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 enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
 553         bool keep_groups = false;
 554         int r;
 555
 556         assert(context);
 557
 558         /* Lookup and set GID and supplementary group list. Here too
 559          * we avoid NSS lookups for gid=0. */
 560
 561         if (context->group || username) {
 562
 563                 if (context->group) {
 564                         const char *g = context->group;
 565
 566                         if ((r = get_group_creds(&g, &gid)) < 0)
 567                                 return r;
 568                 }
 569
 570                 /* First step, initialize groups from /etc/groups */
 571                 if (username && gid != 0) {
 572                         if (initgroups(username, gid) < 0)
 573                                 return -errno;
 574
 575                         keep_groups = true;
 576                 }
 577
 578                 /* Second step, set our gids */
 579                 if (setresgid(gid, gid, gid) < 0)
 580                         return -errno;
 581         }
 582
 583         if (context->supplementary_groups) {
 584                 int ngroups_max, k;
 585                 gid_t *gids;
 586                 char **i;
 587
 588                 /* Final step, initialize any manually set supplementary groups */
 589                 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
 590
 591                 if (!(gids = new(gid_t, ngroups_max)))
 592                         return -ENOMEM;
 593
 594                 if (keep_groups) {
 595                         if ((k = getgroups(ngroups_max, gids)) < 0) {
 596                                 free(gids);
 597                                 return -errno;
 598                         }
 599                 } else
 600                         k = 0;
 601
 602                 STRV_FOREACH(i, context->supplementary_groups) {
 603                         const char *g;
 604
 605                         if (k >= ngroups_max) {
 606                                 free(gids);
 607                                 return -E2BIG;
 608                         }
 609
 610                         g = *i;
 611                         r = get_group_creds(&g, gids+k);
 612                         if (r < 0) {
 613                                 free(gids);
 614                                 return r;
 615                         }
 616
 617                         k++;
 618                 }
 619
 620                 if (setgroups(k, gids) < 0) {
 621                         free(gids);
 622                         return -errno;
 623                 }
 624
 625                 free(gids);
 626         }
 627
 628         return 0;
 629 }
 630
 631 static int enforce_user(const ExecContext *context, uid_t uid) {
 632         int r;
 633         assert(context);
 634
 635         /* Sets (but doesn't lookup) the uid and make sure we keep the
 636          * capabilities while doing so. */
 637
 638         if (context->capabilities) {
 639                 cap_t d;
 640                 static const cap_value_t bits[] = {
 641                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
 642                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
 643                 };
 644
 645                 /* First step: If we need to keep capabilities but
 646                  * drop privileges we need to make sure we keep our
 647                  * caps, whiel we drop privileges. */
 648                 if (uid != 0) {
 649                         int sb = context->secure_bits|SECURE_KEEP_CAPS;
 650
 651                         if (prctl(PR_GET_SECUREBITS) != sb)
 652                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 653                                         return -errno;
 654                 }
 655
 656                 /* Second step: set the capabilities. This will reduce
 657                  * the capabilities to the minimum we need. */
 658
 659                 if (!(d = cap_dup(context->capabilities)))
 660                         return -errno;
 661
 662                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
 663                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
 664                         r = -errno;
 665                         cap_free(d);
 666                         return r;
 667                 }
 668
 669                 if (cap_set_proc(d) < 0) {
 670                         r = -errno;
 671                         cap_free(d);
 672                         return r;
 673                 }
 674
 675                 cap_free(d);
 676         }
 677
 678         /* Third step: actually set the uids */
 679         if (setresuid(uid, uid, uid) < 0)
 680                 return -errno;
 681
 682         /* At this point we should have all necessary capabilities but
 683            are otherwise a normal user. However, the caps might got
 684            corrupted due to the setresuid() so we need clean them up
 685            later. This is done outside of this call. */
 686
 687         return 0;
 688 }
 689
 690 #ifdef HAVE_PAM
 691
 692 static int null_conv(
 693                 int num_msg,
 694                 const struct pam_message **msg,
 695                 struct pam_response **resp,
 696                 void *appdata_ptr) {
 697
 698         /* We don't support conversations */
 699
 700         return PAM_CONV_ERR;
 701 }
 702
 703 static int setup_pam(
 704                 const char *name,
 705                 const char *user,
 706                 const char *tty,
 707                 char ***pam_env,
 708                 int fds[], unsigned n_fds) {
 709
 710         static const struct pam_conv conv = {
 711                 .conv = null_conv,
 712                 .appdata_ptr = NULL
 713         };
 714
 715         pam_handle_t *handle = NULL;
 716         sigset_t ss, old_ss;
 717         int pam_code = PAM_SUCCESS;
 718         char **e = NULL;
 719         bool close_session = false;
 720         pid_t pam_pid = 0, parent_pid;
 721
 722         assert(name);
 723         assert(user);
 724         assert(pam_env);
 725
 726         /* We set up PAM in the parent process, then fork. The child
 727          * will then stay around until killed via PR_GET_PDEATHSIG or
 728          * systemd via the cgroup logic. It will then remove the PAM
 729          * session again. The parent process will exec() the actual
 730          * daemon. We do things this way to ensure that the main PID
 731          * of the daemon is the one we initially fork()ed. */
 732
