail_child;
1232                         }
1233                 }
1234 #endif
1235
1236                 if (strv_length(context->read_write_dirs) > 0 ||
1237                     strv_length(context->read_only_dirs) > 0 ||
1238                     strv_length(context->inaccessible_dirs) > 0 ||
1239                     context->mount_flags != MS_SHARED ||
1240                     context->private_tmp)
1241                         if ((r = setup_namespace(
1242                                              context->read_write_dirs,
1243                                              context->read_only_dirs,
1244                                              context->inaccessible_dirs,
1245                                              context->private_tmp,
1246                                              context->mount_flags)) < 0)
1247                                 goto fail_child;
1248
1249                 if (apply_chroot) {
1250                         if (context->root_directory)
1251                                 if (chroot(context->root_directory) < 0) {
1252                                         r = EXIT_CHROOT;
1253                                         goto fail_child;
1254                                 }
1255
1256                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1257                                 r = EXIT_CHDIR;
1258                                 goto fail_child;
1259                         }
1260                 } else {
1261
1262                         char *d;
1263
1264                         if (asprintf(&d, "%s/%s",
1265                                      context->root_directory ? context->root_directory : "",
1266                                      context->working_directory ? context->working_directory : "") < 0) {
1267                                 r = EXIT_MEMORY;
1268                                 goto fail_child;
1269                         }
1270
1271                         if (chdir(d) < 0) {
1272                                 free(d);
1273                                 r = EXIT_CHDIR;
1274                                 goto fail_child;
1275                         }
1276
1277                         free(d);
1278                 }
1279
1280                 /* We repeat the fd closing here, to make sure that
1281                  * nothing is leaked from the PAM modules */
1282                 if (close_all_fds(fds, n_fds) < 0 ||
1283                     shift_fds(fds, n_fds) < 0 ||
1284                     flags_fds(fds, n_fds, context->non_blocking) < 0) {
1285                         r = EXIT_FDS;
1286                         goto fail_child;
1287                 }
1288
1289                 if (apply_permissions) {
1290
1291                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1292                                 if (!context->rlimit[i])
1293                                         continue;
1294
1295                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1296                                         r = EXIT_LIMITS;
1297                                         goto fail_child;
1298                                 }
1299                         }
1300
1301                         if (context->capability_bounding_set_drop)
1302                                 if (do_capability_bounding_set_drop(context->capability_bounding_set_drop) < 0) {
1303                                         r = EXIT_CAPABILITIES;
1304                                         goto fail_child;
1305                                 }
1306
1307                         if (context->user)
1308                                 if (enforce_user(context, uid) < 0) {
1309                                         r = EXIT_USER;
1310                                         goto fail_child;
1311                                 }
1312
1313                         /* PR_GET_SECUREBITS is not privileged, while
1314                          * PR_SET_SECUREBITS is. So to suppress
1315                          * potential EPERMs we'll try not to call
1316                          * PR_SET_SECUREBITS unless necessary. */
1317                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1318                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1319                                         r = EXIT_SECUREBITS;
1320                                         goto fail_child;
1321                                 }
1322
1323                         if (context->capabilities)
1324                                 if (cap_set_proc(context->capabilities) < 0) {
1325                                         r = EXIT_CAPABILITIES;
1326                                         goto fail_child;
1327                                 }
1328                 }
1329
1330                 if (!(our_env = new0(char*, 7))) {
1331                         r = EXIT_MEMORY;
1332                         goto fail_child;
1333                 }
1334
1335                 if (n_fds > 0)
1336                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1337                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1338                                 r = EXIT_MEMORY;
1339                                 goto fail_child;
1340                         }
1341
1342                 if (home)
1343                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1344                                 r = EXIT_MEMORY;
1345                                 goto fail_child;
1346                         }
