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
  58 /* This assumes there is a 'tty' group */
  59 #define TTY_MODE 0620
  60
  61 static int shift_fds(int fds[], unsigned n_fds) {
  62         int start, restart_from;
  63
  64         if (n_fds <= 0)
  65                 return 0;
  66
  67         /* Modifies the fds array! (sorts it) */
  68
  69         assert(fds);
  70
  71         start = 0;
  72         for (;;) {
  73                 int i;
  74
  75                 restart_from = -1;
  76
  77                 for (i = start; i < (int) n_fds; i++) {
  78                         int nfd;
  79
  80                         /* Already at right index? */
  81                         if (fds[i] == i+3)
  82                                 continue;
  83
  84                         if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
  85                                 return -errno;
  86
  87                         close_nointr_nofail(fds[i]);
  88                         fds[i] = nfd;
  89
  90                         /* Hmm, the fd we wanted isn't free? Then
  91                          * let's remember that and try again from here*/
  92                         if (nfd != i+3 && restart_from < 0)
  93                                 restart_from = i;
  94                 }
  95
  96                 if (restart_from < 0)
  97                         break;
  98
  99                 start = restart_from;
 100         }
 101
 102         return 0;
 103 }
 104
 105 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
 106         unsigned i;
 107         int r;
 108
 109         if (n_fds <= 0)
 110                 return 0;
 111
 112         assert(fds);
 113
 114         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
 115
 116         for (i = 0; i < n_fds; i++) {
 117
 118                 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
 119                         return r;
 120
 121                 /* We unconditionally drop FD_CLOEXEC from the fds,
 122                  * since after all we want to pass these fds to our
 123                  * children */
 124
 125                 if ((r = fd_cloexec(fds[i], false)) < 0)
 126                         return r;
 127         }
 128
 129         return 0;
 130 }
 131
 132 static const char *tty_path(const ExecContext *context) {
 133         assert(context);
 134
 135         if (context->tty_path)
 136                 return context->tty_path;
 137
 138         return "/dev/console";
 139 }
 140
 141 static int open_null_as(int flags, int nfd) {
 142         int fd, r;
 143
 144         assert(nfd >= 0);
 145
 146         if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
 147                 return -errno;
 148
 149         if (fd != nfd) {
 150                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 151                 close_nointr_nofail(fd);
 152         } else
 153                 r = nfd;
 154
 155         return r;
 156 }
 157
 158 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
 159         int fd, r;
 160         union {
 161                 struct sockaddr sa;
 162                 struct sockaddr_un un;
 163         } sa;
 164
 165         assert(context);
 166         assert(output < _EXEC_OUTPUT_MAX);
 167         assert(ident);
 168         assert(nfd >= 0);
 169
 170         if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
 171                 return -errno;
 172
 173         zero(sa);
 174         sa.sa.sa_family = AF_UNIX;
 175         strncpy(sa.un.sun_path+1, LOGGER_SOCKET, sizeof(sa.un.sun_path)-1);
 176
 177         if (connect(fd, &sa.sa, sizeof(sa_family_t) + 1 + sizeof(LOGGER_SOCKET) - 1) < 0) {
 178                 close_nointr_nofail(fd);
 179                 return -errno;
 180         }
 181
 182         if (shutdown(fd, SHUT_RD) < 0) {
 183                 close_nointr_nofail(fd);
 184                 return -errno;
 185         }
 186
 187         /* We speak a very simple protocol between log server
 188          * and client: one line for the log destination (kmsg
 189          * or syslog), followed by the priority field,
 190          * followed by the process name. Since we replaced
 191          * stdin/stderr we simple use stdio to write to
 192          * it. Note that we use stderr, to minimize buffer
 193          * flushing issues. */
 194
 195         dprintf(fd,
 196                 "%s\n"
 197                 "%i\n"
 198                 "%s\n"
 199                 "%i\n",
 200                 output == EXEC_OUTPUT_KMSG ? "kmsg" : "syslog",
 201                 context->syslog_priority,
 202                 context->syslog_identifier ? context->syslog_identifier : ident,
 203                 context->syslog_level_prefix);
 204
 205         if (fd != nfd) {
 206                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 207                 close_nointr_nofail(fd);
 208         } else
 209                 r = nfd;
 210
 211         return r;
 212 }
 213 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
 214         int fd, r;
 215
 216         assert(path);
 217         assert(nfd >= 0);
 218
 219         if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
 220                 return fd;
 221
 222         if (fd != nfd) {
 223                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 224                 close_nointr_nofail(fd);
 225         } else
 226                 r = nfd;
 227
 228         return r;
 229 }
 230
 231 static bool is_terminal_input(ExecInput i) {
 232         return
 233                 i == EXEC_INPUT_TTY ||
 234                 i == EXEC_INPUT_TTY_FORCE ||
 235                 i == EXEC_INPUT_TTY_FAIL;
 236 }
 237
 238 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
 239
 240         if (is_terminal_input(std_input) && !apply_tty_stdin)
 241                 return EXEC_INPUT_NULL;
 242
 243         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 244                 return EXEC_INPUT_NULL;
 245
 246         return std_input;
 247 }
 248
 249 static int fixup_output(ExecOutput std_output, int socket_fd) {
 250
 251         if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 252                 return EXEC_OUTPUT_INHERIT;
 253
 254         return std_output;
 255 }
 256
 257 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
 258         ExecInput i;
 259
 260         assert(context);
 261
 262         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 263
 264         switch (i) {
 265
 266         case EXEC_INPUT_NULL:
 267                 return open_null_as(O_RDONLY, STDIN_FILENO);
 268
 269         case EXEC_INPUT_TTY:
 270         case EXEC_INPUT_TTY_FORCE:
 271         case EXEC_INPUT_TTY_FAIL: {
 272                 int fd, r;
 273
 274                 if ((fd = acquire_terminal(
 275                                      tty_path(context),
 276                                      i == EXEC_INPUT_TTY_FAIL,
 277                                      i == EXEC_INPUT_TTY_FORCE,
 278                                      false)) < 0)
 279                         return fd;
 280
 281                 if (fd != STDIN_FILENO) {
 282                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 283                         close_nointr_nofail(fd);
 284                 } else
 285                         r = STDIN_FILENO;
 286
 287                 return r;
 288         }
 289
 290         case EXEC_INPUT_SOCKET:
 291                 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 292
 293         default:
 294                 assert_not_reached("Unknown input type");
 295         }
 296 }
 297
 298 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 299         ExecOutput o;
 300         ExecInput i;
 301
 302         assert(context);
 303         assert(ident);
 304