 733         if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
 734                 handle = NULL;
 735                 goto fail;
 736         }
 737
 738         if (tty)
 739                 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
 740                         goto fail;
 741
 742         if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
 743                 goto fail;
 744
 745         if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
 746                 goto fail;
 747
 748         close_session = true;
 749
 750         if ((!(e = pam_getenvlist(handle)))) {
 751                 pam_code = PAM_BUF_ERR;
 752                 goto fail;
 753         }
 754
 755         /* Block SIGTERM, so that we know that it won't get lost in
 756          * the child */
 757         if (sigemptyset(&ss) < 0 ||
 758             sigaddset(&ss, SIGTERM) < 0 ||
 759             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 760                 goto fail;
 761
 762         parent_pid = getpid();
 763
 764         if ((pam_pid = fork()) < 0)
 765                 goto fail;
 766
 767         if (pam_pid == 0) {
 768                 int sig;
 769                 int r = EXIT_PAM;
 770
 771                 /* The child's job is to reset the PAM session on
 772                  * termination */
 773
 774                 /* This string must fit in 10 chars (i.e. the length
 775                  * of "/sbin/init") */
 776                 rename_process("sd(PAM)");
 777
 778                 /* Make sure we don't keep open the passed fds in this
 779                 child. We assume that otherwise only those fds are
 780                 open here that have been opened by PAM. */
 781                 close_many(fds, n_fds);
 782
 783                 /* Wait until our parent died. This will most likely
 784                  * not work since the kernel does not allow
 785                  * unprivileged parents kill their privileged children
 786                  * this way. We rely on the control groups kill logic
 787                  * to do the rest for us. */
 788                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 789                         goto child_finish;
 790
 791                 /* Check if our parent process might already have
 792                  * died? */
 793                 if (getppid() == parent_pid) {
 794                         for (;;) {
 795                                 if (sigwait(&ss, &sig) < 0) {
 796                                         if (errno == EINTR)
 797                                                 continue;
 798
 799                                         goto child_finish;
 800                                 }
 801
 802                                 assert(sig == SIGTERM);
 803                                 break;
 804                         }
 805                 }
 806
 807                 /* If our parent died we'll end the session */
 808                 if (getppid() != parent_pid)
 809                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
 810                                 goto child_finish;
 811
 812                 r = 0;
 813
 814         child_finish:
 815                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 816                 _exit(r);
 817         }
 818
 819         /* If the child was forked off successfully it will do all the
 820          * cleanups, so forget about the handle here. */
 821         handle = NULL;
 822
 823         /* Unblock SIGTERM again in the parent */
 824         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 825                 goto fail;
 826
 827         /* We close the log explicitly here, since the PAM modules
 828          * might have opened it, but we don't want this fd around. */
 829         closelog();
 830
 831         *pam_env = e;
 832         e = NULL;
 833
 834         return 0;
 835
 836 fail:
 837         if (handle) {
 838                 if (close_session)
 839                         pam_code = pam_close_session(handle, PAM_DATA_SILENT);
 840
 841                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 842         }
 843
 844         strv_free(e);
 845
 846         closelog();
 847
 848         if (pam_pid > 1) {
 849                 kill(pam_pid, SIGTERM);
 850                 kill(pam_pid, SIGCONT);
 851         }
 852
 853         return EXIT_PAM;
 854 }
 855 #endif
 856
 857 static int do_capability_bounding_set_drop(uint64_t drop) {
 858         unsigned long i;
 859         cap_t old_cap = NULL, new_cap = NULL;
 860         cap_flag_value_t fv;
 861         int r;
 862
 863         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 864          * in the effective set (yes, the kernel drops that when
 865          * executing init!), so get it back temporarily so that we can
 866          * call PR_CAPBSET_DROP. */
 867
 868         old_cap = cap_get_proc();
 869         if (!old_cap)
 870                 return -errno;
 871
 872         if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
 873                 r = -errno;
 874                 goto finish;
 875         }
 876
 877         if (fv != CAP_SET) {
 878                 static const cap_value_t v = CAP_SETPCAP;
 879
 880                 new_cap = cap_dup(old_cap);
 881                 if (!new_cap) {
 882                         r = -errno;
 883                         goto finish;
 884                 }
 885
 886                 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
 887                         r = -errno;
 888                         goto finish;
 889                 }
 890
 891                 if (cap_set_proc(new_cap) < 0) {
 892                         r = -errno;
 893                         goto finish;
 894                 }
 895         }
 896
 897         for (i = 0; i <= MAX(63LU, (unsigned long) CAP_LAST_CAP); i++)
 898                 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
 899                         if (prctl(PR_CAPBSET_DROP, i) < 0) {
 900                                 if (errno == EINVAL)
 901                                         break;
 902
 903                                 r = -errno;
 904                                 goto finish;
 905                         }
 906                 }
 907
 908         r = 0;
 909
 910 finish:
 911         if (new_cap)
 912                 cap_free(new_cap);
 913
 914         if (old_cap) {
 915                 cap_set_proc(old_cap);
 916                 cap_free(old_cap);
 917         }
 918
 919         return r;
 920 }
 921
 922 int exec_spawn(ExecCommand *command,
 923                char **argv,