1347
1348                 if (username)
1349                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1350                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1351                                 r = EXIT_MEMORY;
1352                                 goto fail_child;
1353                         }
1354
1355                 if (is_terminal_input(context->std_input) ||
1356                     context->std_output == EXEC_OUTPUT_TTY ||
1357                     context->std_error == EXEC_OUTPUT_TTY)
1358                         if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1359                                 r = EXIT_MEMORY;
1360                                 goto fail_child;
1361                         }
1362
1363                 assert(n_env <= 7);
1364
1365                 if (!(final_env = strv_env_merge(
1366                                       5,
1367                                       environment,
1368                                       our_env,
1369                                       context->environment,
1370                                       files_env,
1371                                       pam_env,
1372                                       NULL))) {
1373                         r = EXIT_MEMORY;
1374                         goto fail_child;
1375                 }
1376
1377                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1378                         r = EXIT_MEMORY;
1379                         goto fail_child;
1380                 }
1381
1382                 final_env = strv_env_clean(final_env);
1383
1384                 execve(command->path, final_argv, final_env);
1385                 r = EXIT_EXEC;
1386
1387         fail_child:
1388                 strv_free(our_env);
1389                 strv_free(final_env);
1390                 strv_free(pam_env);
1391                 strv_free(files_env);
1392                 strv_free(final_argv);
1393
1394                 if (saved_stdin >= 0)
1395                         close_nointr_nofail(saved_stdin);
1396
1397                 if (saved_stdout >= 0)
1398                         close_nointr_nofail(saved_stdout);
1399
1400                 _exit(r);
1401         }
1402
1403         strv_free(files_env);
1404
1405         /* We add the new process to the cgroup both in the child (so
1406          * that we can be sure that no user code is ever executed
1407          * outside of the cgroup) and in the parent (so that we can be
1408          * sure that when we kill the cgroup the process will be
1409          * killed too). */
1410         if (cgroup_bondings)
1411                 cgroup_bonding_install_list(cgroup_bondings, pid);
1412
1413         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1414
1415         exec_status_start(&command->exec_status, pid);
1416
1417         *ret = pid;
1418         return 0;
1419
1420 fail_parent:
1421         strv_free(files_env);
1422
1423         return r;
1424 }
1425
1426 void exec_context_init(ExecContext *c) {
1427         assert(c);
1428
1429         c->umask = 0002;
1430         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1431         c->cpu_sched_policy = SCHED_OTHER;
1432         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1433         c->syslog_level_prefix = true;
1434         c->mount_flags = MS_SHARED;
1435         c->kill_signal = SIGTERM;
1436         c->send_sigkill = true;
1437 }
1438
1439 void exec_context_done(ExecContext *c) {
1440         unsigned l;
1441
1442         assert(c);
1443
1444         strv_free(c->environment);
1445         c->environment = NULL;
1446
1447         strv_free(c->environment_files);
1448         c->environment_files = NULL;
1449
1450         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1451                 free(c->rlimit[l]);
1452                 c->rlimit[l] = NULL;
1453         }
1454
1455         free(c->working_directory);
1456         c->working_directory = NULL;
1457         free(c->root_directory);
1458         c->root_directory = NULL;
1459
1460         free(c->tty_path);
1461         c->tty_path = NULL;
1462
1463         free(c->tcpwrap_name);
1464         c->tcpwrap_name = NULL;
1465
1466         free(c->syslog_identifier);
1467         c->syslog_identifier = NULL;
1468
1469         free(c->user);
1470         c->user = NULL;
1471
1472         free(c->group);
1473         c->group = NULL;
1474
1475         strv_free(c->supplementary_groups);
1476         c->supplementary_groups = NULL;
1477
1478         free(c->pam_name);
1479         c->pam_name = NULL;
1480
1481         if (c->capabilities) {
1482                 cap_free(c->capabilities);
1483                 c->capabilities = NULL;
1484         }
1485
1486         strv_free(c->read_only_dirs);
1487         c->read_only_dirs = NULL;
1488
1489         strv_free(c->read_write_dirs);
1490         c->read_write_dirs = NULL;
1491
1492         strv_free(c->inaccessible_dirs);
1493         c->inaccessible_dirs = NULL;
1494
1495         if (c->cpuset)
1496                 CPU_FREE(c->cpuset);
1497
1498         free(c->utmp_id);
1499         c->utmp_id = NULL;
1500 }
1501
1502 void exec_command_done(ExecCommand *c) {
1503         assert(c);
1504
1505         free(c->path);
1506         c->path = NULL;
1507
1508         strv_free(c->argv);
1509         c->argv = NULL;
1510 }
1511
1512 void exec_command_done_array(ExecCommand *c, unsigned n) {
1513         unsigned i;
1514
1515         for (i = 0; i < n; i++)