 305         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 306         o = fixup_output(context->std_output, socket_fd);
 307
 308         /* This expects the input is already set up */
 309
 310         switch (o) {
 311
 312         case EXEC_OUTPUT_INHERIT:
 313
 314                 /* If input got downgraded, inherit the original value */
 315                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 316                         return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 317
 318                 /* If the input is connected to anything that's not a /dev/null, inherit that... */
 319                 if (i != EXEC_INPUT_NULL)
 320                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 321
 322                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 323                 if (getppid() != 1)
 324                         return STDOUT_FILENO;
 325
 326                 /* We need to open /dev/null here anew, to get the
 327                  * right access mode. So we fall through */
 328
 329         case EXEC_OUTPUT_NULL:
 330                 return open_null_as(O_WRONLY, STDOUT_FILENO);
 331
 332         case EXEC_OUTPUT_TTY:
 333                 if (is_terminal_input(i))
 334                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 335
 336                 /* We don't reset the terminal if this is just about output */
 337                 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 338
 339         case EXEC_OUTPUT_SYSLOG:
 340         case EXEC_OUTPUT_KMSG:
 341                 return connect_logger_as(context, o, ident, STDOUT_FILENO);
 342
 343         case EXEC_OUTPUT_SOCKET:
 344                 assert(socket_fd >= 0);
 345                 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 346
 347         default:
 348                 assert_not_reached("Unknown output type");
 349         }
 350 }
 351
 352 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 353         ExecOutput o, e;
 354         ExecInput i;
 355
 356         assert(context);
 357         assert(ident);
 358
 359         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 360         o = fixup_output(context->std_output, socket_fd);
 361         e = fixup_output(context->std_error, socket_fd);
 362
 363         /* This expects the input and output are already set up */
 364
 365         /* Don't change the stderr file descriptor if we inherit all
 366          * the way and are not on a tty */
 367         if (e == EXEC_OUTPUT_INHERIT &&
 368             o == EXEC_OUTPUT_INHERIT &&
 369             i == EXEC_INPUT_NULL &&
 370             !is_terminal_input(context->std_input) &&
 371             getppid () != 1)
 372                 return STDERR_FILENO;
 373
 374         /* Duplicate from stdout if possible */
 375         if (e == o || e == EXEC_OUTPUT_INHERIT)
 376                 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 377
 378         switch (e) {
 379
 380         case EXEC_OUTPUT_NULL:
 381                 return open_null_as(O_WRONLY, STDERR_FILENO);
 382
 383         case EXEC_OUTPUT_TTY:
 384                 if (is_terminal_input(i))
 385                         return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 386
 387                 /* We don't reset the terminal if this is just about output */
 388                 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
 389
 390         case EXEC_OUTPUT_SYSLOG:
 391         case EXEC_OUTPUT_KMSG:
 392                 return connect_logger_as(context, e, ident, STDERR_FILENO);
 393
 394         case EXEC_OUTPUT_SOCKET:
 395                 assert(socket_fd >= 0);
 396                 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 397
 398         default:
 399                 assert_not_reached("Unknown error type");
 400         }
 401 }
 402
 403 static int chown_terminal(int fd, uid_t uid) {
 404         struct stat st;
 405
 406         assert(fd >= 0);
 407
 408         /* This might fail. What matters are the results. */
 409         (void) fchown(fd, uid, -1);
 410         (void) fchmod(fd, TTY_MODE);
 411
 412         if (fstat(fd, &st) < 0)
 413                 return -errno;
 414
 415         if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
 416                 return -EPERM;
 417
 418         return 0;
 419 }
 420
 421 static int setup_confirm_stdio(const ExecContext *context,
 422                                int *_saved_stdin,
 423                                int *_saved_stdout) {
 424         int fd = -1, saved_stdin, saved_stdout = -1, r;
 425
 426         assert(context);
 427         assert(_saved_stdin);
 428         assert(_saved_stdout);
 429
 430         /* This returns positive EXIT_xxx return values instead of
 431          * negative errno style values! */
 432
 433         if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
 434                 return EXIT_STDIN;
 435
 436         if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
 437                 r = EXIT_STDOUT;
 438                 goto fail;
 439         }
 440
 441         if ((fd = acquire_terminal(
 442                              tty_path(context),
 443                              context->std_input == EXEC_INPUT_TTY_FAIL,
 444                              context->std_input == EXEC_INPUT_TTY_FORCE,
 445                              false)) < 0) {
 446                 r = EXIT_STDIN;
 447                 goto fail;
 448         }
 449
 450         if (chown_terminal(fd, getuid()) < 0) {
 451                 r = EXIT_STDIN;
 452                 goto fail;
 453         }
 454
 455         if (dup2(fd, STDIN_FILENO) < 0) {
 456                 r = EXIT_STDIN;
 457                 goto fail;
 458         }
 459
 460         if (dup2(fd, STDOUT_FILENO) < 0) {
 461                 r = EXIT_STDOUT;
 462                 goto fail;
 463         }
 464
 465         if (fd >= 2)
 466                 close_nointr_nofail(fd);
 467
 468         *_saved_stdin = saved_stdin;
 469         *_saved_stdout = saved_stdout;
 470
 471         return 0;
 472
 473 fail:
 474         if (saved_stdout >= 0)
 475                 close_nointr_nofail(saved_stdout);
 476
 477         if (saved_stdin >= 0)
 478                 close_nointr_nofail(saved_stdin);
 479
 480         if (fd >= 0)
 481                 close_nointr_nofail(fd);
 482
 483         return r;
 484 }
 485
 486 static int restore_confirm_stdio(const ExecContext *context,
 487                                  int *saved_stdin,
 488                                  int *saved_stdout,
 489                                  bool *keep_stdin,
 490                                  bool *keep_stdout) {
 491
 492         assert(context);
 493         assert(saved_stdin);
 494         assert(*saved_stdin >= 0);
 495         assert(saved_stdout);
 496         assert(*saved_stdout >= 0);
 497
 498         /* This returns positive EXIT_xxx return values instead of
 499          * negative errno style values! */
 500
 501         if (is_terminal_input(context->std_input)) {
 502
 503                 /* The service wants terminal input. */
 504
 505                 *keep_stdin = true;
 506                 *keep_stdout =
 507                         context->std_output == EXEC_OUTPUT_INHERIT ||
 508                         context->std_output == EXEC_OUTPUT_TTY;
 509
 510         } else {
 511                 /* If the service doesn't want a controlling terminal,
 512                  * then we need to get rid entirely of what we have
 513                  * already. */
 514
 515                 if (release_terminal() < 0)
 516                         return EXIT_STDIN;
 517
 518                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 519                         return EXIT_STDIN;
 520
 521                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 522                         return EXIT_STDOUT;