 924                const ExecContext *context,
 925                int fds[], unsigned n_fds,
 926                char **environment,
 927                bool apply_permissions,
 928                bool apply_chroot,
 929                bool apply_tty_stdin,
 930                bool confirm_spawn,
 931                CGroupBonding *cgroup_bondings,
 932                pid_t *ret) {
 933
 934         pid_t pid;
 935         int r;
 936         char *line;
 937         int socket_fd;
 938         char **files_env = NULL;
 939
 940         assert(command);
 941         assert(context);
 942         assert(ret);
 943         assert(fds || n_fds <= 0);
 944
 945         if (context->std_input == EXEC_INPUT_SOCKET ||
 946             context->std_output == EXEC_OUTPUT_SOCKET ||
 947             context->std_error == EXEC_OUTPUT_SOCKET) {
 948
 949                 if (n_fds != 1)
 950                         return -EINVAL;
 951
 952                 socket_fd = fds[0];
 953
 954                 fds = NULL;
 955                 n_fds = 0;
 956         } else
 957                 socket_fd = -1;
 958
 959         if ((r = exec_context_load_environment(context, &files_env)) < 0) {
 960                 log_error("Failed to load environment files: %s", strerror(-r));
 961                 return r;
 962         }
 963
 964         if (!argv)
 965                 argv = command->argv;
 966
 967         if (!(line = exec_command_line(argv))) {
 968                 r = -ENOMEM;
 969                 goto fail_parent;
 970         }
 971
 972         log_debug("About to execute: %s", line);
 973         free(line);
 974
 975         if (cgroup_bondings)
 976                 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
 977                         goto fail_parent;
 978
 979         if ((pid = fork()) < 0) {
 980                 r = -errno;
 981                 goto fail_parent;
 982         }
 983
 984         if (pid == 0) {
 985                 int i;
 986                 sigset_t ss;
 987                 const char *username = NULL, *home = NULL;
 988                 uid_t uid = (uid_t) -1;
 989                 gid_t gid = (gid_t) -1;
 990                 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
 991                 unsigned n_env = 0;
 992                 int saved_stdout = -1, saved_stdin = -1;
 993                 bool keep_stdout = false, keep_stdin = false;
 994
 995                 /* child */
 996
 997                 /* This string must fit in 10 chars (i.e. the length
 998                  * of "/sbin/init") */
 999                 rename_process("sd(EXEC)");
1000
1001                 /* We reset exactly these signals, since they are the
1002                  * only ones we set to SIG_IGN in the main daemon. All
1003                  * others we leave untouched because we set them to
1004                  * SIG_DFL or a valid handler initially, both of which
1005                  * will be demoted to SIG_DFL. */
1006                 default_signals(SIGNALS_CRASH_HANDLER,
1007                                 SIGNALS_IGNORE, -1);
1008
1009                 if (sigemptyset(&ss) < 0 ||
1010                     sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1011                         r = EXIT_SIGNAL_MASK;
1012                         goto fail_child;
1013                 }
1014
1015                 /* Close sockets very early to make sure we don't
1016                  * block init reexecution because it cannot bind its
1017                  * sockets */
1018                 if (close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1019                                   socket_fd >= 0 ? 1 : n_fds) < 0) {
1020                         r = EXIT_FDS;
1021                         goto fail_child;
1022                 }
1023
1024                 if (!context->same_pgrp)
1025                         if (setsid() < 0) {
1026                                 r = EXIT_SETSID;
1027                                 goto fail_child;
1028                         }
1029
1030                 if (context->tcpwrap_name) {
1031                         if (socket_fd >= 0)
1032                                 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1033                                         r = EXIT_TCPWRAP;
1034                                         goto fail_child;
1035                                 }
1036
1037                         for (i = 0; i < (int) n_fds; i++) {
1038                                 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1039                                         r = EXIT_TCPWRAP;
1040                                         goto fail_child;
1041                                 }
1042                         }
1043                 }
1044
1045                 exec_context_tty_reset(context);
1046
1047                 /* We skip the confirmation step if we shall not apply the TTY */
1048                 if (confirm_spawn &&
1049                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1050                         char response;
1051
1052                         /* Set up terminal for the question */
1053                         if ((r = setup_confirm_stdio(context,
1054                                                      &saved_stdin, &saved_stdout)))
1055                                 goto fail_child;
1056
1057                         /* Now ask the question. */
1058                         if (!(line = exec_command_line(argv))) {
1059                                 r = EXIT_MEMORY;
1060                                 goto fail_child;
1061                         }
1062
1063                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1064                         free(line);
1065
1066                         if (r < 0 || response == 'n') {
1067                                 r = EXIT_CONFIRM;
1068                                 goto fail_child;
1069                         } else if (response == 's') {
1070                                 r = 0;
1071                                 goto fail_child;
1072                         }
1073
1074                         /* Release terminal for the question */
1075                         if ((r = restore_confirm_stdio(context,
1076                                                        &saved_stdin, &saved_stdout,
1077                                                        &keep_stdin, &keep_stdout)))
1078                                 goto fail_child;
1079                 }
1080
1081                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1082                  * must sure to drop O_NONBLOCK */
1083                 if (socket_fd >= 0)