1516                 exec_command_done(c+i);
1517 }
1518
1519 void exec_command_free_list(ExecCommand *c) {
1520         ExecCommand *i;
1521
1522         while ((i = c)) {
1523                 LIST_REMOVE(ExecCommand, command, c, i);
1524                 exec_command_done(i);
1525                 free(i);
1526         }
1527 }
1528
1529 void exec_command_free_array(ExecCommand **c, unsigned n) {
1530         unsigned i;
1531
1532         for (i = 0; i < n; i++) {
1533                 exec_command_free_list(c[i]);
1534                 c[i] = NULL;
1535         }
1536 }
1537
1538 int exec_context_load_environment(const ExecContext *c, char ***l) {
1539         char **i, **r = NULL;
1540
1541         assert(c);
1542         assert(l);
1543
1544         STRV_FOREACH(i, c->environment_files) {
1545                 char *fn;
1546                 int k;
1547                 bool ignore = false;
1548                 char **p;
1549
1550                 fn = *i;
1551
1552                 if (fn[0] == '-') {
1553                         ignore = true;
1554                         fn ++;
1555                 }
1556
1557                 if (!path_is_absolute(fn)) {
1558
1559                         if (ignore)
1560                                 continue;
1561
1562                         strv_free(r);
1563                         return -EINVAL;
1564                 }
1565
1566                 if ((k = load_env_file(fn, &p)) < 0) {
1567
1568                         if (ignore)
1569                                 continue;
1570
1571                         strv_free(r);
1572                         return k;
1573                 }
1574
1575                 if (r == NULL)
1576                         r = p;
1577                 else {
1578                         char **m;
1579
1580                         m = strv_env_merge(2, r, p);
1581                         strv_free(r);
1582                         strv_free(p);
1583
1584                         if (!m)
1585                                 return -ENOMEM;
1586
1587                         r = m;
1588                 }
1589         }
1590
1591         *l = r;
1592
1593         return 0;
1594 }
1595
1596 static void strv_fprintf(FILE *f, char **l) {
1597         char **g;
1598
1599         assert(f);
1600
1601         STRV_FOREACH(g, l)
1602                 fprintf(f, " %s", *g);
1603 }
1604
1605 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1606         char ** e;
1607         unsigned i;
1608
1609         assert(c);
1610         assert(f);
1611
1612         if (!prefix)
1613                 prefix = "";
1614
1615         fprintf(f,
1616                 "%sUMask: %04o\n"
1617                 "%sWorkingDirectory: %s\n"
1618                 "%sRootDirectory: %s\n"
1619                 "%sNonBlocking: %s\n"
1620                 "%sPrivateTmp: %s\n"
1621                 "%sControlGroupModify: %s\n",
1622                 prefix, c->umask,
1623                 prefix, c->working_directory ? c->working_directory : "/",
1624                 prefix, c->root_directory ? c->root_directory : "/",
1625                 prefix, yes_no(c->non_blocking),
1626                 prefix, yes_no(c->private_tmp),
1627                 prefix, yes_no(c->control_group_modify));
1628
1629         STRV_FOREACH(e, c->environment)
1630                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1631
1632         STRV_FOREACH(e, c->environment_files)
1633                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1634
1635         if (c->tcpwrap_name)
1636                 fprintf(f,
1637                         "%sTCPWrapName: %s\n",
1638                         prefix, c->tcpwrap_name);
1639
1640         if (c->nice_set)
1641                 fprintf(f,
1642                         "%sNice: %i\n",
1643                         prefix, c->nice);
1644
1645         if (c->oom_score_adjust_set)
1646                 fprintf(f,
1647                         "%sOOMScoreAdjust: %i\n",
1648                         prefix, c->oom_score_adjust);
1649
1650         for (i = 0; i < RLIM_NLIMITS; i++)
1651                 if (c->rlimit[i])
1652                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1653
1654         if (c->ioprio_set)
1655                 fprintf(f,
1656                         "%sIOSchedulingClass: %s\n"
1657                         "%sIOPriority: %i\n",
1658                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1659                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1660
1661         if (c->cpu_sched_set)
1662                 fprintf(f,
1663                         "%sCPUSchedulingPolicy: %s\n"
1664                         "%sCPUSchedulingPriority: %i\n"
1665                         "%sCPUSchedulingResetOnFork: %s\n",
1666                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1667                         prefix, c->cpu_sched_priority,
1668                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1669
1670         if (c->cpuset) {
1671                 fprintf(f, "%sCPUAffinity:", prefix);
1672                 for (i = 0; i < c->cpuset_ncpus; i++)
1673                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1674                                 fprintf(f, " %i", i);
1675                 fputs("\n", f);
1676         }
1677
1678         if (c->timer_slack_nsec_set)
1679                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1680
1681         fprintf(f,
1682                 "%sStandardInput: %s\n"
1683                 "%sStandardOutput: %s\n"