 523
 524                 *keep_stdout = *keep_stdin = false;
 525         }
 526
 527         return 0;
 528 }
 529
 530 static int get_group_creds(const char *groupname, gid_t *gid) {
 531         struct group *g;
 532         unsigned long lu;
 533
 534         assert(groupname);
 535         assert(gid);
 536
 537         /* We enforce some special rules for gid=0: in order to avoid
 538          * NSS lookups for root we hardcode its data. */
 539
 540         if (streq(groupname, "root") || streq(groupname, "0")) {
 541                 *gid = 0;
 542                 return 0;
 543         }
 544
 545         if (safe_atolu(groupname, &lu) >= 0) {
 546                 errno = 0;
 547                 g = getgrgid((gid_t) lu);
 548         } else {
 549                 errno = 0;
 550                 g = getgrnam(groupname);
 551         }
 552
 553         if (!g)
 554                 return errno != 0 ? -errno : -ESRCH;
 555
 556         *gid = g->gr_gid;
 557         return 0;
 558 }
 559
 560 static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
 561         struct passwd *p;
 562         unsigned long lu;
 563
 564         assert(username);
 565         assert(*username);
 566         assert(uid);
 567         assert(gid);
 568         assert(home);
 569
 570         /* We enforce some special rules for uid=0: in order to avoid
 571          * NSS lookups for root we hardcode its data. */
 572
 573         if (streq(*username, "root") || streq(*username, "0")) {
 574                 *username = "root";
 575                 *uid = 0;
 576                 *gid = 0;
 577                 *home = "/root";
 578                 return 0;
 579         }
 580
 581         if (safe_atolu(*username, &lu) >= 0) {
 582                 errno = 0;
 583                 p = getpwuid((uid_t) lu);
 584
 585                 /* If there are multiple users with the same id, make
 586                  * sure to leave $USER to the configured value instead
 587                  * of the first occurence in the database. However if
 588                  * the uid was configured by a numeric uid, then let's
 589                  * pick the real username from /etc/passwd. */
 590                 if (*username && p)
 591                         *username = p->pw_name;
 592         } else {
 593                 errno = 0;
 594                 p = getpwnam(*username);
 595         }
 596
 597         if (!p)
 598                 return errno != 0 ? -errno : -ESRCH;
 599
 600         *uid = p->pw_uid;
 601         *gid = p->pw_gid;
 602         *home = p->pw_dir;
 603         return 0;
 604 }
 605
 606 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
 607         bool keep_groups = false;
 608         int r;
 609
 610         assert(context);
 611
 612         /* Lookup and ser GID and supplementary group list. Here too
 613          * we avoid NSS lookups for gid=0. */
 614
 615         if (context->group || username) {
 616
 617                 if (context->group)
 618                         if ((r = get_group_creds(context->group, &gid)) < 0)
 619                                 return r;
 620
 621                 /* First step, initialize groups from /etc/groups */
 622                 if (username && gid != 0) {
 623                         if (initgroups(username, gid) < 0)
 624                                 return -errno;
 625
 626                         keep_groups = true;
 627                 }
 628
 629                 /* Second step, set our gids */
 630                 if (setresgid(gid, gid, gid) < 0)
 631                         return -errno;
 632         }
 633
 634         if (context->supplementary_groups) {
 635                 int ngroups_max, k;
 636                 gid_t *gids;
 637                 char **i;
 638
 639                 /* Final step, initialize any manually set supplementary groups */
 640                 ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
 641
 642                 if (!(gids = new(gid_t, ngroups_max)))
 643                         return -ENOMEM;
 644
 645                 if (keep_groups) {
 646                         if ((k = getgroups(ngroups_max, gids)) < 0) {
 647                                 free(gids);
 648                                 return -errno;
 649                         }
 650                 } else
 651                         k = 0;
 652
 653                 STRV_FOREACH(i, context->supplementary_groups) {
 654
 655                         if (k >= ngroups_max) {
 656                                 free(gids);
 657                                 return -E2BIG;
 658                         }
 659
 660                         if ((r = get_group_creds(*i, gids+k)) < 0) {
 661                                 free(gids);
 662                                 return r;
 663                         }
 664
 665                         k++;
 666                 }
 667
 668                 if (setgroups(k, gids) < 0) {
 669                         free(gids);
 670                         return -errno;
 671                 }
 672
 673                 free(gids);
 674         }
 675
 676         return 0;
 677 }
 678
 679 static int enforce_user(const ExecContext *context, uid_t uid) {
 680         int r;
 681         assert(context);
 682
 683         /* Sets (but doesn't lookup) the uid and make sure we keep the
 684          * capabilities while doing so. */
 685
 686         if (context->capabilities) {
 687                 cap_t d;
 688                 static const cap_value_t bits[] = {
 689                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
 690                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
 691                 };
 692
 693                 /* First step: If we need to keep capabilities but
 694                  * drop privileges we need to make sure we keep our
 695                  * caps, whiel we drop priviliges. */
 696                 if (uid != 0) {
 697                         int sb = context->secure_bits|SECURE_KEEP_CAPS;
 698
 699                         if (prctl(PR_GET_SECUREBITS) != sb)
 700                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 701                                         return -errno;
 702                 }
 703
 704                 /* Second step: set the capabilites. This will reduce
 705                  * the capabilities to the minimum we need. */
 706
 707                 if (!(d = cap_dup(context->capabilities)))
 708                         return -errno;
 709
 710                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
 711                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
 712                         r = -errno;
 713                         cap_free(d);
 714                         return r;
 715                 }
 716
 717                 if (cap_set_proc(d) < 0) {
 718                         r = -errno;
 719                         cap_free(d);
 720                         return r;
 721                 }
 722
 723                 cap_free(d);
 724         }
 725
 726         /* Third step: actually set the uids */
 727         if (setresuid(uid, uid, uid) < 0)
 728                 return -errno;
 729
 730         /* At this point we should have all necessary capabilities but
 731            are otherwise a normal user. However, the caps might got
 732            corrupted due to the setresuid() so we need clean them up
 733            later. This is done outside of this call. */
 734
 735         return 0;
 736 }
 737
 738 #ifdef HAVE_PAM
 739
 740 static int null_conv(
 741                 int num_msg,
 742                 const struct pam_message **msg,
 743                 struct pam_response **resp,
 744                 void *appdata_ptr) {
 745
 746         /* We don't support conversations */
 747
 748         return PAM_CONV_ERR;