1084                         fd_nonblock(socket_fd, false);
1085
1086                 if (!keep_stdin)
1087                         if (setup_input(context, socket_fd, apply_tty_stdin) < 0) {
1088                                 r = EXIT_STDIN;
1089                                 goto fail_child;
1090                         }
1091
1092                 if (!keep_stdout)
1093                         if (setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1094                                 r = EXIT_STDOUT;
1095                                 goto fail_child;
1096                         }
1097
1098                 if (setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1099                         r = EXIT_STDERR;
1100                         goto fail_child;
1101                 }
1102
1103                 if (cgroup_bondings)
1104                         if (cgroup_bonding_install_list(cgroup_bondings, 0) < 0) {
1105                                 r = EXIT_CGROUP;
1106                                 goto fail_child;
1107                         }
1108
1109                 if (context->oom_score_adjust_set) {
1110                         char t[16];
1111
1112                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1113                         char_array_0(t);
1114
1115                         if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1116                                 /* Compatibility with Linux <= 2.6.35 */
1117
1118                                 int adj;
1119
1120                                 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1121                                 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1122
1123                                 snprintf(t, sizeof(t), "%i", adj);
1124                                 char_array_0(t);
1125
1126                                 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1127                                     && errno != EACCES) {
1128                                         r = EXIT_OOM_ADJUST;
1129                                         goto fail_child;
1130                                 }
1131                         }
1132                 }
1133
1134                 if (context->nice_set)
1135                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1136                                 r = EXIT_NICE;
1137                                 goto fail_child;
1138                         }
1139
1140                 if (context->cpu_sched_set) {
1141                         struct sched_param param;
1142
1143                         zero(param);
1144                         param.sched_priority = context->cpu_sched_priority;
1145
1146                         if (sched_setscheduler(0, context->cpu_sched_policy |
1147                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
1148                                 r = EXIT_SETSCHEDULER;
1149                                 goto fail_child;
1150                         }
1151                 }
1152
1153                 if (context->cpuset)
1154                         if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1155                                 r = EXIT_CPUAFFINITY;
1156                                 goto fail_child;
1157                         }
1158
1159                 if (context->ioprio_set)
1160                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1161                                 r = EXIT_IOPRIO;
1162                                 goto fail_child;
1163                         }
1164
1165                 if (context->timer_slack_nsec_set)
1166                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1167                                 r = EXIT_TIMERSLACK;
1168                                 goto fail_child;
1169                         }
1170
1171                 if (context->utmp_id)
1172                         utmp_put_init_process(0, context->utmp_id, getpid(), getsid(0), context->tty_path);
1173
1174                 if (context->user) {
1175                         username = context->user;
1176                         if (get_user_creds(&username, &uid, &gid, &home) < 0) {
1177                                 r = EXIT_USER;
1178                                 goto fail_child;
1179                         }
1180
1181                         if (is_terminal_input(context->std_input))
1182                                 if (chown_terminal(STDIN_FILENO, uid) < 0) {
1183                                         r = EXIT_STDIN;
1184                                         goto fail_child;
1185                                 }
1186
1187                         if (cgroup_bondings && context->control_group_modify)
1188                                 if (cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid) < 0 ||
1189                                     cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid) < 0) {
1190                                         r = EXIT_CGROUP;
1191                                         goto fail_child;
1192                                 }
1193                 }
1194
1195                 if (apply_permissions)
1196                         if (enforce_groups(context, username, uid) < 0) {
1197                                 r = EXIT_GROUP;
1198                                 goto fail_child;
1199                         }
1200
1201                 umask(context->umask);
1202
1203 #ifdef HAVE_PAM
1204                 if (context->pam_name && username) {
1205                         if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) != 0) {
1206                                 r = EXIT_PAM;
1207                                 goto fail_child;
1208                         }
1209                 }
1210 #endif
1211
1212                 if (strv_length(context->read_write_dirs) > 0 ||
1213                     strv_length(context->read_only_dirs) > 0 ||
1214                     strv_length(context->inaccessible_dirs) > 0 ||
1215                     context->mount_flags != MS_SHARED ||
1216                     context->private_tmp)
1217                         if ((r = setup_namespace(
1218                                              context->read_write_dirs,
1219                                              context->read_only_dirs,
1220                                              context->inaccessible_dirs,
1221                                              context->private_tmp,
1222                                              context->mount_flags)) < 0)
1223                                 goto fail_child;
1224
1225                 if (apply_chroot) {