1684                 "%sStandardError: %s\n",
1685                 prefix, exec_input_to_string(c->std_input),
1686                 prefix, exec_output_to_string(c->std_output),
1687                 prefix, exec_output_to_string(c->std_error));
1688
1689         if (c->tty_path)
1690                 fprintf(f,
1691                         "%sTTYPath: %s\n"
1692                         "%sTTYReset: %s\n"
1693                         "%sTTYVHangup: %s\n"
1694                         "%sTTYVTDisallocate: %s\n",
1695                         prefix, c->tty_path,
1696                         prefix, yes_no(c->tty_reset),
1697                         prefix, yes_no(c->tty_vhangup),
1698                         prefix, yes_no(c->tty_vt_disallocate));
1699
1700         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1701             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1702             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1703             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1704                 fprintf(f,
1705                         "%sSyslogFacility: %s\n"
1706                         "%sSyslogLevel: %s\n",
1707                         prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1708                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1709
1710         if (c->capabilities) {
1711                 char *t;
1712                 if ((t = cap_to_text(c->capabilities, NULL))) {
1713                         fprintf(f, "%sCapabilities: %s\n",
1714                                 prefix, t);
1715                         cap_free(t);
1716                 }
1717         }
1718
1719         if (c->secure_bits)
1720                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1721                         prefix,
1722                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1723                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1724                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1725                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1726                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1727                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1728
1729         if (c->capability_bounding_set_drop) {
1730                 unsigned long l;
1731                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1732
1733                 for (l = 0; l <= (unsigned long) CAP_LAST_CAP; l++)
1734                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1735                                 char *t;
1736
1737                                 if ((t = cap_to_name(l))) {
1738                                         fprintf(f, " %s", t);
1739                                         cap_free(t);
1740                                 }
1741                         }
1742
1743                 fputs("\n", f);
1744         }
1745
1746         if (c->user)
1747                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1748         if (c->group)
1749                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1750
1751         if (strv_length(c->supplementary_groups) > 0) {
1752                 fprintf(f, "%sSupplementaryGroups:", prefix);
1753                 strv_fprintf(f, c->supplementary_groups);
1754                 fputs("\n", f);
1755         }
1756
1757         if (c->pam_name)
1758                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1759
1760         if (strv_length(c->read_write_dirs) > 0) {
1761                 fprintf(f, "%sReadWriteDirs:", prefix);
1762                 strv_fprintf(f, c->read_write_dirs);
1763                 fputs("\n", f);
1764         }
1765
1766         if (strv_length(c->read_only_dirs) > 0) {
1767                 fprintf(f, "%sReadOnlyDirs:", prefix);
1768                 strv_fprintf(f, c->read_only_dirs);
1769                 fputs("\n", f);
1770         }
1771
1772         if (strv_length(c->inaccessible_dirs) > 0) {
1773                 fprintf(f, "%sInaccessibleDirs:", prefix);
1774                 strv_fprintf(f, c->inaccessible_dirs);
1775                 fputs("\n", f);
1776         }
1777
1778         fprintf(f,
1779                 "%sKillMode: %s\n"
1780                 "%sKillSignal: SIG%s\n"
1781                 "%sSendSIGKILL: %s\n",
1782                 prefix, kill_mode_to_string(c->kill_mode),
1783                 prefix, signal_to_string(c->kill_signal),
1784                 prefix, yes_no(c->send_sigkill));
1785
1786         if (c->utmp_id)
1787                 fprintf(f,
1788                         "%sUtmpIdentifier: %s\n",
1789                         prefix, c->utmp_id);
1790 }
1791
1792 void exec_status_start(ExecStatus *s, pid_t pid) {
1793         assert(s);
1794
1795         zero(*s);
1796         s->pid = pid;
1797         dual_timestamp_get(&s->start_timestamp);
1798 }
1799
1800 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1801         assert(s);
1802
1803         if ((s->pid && s->pid != pid) ||
1804             !s->start_timestamp.realtime <= 0)
1805                 zero(*s);
1806
1807         s->pid = pid;
1808         dual_timestamp_get(&s->exit_timestamp);
1809
1810         s->code = code;
1811         s->status = status;
1812
1813         if (context) {
1814                 if (context->utmp_id)
1815                         utmp_put_dead_process(context->utmp_id, pid, code, status);
1816
1817                 exec_context_tty_reset(context);
1818         }