 749 }
 750
 751 static int setup_pam(
 752                 const char *name,
 753                 const char *user,
 754                 const char *tty,
 755                 char ***pam_env,
 756                 int fds[], unsigned n_fds) {
 757
 758         static const struct pam_conv conv = {
 759                 .conv = null_conv,
 760                 .appdata_ptr = NULL
 761         };
 762
 763         pam_handle_t *handle = NULL;
 764         sigset_t ss, old_ss;
 765         int pam_code = PAM_SUCCESS;
 766         char **e = NULL;
 767         bool close_session = false;
 768         pid_t pam_pid = 0, parent_pid;
 769
 770         assert(name);
 771         assert(user);
 772         assert(pam_env);
 773
 774         /* We set up PAM in the parent process, then fork. The child
 775          * will then stay around untill killed via PR_GET_PDEATHSIG or
 776          * systemd via the cgroup logic. It will then remove the PAM
 777          * session again. The parent process will exec() the actual
 778          * daemon. We do things this way to ensure that the main PID
 779          * of the daemon is the one we initially fork()ed. */
 780
 781         if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
 782                 handle = NULL;
 783                 goto fail;
 784         }
 785
 786         if (tty)
 787                 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
 788                         goto fail;
 789
 790         if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
 791                 goto fail;
 792
 793         if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
 794                 goto fail;
 795
 796         close_session = true;
 797
 798         if ((pam_code = pam_setcred(handle, PAM_ESTABLISH_CRED | PAM_SILENT)) != PAM_SUCCESS)
 799                 goto fail;
 800
 801         if ((!(e = pam_getenvlist(handle)))) {
 802                 pam_code = PAM_BUF_ERR;
 803                 goto fail;
 804         }
 805
 806         /* Block SIGTERM, so that we know that it won't get lost in
 807          * the child */
 808         if (sigemptyset(&ss) < 0 ||
 809             sigaddset(&ss, SIGTERM) < 0 ||
 810             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 811                 goto fail;
 812
 813         parent_pid = getpid();
 814
 815         if ((pam_pid = fork()) < 0)
 816                 goto fail;
 817
 818         if (pam_pid == 0) {
 819                 int sig;
 820                 int r = EXIT_PAM;
 821
 822                 /* The child's job is to reset the PAM session on
 823                  * termination */
 824
 825                 /* This string must fit in 10 chars (i.e. the length
 826                  * of "/sbin/init") */
 827                 rename_process("sd:pam");
 828
 829                 /* Make sure we don't keep open the passed fds in this
 830                 child. We assume that otherwise only those fds are
 831                 open here that have been opened by PAM. */
 832                 close_many(fds, n_fds);
 833
 834                 /* Wait until our parent died. This will most likely
 835                  * not work since the kernel does not allow
 836                  * unpriviliged paretns kill their priviliged children
 837                  * this way. We rely on the control groups kill logic
 838                  * to do the rest for us. */
 839                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 840                         goto child_finish;
 841
 842                 /* Check if our parent process might already have
 843                  * died? */
 844                 if (getppid() == parent_pid) {
 845                         if (sigwait(&ss, &sig) < 0)
 846                                 goto child_finish;
 847
 848                         assert(sig == SIGTERM);
 849                 }
 850
 851                 /* Only if our parent died we'll end the session */
 852                 if (getppid() != parent_pid)
 853                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
 854                                 goto child_finish;
 855
 856                 r = 0;
 857
 858         child_finish:
 859                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 860                 _exit(r);
 861         }
 862
 863         /* If the child was forked off successfully it will do all the
 864          * cleanups, so forget about the handle here. */
 865         handle = NULL;
 866
 867         /* Unblock SIGSUR1 again in the parent */
 868         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 869                 goto fail;
 870
 871         /* We close the log explicitly here, since the PAM modules
 872          * might have opened it, but we don't want this fd around. */
 873         closelog();
 874
 875         return 0;
 876
 877 fail:
 878         if (handle) {
 879                 if (close_session)
 880                         pam_code = pam_close_session(handle, PAM_DATA_SILENT);
 881
 882                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 883         }
 884
 885         strv_free(e);
 886
 887         closelog();
 888
 889         if (pam_pid > 1)
 890                 kill(pam_pid, SIGTERM);
 891
 892         return EXIT_PAM;
 893 }
 894 #endif
 895
 896 int exec_spawn(ExecCommand *command,
 897                char **argv,
 898                const ExecContext *context,
 899                int fds[], unsigned n_fds,
 900                char **environment,
 901                bool apply_permissions,
 902                bool apply_chroot,
 903                bool apply_tty_stdin,
 904                bool confirm_spawn,
 905                CGroupBonding *cgroup_bondings,
 906                pid_t *ret) {
 907
 908         pid_t pid;
 909         int r;
 910         char *line;
 911         int socket_fd;
 912
 913         assert(command);
 914         assert(context);
 915         assert(ret);
 916         assert(fds || n_fds <= 0);
 917
 918         if (context->std_input == EXEC_INPUT_SOCKET ||
 919             context->std_output == EXEC_OUTPUT_SOCKET ||
 920             context->std_error == EXEC_OUTPUT_SOCKET) {
 921
 922                 if (n_fds != 1)
 923                         return -EINVAL;
 924
 925                 socket_fd = fds[0];
 926
 927                 fds = NULL;
 928                 n_fds = 0;
 929         } else
 930                 socket_fd = -1;
 931
 932         if (!argv)
 933                 argv = command->argv;
 934
 935         if (!(line = exec_command_line(argv)))
 936                 return -ENOMEM;
 937
 938         log_debug("About to execute: %s", line);
 939         free(line);
 940
 941         if (cgroup_bondings)
 942                 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
 943                         return r;
 944
 945         if ((pid = fork()) < 0)
 946                 return -errno;
 947
 948         if (pid == 0) {
 949                 int i;
 950                 sigset_t ss;
 951                 const char *username = NULL, *home = NULL;
 952                 uid_t uid = (uid_t) -1;
 953                 gid_t gid = (gid_t) -1;
 954                 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
 955                 unsigned n_env = 0;
 956                 int saved_stdout = -1, saved_stdin = -1;
 957                 bool keep_stdout = false, keep_stdin = false;
 958
 959                 /* child */
 960
 961                 /* This string must fit in 10 chars (i.e. the length
 962                  * of "/sbin/init") */
 963                 rename_process("sd:exec");
 964
 965                 /* We reset exactly these signals, since they are the
 966                  * only ones we set to SIG_IGN in the main daemon. All
 967                  * others we leave untouched because we set them to