1226                         if (context->root_directory)
1227                                 if (chroot(context->root_directory) < 0) {
1228                                         r = EXIT_CHROOT;
1229                                         goto fail_child;
1230                                 }
1231
1232                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1233                                 r = EXIT_CHDIR;
1234                                 goto fail_child;
1235                         }
1236                 } else {
1237
1238                         char *d;
1239
1240                         if (asprintf(&d, "%s/%s",
1241                                      context->root_directory ? context->root_directory : "",
1242                                      context->working_directory ? context->working_directory : "") < 0) {
1243                                 r = EXIT_MEMORY;
1244                                 goto fail_child;
1245                         }
1246
1247                         if (chdir(d) < 0) {
1248                                 free(d);
1249                                 r = EXIT_CHDIR;
1250                                 goto fail_child;
1251                         }
1252
1253                         free(d);
1254                 }
1255
1256                 /* We repeat the fd closing here, to make sure that
1257                  * nothing is leaked from the PAM modules */
1258                 if (close_all_fds(fds, n_fds) < 0 ||
1259                     shift_fds(fds, n_fds) < 0 ||
1260                     flags_fds(fds, n_fds, context->non_blocking) < 0) {
1261                         r = EXIT_FDS;
1262                         goto fail_child;
1263                 }
1264
1265                 if (apply_permissions) {
1266
1267                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1268                                 if (!context->rlimit[i])
1269                                         continue;
1270
1271                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1272                                         r = EXIT_LIMITS;
1273                                         goto fail_child;
1274                                 }
1275                         }
1276
1277                         if (context->capability_bounding_set_drop)
1278                                 if (do_capability_bounding_set_drop(context->capability_bounding_set_drop) < 0) {
1279                                         r = EXIT_CAPABILITIES;
1280                                         goto fail_child;
1281                                 }
1282
1283                         if (context->user)
1284                                 if (enforce_user(context, uid) < 0) {
1285                                         r = EXIT_USER;
1286                                         goto fail_child;
1287                                 }
1288
1289                         /* PR_GET_SECUREBITS is not privileged, while
1290                          * PR_SET_SECUREBITS is. So to suppress
1291                          * potential EPERMs we'll try not to call
1292                          * PR_SET_SECUREBITS unless necessary. */
1293                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1294                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1295                                         r = EXIT_SECUREBITS;
1296                                         goto fail_child;
1297                                 }
1298
1299                         if (context->capabilities)
1300                                 if (cap_set_proc(context->capabilities) < 0) {
1301                                         r = EXIT_CAPABILITIES;
1302                                         goto fail_child;
1303                                 }
1304                 }
1305
1306                 if (!(our_env = new0(char*, 7))) {
1307                         r = EXIT_MEMORY;
1308                         goto fail_child;
1309                 }
1310
1311                 if (n_fds > 0)
1312                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1313                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1314                                 r = EXIT_MEMORY;
1315                                 goto fail_child;
1316                         }
1317
1318                 if (home)
1319                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1320                                 r = EXIT_MEMORY;
1321                                 goto fail_child;
1322                         }
1323
1324                 if (username)
1325                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1326                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1327                                 r = EXIT_MEMORY;
1328                                 goto fail_child;
1329                         }
1330
1331                 if (is_terminal_input(context->std_input) ||
1332                     context->std_output == EXEC_OUTPUT_TTY ||
1333                     context->std_error == EXEC_OUTPUT_TTY)
1334                         if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1335                                 r = EXIT_MEMORY;
1336                                 goto fail_child;
1337                         }
1338
1339                 assert(n_env <= 7);
1340
1341                 if (!(final_env = strv_env_merge(
1342                                       5,
1343                                       environment,
1344                                       our_env,
1345                                       context->environment,
1346                                       files_env,
1347                                       pam_env,
1348                                       NULL))) {
1349                         r = EXIT_MEMORY;
1350                         goto fail_child;
1351                 }
1352
1353                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1354                         r = EXIT_MEMORY;
1355                         goto fail_child;
1356                 }
1357
1358                 final_env = strv_env_clean(final_env);
1359
1360                 execve(command->path, final_argv, final_env);
1361                 r = EXIT_EXEC;
1362
1363         fail_child:
1364                 strv_free(our_env);
1365                 strv_free(final_env);
1366                 strv_free(pam_env);
1367                 strv_free(files_env);
1368                 strv_free(final_argv);
1369
1370                 if (saved_stdin >= 0)
1371                         close_nointr_nofail(saved_stdin);
1372