1819 }
1820
1821 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1822         char buf[FORMAT_TIMESTAMP_MAX];
1823
1824         assert(s);
1825         assert(f);
1826
1827         if (!prefix)
1828                 prefix = "";
1829
1830         if (s->pid <= 0)
1831                 return;
1832
1833         fprintf(f,
1834                 "%sPID: %lu\n",
1835                 prefix, (unsigned long) s->pid);
1836
1837         if (s->start_timestamp.realtime > 0)
1838                 fprintf(f,
1839                         "%sStart Timestamp: %s\n",
1840                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1841
1842         if (s->exit_timestamp.realtime > 0)
1843                 fprintf(f,
1844                         "%sExit Timestamp: %s\n"
1845                         "%sExit Code: %s\n"
1846                         "%sExit Status: %i\n",
1847                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1848                         prefix, sigchld_code_to_string(s->code),
1849                         prefix, s->status);
1850 }
1851
1852 char *exec_command_line(char **argv) {
1853         size_t k;
1854         char *n, *p, **a;
1855         bool first = true;
1856
1857         assert(argv);
1858
1859         k = 1;
1860         STRV_FOREACH(a, argv)
1861                 k += strlen(*a)+3;
1862
1863         if (!(n = new(char, k)))
1864                 return NULL;
1865
1866         p = n;
1867         STRV_FOREACH(a, argv) {
1868
1869                 if (!first)
1870                         *(p++) = ' ';
1871                 else
1872                         first = false;
1873
1874                 if (strpbrk(*a, WHITESPACE)) {
1875                         *(p++) = '\'';
1876                         p = stpcpy(p, *a);
1877                         *(p++) = '\'';
1878                 } else
1879                         p = stpcpy(p, *a);
1880
1881         }
1882
1883         *p = 0;
1884
1885         /* FIXME: this doesn't really handle arguments that have
1886          * spaces and ticks in them */
1887
1888         return n;
1889 }
1890
1891 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1892         char *p2;
1893         const char *prefix2;
1894
1895         char *cmd;
1896
1897         assert(c);
1898         assert(f);
1899
1900         if (!prefix)
1901                 prefix = "";
1902         p2 = strappend(prefix, "\t");
1903         prefix2 = p2 ? p2 : prefix;
1904
1905         cmd = exec_command_line(c->argv);
1906
1907         fprintf(f,
1908                 "%sCommand Line: %s\n",
1909                 prefix, cmd ? cmd : strerror(ENOMEM));
1910
1911         free(cmd);
1912
1913         exec_status_dump(&c->exec_status, f, prefix2);
1914
1915         free(p2);
1916 }
1917
1918 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1919         assert(f);
1920
1921         if (!prefix)
1922                 prefix = "";
1923
1924         LIST_FOREACH(command, c, c)
1925                 exec_command_dump(c, f, prefix);
1926 }
1927
1928 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1929         ExecCommand *end;
1930
1931         assert(l);
1932         assert(e);
1933
1934         if (*l) {
1935                 /* It's kind of important, that we keep the order here */
1936                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1937                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1938         } else
1939               *l = e;
1940 }
1941
1942 int exec_command_set(ExecCommand *c, const char *path, ...) {
1943         va_list ap;
1944         char **l, *p;
1945
1946         assert(c);
1947         assert(path);
1948
1949         va_start(ap, path);
1950         l = strv_new_ap(path, ap);
1951         va_end(ap);
1952
1953         if (!l)
1954                 return -ENOMEM;
1955
1956         if (!(p = strdup(path))) {
1957                 strv_free(l);
1958                 return -ENOMEM;
1959         }
1960
1961         free(c->path);
1962         c->path = p;
1963
1964         strv_free(c->argv);
1965         c->argv = l;
1966
1967         return 0;
1968 }
1969
1970 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1971         [EXEC_INPUT_NULL] = "null",
1972         [EXEC_INPUT_TTY] = "tty",
1973         [EXEC_INPUT_TTY_FORCE] = "tty-force",
1974         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
1975         [EXEC_INPUT_SOCKET] = "socket"
1976 };
1977
1978 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
1979
1980 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
1981         [EXEC_OUTPUT_INHERIT] = "inherit",
1982         [EXEC_OUTPUT_NULL] = "null",
1983         [EXEC_OUTPUT_TTY] = "tty",
1984         [EXEC_OUTPUT_SYSLOG] = "syslog",
1985         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
1986         [EXEC_OUTPUT_KMSG] = "kmsg",
1987         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
1988         [EXEC_OUTPUT_SOCKET] = "socket"
1989 };
1990
1991 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
1992
1993 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
1994         [KILL_CONTROL_GROUP] = "control-group",
1995         [KILL_PROCESS] = "process",
1996         [KILL_NONE] = "none"
1997 };
1998
1999 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2000
2001 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2002         [KILL_MAIN] = "main",
2003         [KILL_CONTROL] = "control",
2004         [KILL_ALL] = "all"
2005 };
2006
2007 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);