 968                  * SIG_DFL or a valid handler initially, both of which
 969                  * will be demoted to SIG_DFL. */
 970                 default_signals(SIGNALS_CRASH_HANDLER,
 971                                 SIGNALS_IGNORE, -1);
 972
 973                 if (sigemptyset(&ss) < 0 ||
 974                     sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
 975                         r = EXIT_SIGNAL_MASK;
 976                         goto fail;
 977                 }
 978
 979                 /* Close sockets very early to make sure we don't
 980                  * block init reexecution because it cannot bind its
 981                  * sockets */
 982                 if (close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
 983                                   socket_fd >= 0 ? 1 : n_fds) < 0) {
 984                         r = EXIT_FDS;
 985                         goto fail;
 986                 }
 987
 988                 if (!context->same_pgrp)
 989                         if (setsid() < 0) {
 990                                 r = EXIT_SETSID;
 991                                 goto fail;
 992                         }
 993
 994                 if (context->tcpwrap_name) {
 995                         if (socket_fd >= 0)
 996                                 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
 997                                         r = EXIT_TCPWRAP;
 998                                         goto fail;
 999                                 }
1000
1001                         for (i = 0; i < (int) n_fds; i++) {
1002                                 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1003                                         r = EXIT_TCPWRAP;
1004                                         goto fail;
1005                                 }
1006                         }
1007                 }
1008
1009                 /* We skip the confirmation step if we shall not apply the TTY */
1010                 if (confirm_spawn &&
1011                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1012                         char response;
1013
1014                         /* Set up terminal for the question */
1015                         if ((r = setup_confirm_stdio(context,
1016                                                      &saved_stdin, &saved_stdout)))
1017                                 goto fail;
1018
1019                         /* Now ask the question. */
1020                         if (!(line = exec_command_line(argv))) {
1021                                 r = EXIT_MEMORY;
1022                                 goto fail;
1023                         }
1024
1025                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1026                         free(line);
1027
1028                         if (r < 0 || response == 'n') {
1029                                 r = EXIT_CONFIRM;
1030                                 goto fail;
1031                         } else if (response == 's') {
1032                                 r = 0;
1033                                 goto fail;
1034                         }
1035
1036                         /* Release terminal for the question */
1037                         if ((r = restore_confirm_stdio(context,
1038                                                        &saved_stdin, &saved_stdout,
1039                                                        &keep_stdin, &keep_stdout)))
1040                                 goto fail;
1041                 }
1042
1043                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1044                  * must sure to drop O_NONBLOCK */
1045                 if (socket_fd >= 0)
1046                         fd_nonblock(socket_fd, false);
1047
1048                 if (!keep_stdin)
1049                         if (setup_input(context, socket_fd, apply_tty_stdin) < 0) {
1050                                 r = EXIT_STDIN;
1051                                 goto fail;
1052                         }
1053
1054                 if (!keep_stdout)
1055                         if (setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1056                                 r = EXIT_STDOUT;
1057                                 goto fail;
1058                         }
1059
1060                 if (setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1061                         r = EXIT_STDERR;
1062                         goto fail;
1063                 }
1064
1065                 if (cgroup_bondings)
1066                         if (cgroup_bonding_install_list(cgroup_bondings, 0) < 0) {
1067                                 r = EXIT_CGROUP;
1068                                 goto fail;
1069                         }
1070
1071                 if (context->oom_score_adjust_set) {
1072                         char t[16];
1073
1074                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1075                         char_array_0(t);
1076
1077                         if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1078                                 /* Compatibility with Linux <= 2.6.35 */
1079
1080                                 int adj;
1081
1082                                 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1083                                 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1084
1085                                 snprintf(t, sizeof(t), "%i", adj);
1086                                 char_array_0(t);
1087
1088                                 if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
1089                                         r = EXIT_OOM_ADJUST;
1090                                         goto fail;
1091                                 }
1092                         }
1093                 }
1094
1095                 if (context->nice_set)
1096                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1097                                 r = EXIT_NICE;
1098                                 goto fail;
1099                         }
1100
1101                 if (context->cpu_sched_set) {
1102                         struct sched_param param;
1103
1104                         zero(param);
1105                         param.sched_priority = context->cpu_sched_priority;
1106
1107                         if (sched_setscheduler(0, context->cpu_sched_policy |
1108                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
1109                                 r = EXIT_SETSCHEDULER;
1110                                 goto fail;
1111                         }
1112                 }
1113
1114                 if (context->cpuset)
1115                         if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1116                                 r = EXIT_CPUAFFINITY;
1117                                 goto fail;
1118                         }
1119
1120                 if (context->ioprio_set)
1121                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1122                                 r = EXIT_IOPRIO;
1123                                 goto fail;
1124                         }
1125
1126                 if (context->timer_slack_nsec_set)
1127                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1128                                 r = EXIT_TIMERSLACK;
1129                                 goto fail;
1130                         }
1131
1132                 if (context->user) {
1133                         username = context->user;
1134                         if (get_user_creds(&username, &uid, &gid, &home) < 0) {
1135                                 r = EXIT_USER;
1136                                 goto fail;
1137                         }
1138
1139                         if (is_terminal_input(context->std_input))