1373                 if (saved_stdout >= 0)
1374                         close_nointr_nofail(saved_stdout);
1375
1376                 _exit(r);
1377         }
1378
1379         strv_free(files_env);
1380
1381         /* We add the new process to the cgroup both in the child (so
1382          * that we can be sure that no user code is ever executed
1383          * outside of the cgroup) and in the parent (so that we can be
1384          * sure that when we kill the cgroup the process will be
1385          * killed too). */
1386         if (cgroup_bondings)
1387                 cgroup_bonding_install_list(cgroup_bondings, pid);
1388
1389         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1390
1391         exec_status_start(&command->exec_status, pid);
1392
1393         *ret = pid;
1394         return 0;
1395
1396 fail_parent:
1397         strv_free(files_env);
1398
1399         return r;
1400 }
1401
1402 void exec_context_init(ExecContext *c) {
1403         assert(c);
1404
1405         c->umask = 0002;
1406         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1407         c->cpu_sched_policy = SCHED_OTHER;
1408         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1409         c->syslog_level_prefix = true;
1410         c->mount_flags = MS_SHARED;
1411         c->kill_signal = SIGTERM;
1412         c->send_sigkill = true;
1413 }
1414
1415 void exec_context_done(ExecContext *c) {
1416         unsigned l;
1417
1418         assert(c);
1419
1420         strv_free(c->environment);
1421         c->environment = NULL;
1422
1423         strv_free(c->environment_files);
1424         c->environment_files = NULL;
1425
1426         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1427                 free(c->rlimit[l]);
1428                 c->rlimit[l] = NULL;
1429         }
1430
1431         free(c->working_directory);
1432         c->working_directory = NULL;
1433         free(c->root_directory);
1434         c->root_directory = NULL;
1435
1436         free(c->tty_path);
1437         c->tty_path = NULL;
1438
1439         free(c->tcpwrap_name);
1440         c->tcpwrap_name = NULL;
1441
1442         free(c->syslog_identifier);
1443         c->syslog_identifier = NULL;
1444
1445         free(c->user);
1446         c->user = NULL;
1447
1448         free(c->group);
1449         c->group = NULL;
1450
1451         strv_free(c->supplementary_groups);
1452         c->supplementary_groups = NULL;
1453
1454         free(c->pam_name);
1455         c->pam_name = NULL;
1456
1457         if (c->capabilities) {
1458                 cap_free(c->capabilities);
1459                 c->capabilities = NULL;
1460         }
1461
1462         strv_free(c->read_only_dirs);
1463         c->read_only_dirs = NULL;
1464
1465         strv_free(c->read_write_dirs);
1466         c->read_write_dirs = NULL;
1467
1468         strv_free(c->inaccessible_dirs);
1469         c->inaccessible_dirs = NULL;
1470
1471         if (c->cpuset)
1472                 CPU_FREE(c->cpuset);
1473
1474         free(c->utmp_id);
1475         c->utmp_id = NULL;
1476 }
1477
1478 void exec_command_done(ExecCommand *c) {
1479         assert(c);
1480
1481         free(c->path);
1482         c->path = NULL;
1483
1484         strv_free(c->argv);
1485         c->argv = NULL;
1486 }
1487
1488 void exec_command_done_array(ExecCommand *c, unsigned n) {
1489         unsigned i;
1490
1491         for (i = 0; i < n; i++)
1492                 exec_command_done(c+i);
1493 }
1494
1495 void exec_command_free_list(ExecCommand *c) {
1496         ExecCommand *i;
1497
1498         while ((i = c)) {
1499                 LIST_REMOVE(ExecCommand, command, c, i);
1500                 exec_command_done(i);
1501                 free(i);
1502         }
1503 }
1504
1505 void exec_command_free_array(ExecCommand **c, unsigned n) {
1506         unsigned i;
1507
1508         for (i = 0; i < n; i++) {
1509                 exec_command_free_list(c[i]);
1510                 c[i] = NULL;
1511         }
1512 }
1513
1514 int exec_context_load_environment(const ExecContext *c, char ***l) {
1515         char **i, **r = NULL;
1516
1517         assert(c);
1518         assert(l);
1519
1520         STRV_FOREACH(i, c->environment_files) {
1521                 char *fn;
1522                 int k;
1523                 bool ignore = false;
1524                 char **p;
1525
1526                 fn = *i;
1527
1528                 if (fn[0] == '-') {
1529                         ignore = true;
1530                         fn ++;
1531                 }
1532
1533                 if (!path_is_absolute(fn)) {
1534
1535                         if (ignore)
1536                                 continue;
1537
1538                         strv_free(r);
1539                         return -EINVAL;
1540                 }
1541
1542                 if ((k = load_env_file(fn, &p)) < 0) {
1543
1544                         if (ignore)
1545                                 continue;
1546
1547                         strv_free(r);
1548                         return k;
1549                 }
1550
1551                 if (r == NULL)
1552                         r = p;
1553                 else {
1554                         char **m;
1555
1556                         m = strv_env_merge(2, r, p);
1557                         strv_free(r);
1558                         strv_free(p);
1559
1560                         if (!m)
1561                                 return -ENOMEM;
1562
1563                         r = m;
1564                 }
1565         }
1566
1567         *l = r;
1568
1569         return 0;
1570 }
1571
1572 static void strv_fprintf(FILE *f, char **l) {
1573         char **g;
1574
1575         assert(f);
1576
1577         STRV_FOREACH(g, l)
1578                 fprintf(f, " %s", *g);
1579 }
1580
1581 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1582         char ** e;
1583         unsigned i;
1584
1585         assert(c);
1586         assert(f);
1587
1588         if (!prefix)
1589                 prefix = "";
1590
1591         fprintf(f,
1592                 "%sUMask: %04o\n"
1593                 "%sWorkingDirectory: %s\n"
1594                 "%sRootDirectory: %s\n"
1595                 "%sNonBlocking: %s\n"
1596                 "%sPrivateTmp: %s\n"
1597                 "%sControlGroupModify: %s\n",
1598                 prefix, c->umask,
1599                 prefix, c->working_directory ? c->working_directory : "/",
1600                 prefix, c->root_directory ? c->root_directory : "/",
1601                 prefix, yes_no(c->non_blocking),
1602                 prefix, yes_no(c->private_tmp),
1603                 prefix, yes_no(c->control_group_modify));
1604