1140                                 if (chown_terminal(STDIN_FILENO, uid) < 0) {
1141                                         r = EXIT_STDIN;
1142                                         goto fail;
1143                                 }
1144                 }
1145
1146 #ifdef HAVE_PAM
1147                 if (context->pam_name && username) {
1148                         if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) < 0) {
1149                                 r = EXIT_PAM;
1150                                 goto fail;
1151                         }
1152                 }
1153 #endif
1154
1155                 if (apply_permissions)
1156                         if (enforce_groups(context, username, uid) < 0) {
1157                                 r = EXIT_GROUP;
1158                                 goto fail;
1159                         }
1160
1161                 umask(context->umask);
1162
1163                 if (strv_length(context->read_write_dirs) > 0 ||
1164                     strv_length(context->read_only_dirs) > 0 ||
1165                     strv_length(context->inaccessible_dirs) > 0 ||
1166                     context->mount_flags != MS_SHARED ||
1167                     context->private_tmp)
1168                         if ((r = setup_namespace(
1169                                              context->read_write_dirs,
1170                                              context->read_only_dirs,
1171                                              context->inaccessible_dirs,
1172                                              context->private_tmp,
1173                                              context->mount_flags)) < 0)
1174                                 goto fail;
1175
1176                 if (apply_chroot) {
1177                         if (context->root_directory)
1178                                 if (chroot(context->root_directory) < 0) {
1179                                         r = EXIT_CHROOT;
1180                                         goto fail;
1181                                 }
1182
1183                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1184                                 r = EXIT_CHDIR;
1185                                 goto fail;
1186                         }
1187                 } else {
1188
1189                         char *d;
1190
1191                         if (asprintf(&d, "%s/%s",
1192                                      context->root_directory ? context->root_directory : "",
1193                                      context->working_directory ? context->working_directory : "") < 0) {
1194                                 r = EXIT_MEMORY;
1195                                 goto fail;
1196                         }
1197
1198                         if (chdir(d) < 0) {
1199                                 free(d);
1200                                 r = EXIT_CHDIR;
1201                                 goto fail;
1202                         }
1203
1204                         free(d);
1205                 }
1206
1207                 /* We repeat the fd closing here, to make sure that
1208                  * nothing is leaked from the PAM modules */
1209                 if (close_all_fds(fds, n_fds) < 0 ||
1210                     shift_fds(fds, n_fds) < 0 ||
1211                     flags_fds(fds, n_fds, context->non_blocking) < 0) {
1212                         r = EXIT_FDS;
1213                         goto fail;
1214                 }
1215
1216                 if (apply_permissions) {
1217
1218                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1219                                 if (!context->rlimit[i])
1220                                         continue;
1221
1222                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1223                                         r = EXIT_LIMITS;
1224                                         goto fail;
1225                                 }
1226                         }
1227
1228                         if (context->user)
1229                                 if (enforce_user(context, uid) < 0) {
1230                                         r = EXIT_USER;
1231                                         goto fail;
1232                                 }
1233
1234                         /* PR_GET_SECUREBITS is not priviliged, while
1235                          * PR_SET_SECUREBITS is. So to suppress
1236                          * potential EPERMs we'll try not to call
1237                          * PR_SET_SECUREBITS unless necessary. */
1238                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1239                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1240                                         r = EXIT_SECUREBITS;
1241                                         goto fail;
1242                                 }
1243
1244                         if (context->capabilities)
1245                                 if (cap_set_proc(context->capabilities) < 0) {
1246                                         r = EXIT_CAPABILITIES;
1247                                         goto fail;
1248                                 }
1249                 }
1250
1251                 if (!(our_env = new0(char*, 6))) {
1252                         r = EXIT_MEMORY;
1253                         goto fail;
1254                 }
1255
1256                 if (n_fds > 0)
1257                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1258                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1259                                 r = EXIT_MEMORY;
1260                                 goto fail;
1261                         }
1262
1263                 if (home)
1264                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1265                                 r = EXIT_MEMORY;
1266                                 goto fail;
1267                         }
1268
1269                 if (username)
1270                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1271                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1272                                 r = EXIT_MEMORY;
1273                                 goto fail;
1274                         }
1275
1276                 assert(n_env <= 6);
1277
1278                 if (!(final_env = strv_env_merge(
1279                                       4,
1280                                       environment,
1281                                       our_env,
1282                                       context->environment,
1283                                       pam_env,
1284                                       NULL))) {
1285                         r = EXIT_MEMORY;
1286                         goto fail;
1287                 }
1288
1289                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1290                         r = EXIT_MEMORY;
1291                         goto fail;
1292                 }
1293
1294                 execve(command->path, final_argv, final_env);
1295                 r = EXIT_EXEC;
1296
1297         fail:
1298                 strv_free(our_env);
1299                 strv_free(final_env);
1300                 strv_free(pam_env);
1301                 strv_free(final_argv);
1302
1303                 if (saved_stdin >= 0)
1304                         close_nointr_nofail(saved_stdin);
1305
1306                 if (saved_stdout >= 0)
1307                         close_nointr_nofail(saved_stdout);
1308
1309                 _exit(r);
1310         }
1311
1312         /* We add the new process to the cgroup both in the child (so
1313          * that we can be sure that no user code is ever executed
1314          * outside of the cgroup) and in the parent (so that we can be
1315          * sure that when we kill the cgroup the process will be
1316          * killed too). */
1317         if (cgroup_bondings)
1318                 cgroup_bonding_install_list(cgroup_bondings, pid);
1319
1320         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1321
1322         exec_status_start(&command->exec_status, pid);
1323
1324         *ret = pid;
1325         return 0;
1326 }
1327
1328 void exec_context_init(ExecContext *c) {
1329         assert(c);
1330
1331         c->umask = 0002;
1332         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1333         c->cpu_sched_policy = SCHED_OTHER;
1334         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1335         c->syslog_level_prefix = true;
1336         c->mount_flags = MS_SHARED;
1337         c->kill_signal = SIGTERM;
1338 }
1339
1340 void exec_context_done(ExecContext *c) {
1341         unsigned l;
1342
1343         assert(c);
1344
1345         strv_free(c->environment);
1346         c->environment = NULL;
1347
1348         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1349                 free(c->rlimit[l]);
1350                 c->rlimit[l] = NULL;
1351         }
1352
1353         free(c->working_directory);
1354         c->working_directory = NULL;
1355         free(c->root_directory);
1356         c->root_directory = NULL;
1357
1358         free(c->tty_path);
1359         c->tty_path = NULL;
1360
1361         free(c->tcpwrap_name);
1362         c->tcpwrap_name = NULL;
1363
1364         free(c->syslog_identifier);
1365         c->syslog_identifier = NULL;
1366
1367         free(c->user);
1368         c->user = NULL;
1369
1370         free(c->group);
1371         c->group = NULL;
1372
1373         strv_free(c->supplementary_groups);
1374         c->supplementary_groups = NULL;
1375
1376         free(c->pam_name);
1377         c->pam_name = NULL;
1378
1379         if (c->capabilities) {
1380                 cap_free(c->capabilities);
1381                 c->capabilities = NULL;
1382         }
1383
1384         strv_free(c->read_only_dirs);
1385         c->read_only_dirs = NULL;
1386
1387         strv_free(c->read_write_dirs);
1388         c->read_write_dirs = NULL;
1389
1390         strv_free(c->inaccessible_dirs);
1391         c->inaccessible_dirs = NULL;
1392
1393         if (c->cpuset)
1394                 CPU_FREE(c->cpuset);
1395 }
1396
1397 void exec_command_done(ExecCommand *c) {
1398         assert(c);
1399
1400         free(c->path);
1401         c->path = NULL;
1402
1403         strv_free(c->argv);
1404         c->argv = NULL;
1405 }
1406
1407 void exec_command_done_array(ExecCommand *c, unsigned n) {
1408         unsigned i;
1409
1410         for (i = 0; i < n; i++)
1411                 exec_command_done(c+i);
1412 }
1413
1414 void exec_command_free_list(ExecCommand *c) {
1415         ExecCommand *i;
1416
1417         while ((i = c)) {
1418                 LIST_REMOVE(ExecCommand, command, c, i);
1419                 exec_command_done(i);
1420                 free(i);
1421         }
1422 }
1423
1424 void exec_command_free_array(ExecCommand **c, unsigned n) {
1425         unsigned i;
1426
1427         for (i = 0; i < n; i++) {
1428                 exec_command_free_list(c[i]);
1429                 c[i] = NULL;
1430         }
1431 }
1432
1433 static void strv_fprintf(FILE *f, char **l) {
1434         char **g;
1435
1436         assert(f);
1437
1438         STRV_FOREACH(g, l)
1439                 fprintf(f, " %s", *g);
1440 }
1441
1442 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1443         char ** e;
1444         unsigned i;
1445
1446         assert(c);
1447         assert(f);
1448
1449         if (!prefix)
1450                 prefix = "";
1451
1452         fprintf(f,
1453                 "%sUMask: %04o\n"
1454                 "%sWorkingDirectory: %s\n"
1455                 "%sRootDirectory: %s\n"
1456                 "%sNonBlocking: %s\n"
1457                 "%sPrivateTmp: %s\n",
1458                 prefix, c->umask,
1459                 prefix, c->working_directory ? c->working_directory : "/",
1460                 prefix, c->root_directory ? c->root_directory : "/",
1461                 prefix, yes_no(c->non_blocking),
1462                 prefix, yes_no(c->private_tmp));
1463
1464         if (c->environment)
1465                 for (e = c->environment; *e; e++)
1466                         fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1467
1468         if (c->tcpwrap_name)
1469                 fprintf(f,
1470                         "%sTCPWrapName: %s\n",
1471                         prefix, c->tcpwrap_name);
1472
1473         if (c->nice_set)
1474                 fprintf(f,
1475                         "%sNice: %i\n",
1476                         prefix, c->nice);
1477
1478         if (c->oom_score_adjust_set)
1479                 fprintf(f,
1480                         "%sOOMScoreAdjust: %i\n",
1481                         prefix, c->oom_score_adjust);
1482
1483         for (i = 0; i < RLIM_NLIMITS; i++)
1484                 if (c->rlimit[i])
1485                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1486
1487         if (c->ioprio_set)
1488                 fprintf(f,
1489                         "%sIOSchedulingClass: %s\n"
1490                         "%sIOPriority: %i\n",
1491                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1492                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1493
1494         if (c->cpu_sched_set)
1495                 fprintf(f,
1496                         "%sCPUSchedulingPolicy: %s\n"
1497                         "%sCPUSchedulingPriority: %i\n"
1498                         "%sCPUSchedulingResetOnFork: %s\n",
1499                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1500                         prefix, c->cpu_sched_priority,
1501                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1502
1503         if (c->cpuset) {
1504                 fprintf(f, "%sCPUAffinity:", prefix);
1505                 for (i = 0; i < c->cpuset_ncpus; i++)
1506                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1507                                 fprintf(f, " %i", i);
1508                 fputs("\n", f);
1509         }
1510
1511         if (c->timer_slack_nsec_set)
1512                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1513
1514         fprintf(f,
1515                 "%sStandardInput: %s\n"
1516                 "%sStandardOutput: %s\n"
1517                 "%sStandardError: %s\n",
1518                 prefix, exec_input_to_string(c->std_input),
1519                 prefix, exec_output_to_string(c->std_output),
1520                 prefix, exec_output_to_string(c->std_error));
1521
1522         if (c->tty_path)
1523                 fprintf(f,
1524                         "%sTTYPath: %s\n",
1525                         prefix, c->tty_path);
1526
1527         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1528             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG)
1529                 fprintf(f,
1530                         "%sSyslogFacility: %s\n"
1531                         "%sSyslogLevel: %s\n",
1532                         prefix, log_facility_to_string(LOG_FAC(c->syslog_priority)),
1533                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1534
1535         if (c->capabilities) {
1536                 char *t;
1537                 if ((t = cap_to_text(c->capabilities, NULL))) {
1538                         fprintf(f, "%sCapabilities: %s\n",
1539                                 prefix, t);
1540                         cap_free(t);
1541                 }
1542         }
1543
1544         if (c->secure_bits)
1545                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1546                         prefix,