1605         STRV_FOREACH(e, c->environment)
1606                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1607
1608         STRV_FOREACH(e, c->environment_files)
1609                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1610
1611         if (c->tcpwrap_name)
1612                 fprintf(f,
1613                         "%sTCPWrapName: %s\n",
1614                         prefix, c->tcpwrap_name);
1615
1616         if (c->nice_set)
1617                 fprintf(f,
1618                         "%sNice: %i\n",
1619                         prefix, c->nice);
1620
1621         if (c->oom_score_adjust_set)
1622                 fprintf(f,
1623                         "%sOOMScoreAdjust: %i\n",
1624                         prefix, c->oom_score_adjust);
1625
1626         for (i = 0; i < RLIM_NLIMITS; i++)
1627                 if (c->rlimit[i])
1628                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1629
1630         if (c->ioprio_set)
1631                 fprintf(f,
1632                         "%sIOSchedulingClass: %s\n"
1633                         "%sIOPriority: %i\n",
1634                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1635                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1636
1637         if (c->cpu_sched_set)
1638                 fprintf(f,
1639                         "%sCPUSchedulingPolicy: %s\n"
1640                         "%sCPUSchedulingPriority: %i\n"
1641                         "%sCPUSchedulingResetOnFork: %s\n",
1642                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1643                         prefix, c->cpu_sched_priority,
1644                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1645
1646         if (c->cpuset) {
1647                 fprintf(f, "%sCPUAffinity:", prefix);
1648                 for (i = 0; i < c->cpuset_ncpus; i++)
1649                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1650                                 fprintf(f, " %i", i);
1651                 fputs("\n", f);
1652         }
1653
1654         if (c->timer_slack_nsec_set)
1655                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1656
1657         fprintf(f,
1658                 "%sStandardInput: %s\n"
1659                 "%sStandardOutput: %s\n"
1660                 "%sStandardError: %s\n",
1661                 prefix, exec_input_to_string(c->std_input),
1662                 prefix, exec_output_to_string(c->std_output),
1663                 prefix, exec_output_to_string(c->std_error));
1664
1665         if (c->tty_path)
1666                 fprintf(f,
1667                         "%sTTYPath: %s\n"
1668                         "%sTTYReset: %s\n"
1669                         "%sTTYVHangup: %s\n"
1670                         "%sTTYVTDisallocate: %s\n",
1671                         prefix, c->tty_path,
1672                         prefix, yes_no(c->tty_reset),
1673                         prefix, yes_no(c->tty_vhangup),
1674                         prefix, yes_no(c->tty_vt_disallocate));
1675
1676         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1677             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1678             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1679             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1680                 fprintf(f,
1681                         "%sSyslogFacility: %s\n"
1682                         "%sSyslogLevel: %s\n",
1683                         prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1684                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1685
1686         if (c->capabilities) {
1687                 char *t;
1688                 if ((t = cap_to_text(c->capabilities, NULL))) {
1689                         fprintf(f, "%sCapabilities: %s\n",
1690                                 prefix, t);
1691                         cap_free(t);
1692                 }
1693         }
1694
1695         if (c->secure_bits)
1696                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1697                         prefix,
1698                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1699                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1700                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1701                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1702                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1703                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1704
1705         if (c->capability_bounding_set_drop) {
1706                 unsigned long l;
1707                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1708
1709                 for (l = 0; l <= (unsigned long) CAP_LAST_CAP; l++)
1710                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1711                                 char *t;
1712
1713                                 if ((t = cap_to_name(l))) {
1714                                         fprintf(f, " %s", t);
1715                                         cap_free(t);
1716                                 }
1717                         }
1718
1719                 fputs("\n", f);
1720         }
1721
1722         if (c->user)
1723                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1724         if (c->group)
1725                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1726
1727         if (strv_length(c->supplementary_groups) > 0) {
1728                 fprintf(f, "%sSupplementaryGroups:", prefix);
1729                 strv_fprintf(f, c->supplementary_groups);
1730                 fputs("\n", f);
1731         }
1732
1733         if (c->pam_name)
1734                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1735
1736         if (strv_length(c->read_write_dirs) > 0) {
1737                 fprintf(f, "%sReadWriteDirs:", prefix);
1738                 strv_fprintf(f, c->read_write_dirs);
1739                 fputs("\n", f);
1740         }
1741
1742         if (strv_length(c->read_only_dirs) > 0) {
1743                 fprintf(f, "%sReadOnlyDirs:", prefix);
1744                 strv_fprintf(f, c->read_only_dirs);
1745                 fputs("\n", f);
1746         }
1747
1748         if (strv_length(c->inaccessible_dirs) > 0) {
1749                 fprintf(f, "%sInaccessibleDirs:", prefix);
1750                 strv_fprintf(f, c->inaccessible_dirs);
1751                 fputs("\n", f);
1752         }
1753
1754         fprintf(f,
1755                 "%sKillMode: %s\n"
1756                 "%sKillSignal: SIG%s\n"
1757                 "%sSendSIGKILL: %s\n",