1547                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1548                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1549                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1550                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1551                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1552                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1553
1554         if (c->capability_bounding_set_drop) {
1555                 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
1556
1557                 for (i = 0; i <= CAP_LAST_CAP; i++)
1558                         if (c->capability_bounding_set_drop & (1 << i)) {
1559                                 char *t;
1560
1561                                 if ((t = cap_to_name(i))) {
1562                                         fprintf(f, " %s", t);
1563                                         free(t);
1564                                 }
1565                         }
1566
1567                 fputs("\n", f);
1568         }
1569
1570         if (c->user)
1571                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1572         if (c->group)
1573                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1574
1575         if (strv_length(c->supplementary_groups) > 0) {
1576                 fprintf(f, "%sSupplementaryGroups:", prefix);
1577                 strv_fprintf(f, c->supplementary_groups);
1578                 fputs("\n", f);
1579         }
1580
1581         if (c->pam_name)
1582                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1583
1584         if (strv_length(c->read_write_dirs) > 0) {
1585                 fprintf(f, "%sReadWriteDirs:", prefix);
1586                 strv_fprintf(f, c->read_write_dirs);
1587                 fputs("\n", f);
1588         }
1589
1590         if (strv_length(c->read_only_dirs) > 0) {
1591                 fprintf(f, "%sReadOnlyDirs:", prefix);
1592                 strv_fprintf(f, c->read_only_dirs);
1593                 fputs("\n", f);
1594         }
1595
1596         if (strv_length(c->inaccessible_dirs) > 0) {
1597                 fprintf(f, "%sInaccessibleDirs:", prefix);
1598                 strv_fprintf(f, c->inaccessible_dirs);
1599                 fputs("\n", f);
1600         }
1601
1602         fprintf(f,
1603                 "%sKillMode: %s\n"
1604                 "%sKillSignal: SIG%s\n",
1605                 prefix, kill_mode_to_string(c->kill_mode),
1606                 prefix, signal_to_string(c->kill_signal));
1607 }
1608
1609 void exec_status_start(ExecStatus *s, pid_t pid) {
1610         assert(s);
1611
1612         zero(*s);
1613         s->pid = pid;
1614         dual_timestamp_get(&s->start_timestamp);
1615 }
1616
1617 void exec_status_exit(ExecStatus *s, pid_t pid, int code, int status) {
1618         assert(s);
1619
1620         if ((s->pid && s->pid != pid) ||
1621             !s->start_timestamp.realtime <= 0)
1622                 zero(*s);
1623
1624         s->pid = pid;
1625         dual_timestamp_get(&s->exit_timestamp);
1626
1627         s->code = code;
1628         s->status = status;
1629 }
1630
1631 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1632         char buf[FORMAT_TIMESTAMP_MAX];
1633
1634         assert(s);
1635         assert(f);
1636
1637         if (!prefix)
1638                 prefix = "";
1639
1640         if (s->pid <= 0)
1641                 return;
1642
1643         fprintf(f,
1644                 "%sPID: %lu\n",
1645                 prefix, (unsigned long) s->pid);
1646
1647         if (s->start_timestamp.realtime > 0)
1648                 fprintf(f,
1649                         "%sStart Timestamp: %s\n",
1650                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1651
1652         if (s->exit_timestamp.realtime > 0)
1653                 fprintf(f,
1654                         "%sExit Timestamp: %s\n"
1655                         "%sExit Code: %s\n"
1656                         "%sExit Status: %i\n",
1657                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1658                         prefix, sigchld_code_to_string(s->code),
1659                         prefix, s->status);
1660 }
1661
1662 char *exec_command_line(char **argv) {
1663         size_t k;
1664         char *n, *p, **a;
1665         bool first = true;
1666
1667         assert(argv);
1668
1669         k = 1;
1670         STRV_FOREACH(a, argv)
1671                 k += strlen(*a)+3;
1672
1673         if (!(n = new(char, k)))
1674                 return NULL;
1675
1676         p = n;
1677         STRV_FOREACH(a, argv) {
1678
1679                 if (!first)
1680                         *(p++) = ' ';
1681                 else
1682                         first = false;
1683
1684                 if (strpbrk(*a, WHITESPACE)) {
1685                         *(p++) = '\'';
1686                         p = stpcpy(p, *a);
1687                         *(p++) = '\'';
1688                 } else
1689                         p = stpcpy(p, *a);
1690
1691         }
1692
1693         *p = 0;
1694
1695         /* FIXME: this doesn't really handle arguments that have
1696          * spaces and ticks in them */
1697
1698         return n;
1699 }
1700
1701 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1702         char *p2;
1703         const char *prefix2;
1704
1705         char *cmd;
1706
1707         assert(c);
1708         assert(f);
1709
1710         if (!prefix)
1711                 prefix = "";
1712         p2 = strappend(prefix, "\t");
1713         prefix2 = p2 ? p2 : prefix;
1714
1715         cmd = exec_command_line(c->argv);
1716
1717         fprintf(f,
1718                 "%sCommand Line: %s\n",
1719                 prefix, cmd ? cmd : strerror(ENOMEM));
1720
1721         free(cmd);
1722
1723         exec_status_dump(&c->exec_status, f, prefix2);
1724
1725         free(p2);
1726 }
1727
1728 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1729         assert(f);
1730
1731         if (!prefix)
1732                 prefix = "";
1733
1734         LIST_FOREACH(command, c, c)
1735                 exec_command_dump(c, f, prefix);
1736 }
1737
1738 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1739         ExecCommand *end;
1740
1741         assert(l);
1742         assert(e);
1743
1744         if (*l) {
1745                 /* It's kinda important that we keep the order here */
1746                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1747                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1748         } else
1749               *l = e;
1750 }
1751
1752 int exec_command_set(ExecCommand *c, const char *path, ...) {
1753         va_list ap;
1754         char **l, *p;
1755
1756         assert(c);
1757         assert(path);
1758
1759         va_start(ap, path);
1760         l = strv_new_ap(path, ap);
1761         va_end(ap);
1762
1763         if (!l)
1764                 return -ENOMEM;
1765
1766         if (!(p = strdup(path))) {
1767                 strv_free(l);
1768                 return -ENOMEM;
1769         }
1770
1771         free(c->path);
1772         c->path = p;
1773
1774         strv_free(c->argv);
1775         c->argv = l;
1776
1777         return 0;
1778 }
1779
1780 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1781         [EXEC_INPUT_NULL] = "null",
1782         [EXEC_INPUT_TTY] = "tty",
1783         [EXEC_INPUT_TTY_FORCE] = "tty-force",
1784         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
1785         [EXEC_INPUT_SOCKET] = "socket"
1786 };
1787
1788 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
1789         [EXEC_OUTPUT_INHERIT] = "inherit",
1790         [EXEC_OUTPUT_NULL] = "null",
1791         [EXEC_OUTPUT_TTY] = "tty",
1792         [EXEC_OUTPUT_SYSLOG] = "syslog",
1793         [EXEC_OUTPUT_KMSG] = "kmsg",
1794         [EXEC_OUTPUT_SOCKET] = "socket"
1795 };
1796
1797 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
1798
1799 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);