1758                 prefix, kill_mode_to_string(c->kill_mode),
1759                 prefix, signal_to_string(c->kill_signal),
1760                 prefix, yes_no(c->send_sigkill));
1761
1762         if (c->utmp_id)
1763                 fprintf(f,
1764                         "%sUtmpIdentifier: %s\n",
1765                         prefix, c->utmp_id);
1766 }
1767
1768 void exec_status_start(ExecStatus *s, pid_t pid) {
1769         assert(s);
1770
1771         zero(*s);
1772         s->pid = pid;
1773         dual_timestamp_get(&s->start_timestamp);
1774 }
1775
1776 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1777         assert(s);
1778
1779         if ((s->pid && s->pid != pid) ||
1780             !s->start_timestamp.realtime <= 0)
1781                 zero(*s);
1782
1783         s->pid = pid;
1784         dual_timestamp_get(&s->exit_timestamp);
1785
1786         s->code = code;
1787         s->status = status;
1788
1789         if (context) {
1790                 if (context->utmp_id)
1791                         utmp_put_dead_process(context->utmp_id, pid, code, status);
1792
1793                 exec_context_tty_reset(context);
1794         }
1795 }
1796
1797 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1798         char buf[FORMAT_TIMESTAMP_MAX];
1799
1800         assert(s);
1801         assert(f);
1802
1803         if (!prefix)
1804                 prefix = "";
1805
1806         if (s->pid <= 0)
1807                 return;
1808
1809         fprintf(f,
1810                 "%sPID: %lu\n",
1811                 prefix, (unsigned long) s->pid);
1812
1813         if (s->start_timestamp.realtime > 0)
1814                 fprintf(f,
1815                         "%sStart Timestamp: %s\n",
1816                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1817
1818         if (s->exit_timestamp.realtime > 0)
1819                 fprintf(f,
1820                         "%sExit Timestamp: %s\n"
1821                         "%sExit Code: %s\n"
1822                         "%sExit Status: %i\n",
1823                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1824                         prefix, sigchld_code_to_string(s->code),
1825                         prefix, s->status);
1826 }
1827
1828 char *exec_command_line(char **argv) {
1829         size_t k;
1830         char *n, *p, **a;
1831         bool first = true;
1832
1833         assert(argv);
1834
1835         k = 1;
1836         STRV_FOREACH(a, argv)
1837                 k += strlen(*a)+3;
1838
1839         if (!(n = new(char, k)))
1840                 return NULL;
1841
1842         p = n;
1843         STRV_FOREACH(a, argv) {
1844
1845                 if (!first)
1846                         *(p++) = ' ';
1847                 else
1848                         first = false;
1849
1850                 if (strpbrk(*a, WHITESPACE)) {
1851                         *(p++) = '\'';
1852                         p = stpcpy(p, *a);
1853                         *(p++) = '\'';
1854                 } else
1855                         p = stpcpy(p, *a);
1856
1857         }
1858
1859         *p = 0;
1860
1861         /* FIXME: this doesn't really handle arguments that have
1862          * spaces and ticks in them */
1863
1864         return n;
1865 }
1866
1867 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1868         char *p2;
1869         const char *prefix2;
1870
1871         char *cmd;
1872
1873         assert(c);
1874         assert(f);
1875
1876         if (!prefix)
1877                 prefix = "";
1878         p2 = strappend(prefix, "\t");
1879         prefix2 = p2 ? p2 : prefix;
1880
1881         cmd = exec_command_line(c->argv);
1882
1883         fprintf(f,
1884                 "%sCommand Line: %s\n",
1885                 prefix, cmd ? cmd : strerror(ENOMEM));
1886
1887         free(cmd);
1888
1889         exec_status_dump(&c->exec_status, f, prefix2);
1890
1891         free(p2);
1892 }
1893
1894 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1895         assert(f);
1896
1897         if (!prefix)
1898                 prefix = "";
1899
1900         LIST_FOREACH(command, c, c)
1901                 exec_command_dump(c, f, prefix);
1902 }
1903
1904 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1905         ExecCommand *end;
1906
1907         assert(l);
1908         assert(e);
1909
1910         if (*l) {
1911                 /* It's kind of important, that we keep the order here */
1912                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1913                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1914         } else
1915               *l = e;
1916 }
1917
1918 int exec_command_set(ExecCommand *c, const char *path, ...) {
1919         va_list ap;
1920         char **l, *p;
1921
1922         assert(c);
1923         assert(path);
1924
1925         va_start(ap, path);
1926         l = strv_new_ap(path, ap);
1927         va_end(ap);
1928
1929         if (!l)
1930                 return -ENOMEM;
1931
1932         if (!(p = strdup(path))) {
1933                 strv_free(l);
1934                 return -ENOMEM;
1935         }
1936
1937         free(c->path);
1938         c->path = p;
1939
1940         strv_free(c->argv);
1941         c->argv = l;
1942
1943         return 0;
1944 }
1945
1946 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1947         [EXEC_INPUT_NULL] = "null",
1948         [EXEC_INPUT_TTY] = "tty",
1949         [EXEC_INPUT_TTY_FORCE] = "tty-force",
1950         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
1951         [EXEC_INPUT_SOCKET] = "socket"
1952 };
1953
1954 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
1955
1956 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
1957         [EXEC_OUTPUT_INHERIT] = "inherit",
1958         [EXEC_OUTPUT_NULL] = "null",
1959         [EXEC_OUTPUT_TTY] = "tty",
1960         [EXEC_OUTPUT_SYSLOG] = "syslog",
1961         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
1962         [EXEC_OUTPUT_KMSG] = "kmsg",
1963         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
1964         [EXEC_OUTPUT_SOCKET] = "socket"
1965 };
1966
1967 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
1968
1969 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
1970         [KILL_CONTROL_GROUP] = "control-group",
1971         [KILL_PROCESS] = "process",
1972         [KILL_NONE] = "none"
1973 };
1974
1975 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
1976
1977 static const char* const kill_who_table[_KILL_WHO_MAX] = {
1978         [KILL_MAIN] = "main",
1979         [KILL_CONTROL] = "control",
1980         [KILL_ALL] = "all"
1981 };
1982
1983 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);