src/execute.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU General Public License as published by
  10   the Free Software Foundation; either version 2 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   General Public License for more details.
  17
  18   You should have received a copy of the GNU General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <assert.h>
  23 #include <dirent.h>
  24 #include <errno.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include <string.h>
  28 #include <signal.h>
  29 #include <sys/socket.h>
  30 #include <sys/un.h>
  31 #include <sys/prctl.h>
  32 #include <linux/sched.h>
  33 #include <sys/types.h>
  34 #include <sys/stat.h>
  35 #include <grp.h>
  36 #include <pwd.h>
  37 #include <sys/mount.h>
  38 #include <linux/fs.h>
  39 #include <linux/oom.h>
  40
  41 #ifdef HAVE_PAM
  42 #include <security/pam_appl.h>
  43 #endif
  44
  45 #include "execute.h"
  46 #include "strv.h"
  47 #include "macro.h"
  48 #include "util.h"
  49 #include "log.h"
  50 #include "ioprio.h"
  51 #include "securebits.h"
  52 #include "cgroup.h"
  53 #include "namespace.h"
  54 #include "tcpwrap.h"
  55 #include "exit-status.h"
  56 #include "missing.h"
  57 #include "utmp-wtmp.h"
  58 #include "def.h"
  59
  60 /* This assumes there is a 'tty' group */
  61 #define TTY_MODE 0620
  62
  63 static int shift_fds(int fds[], unsigned n_fds) {
  64         int start, restart_from;
  65
  66         if (n_fds <= 0)
  67                 return 0;
  68
  69         /* Modifies the fds array! (sorts it) */
  70
  71         assert(fds);
  72
  73         start = 0;
  74         for (;;) {
  75                 int i;
  76
  77                 restart_from = -1;
  78
  79                 for (i = start; i < (int) n_fds; i++) {
  80                         int nfd;
  81
  82                         /* Already at right index? */
  83                         if (fds[i] == i+3)
  84                                 continue;
  85
  86                         if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
  87                                 return -errno;
  88
  89                         close_nointr_nofail(fds[i]);
  90                         fds[i] = nfd;
  91
  92                         /* Hmm, the fd we wanted isn't free? Then
  93                          * let's remember that and try again from here*/
  94                         if (nfd != i+3 && restart_from < 0)
  95                                 restart_from = i;
  96                 }
  97
  98                 if (restart_from < 0)
  99                         break;
 100
 101                 start = restart_from;
 102         }
 103
 104         return 0;
 105 }
 106
 107 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
 108         unsigned i;
 109         int r;
 110
 111         if (n_fds <= 0)
 112                 return 0;
 113
 114         assert(fds);
 115
 116         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
 117
 118         for (i = 0; i < n_fds; i++) {
 119
 120                 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
 121                         return r;
 122
 123                 /* We unconditionally drop FD_CLOEXEC from the fds,
 124                  * since after all we want to pass these fds to our
 125                  * children */
 126
 127                 if ((r = fd_cloexec(fds[i], false)) < 0)
 128                         return r;
 129         }
 130
 131         return 0;
 132 }
 133
 134 static const char *tty_path(const ExecContext *context) {
 135         assert(context);
 136
 137         if (context->tty_path)
 138                 return context->tty_path;
 139
 140         return "/dev/console";
 141 }
 142
 143 void exec_context_tty_reset(const ExecContext *context) {
 144         assert(context);
 145
 146         if (context->tty_vhangup)
 147                 terminal_vhangup(tty_path(context));
 148
 149         if (context->tty_reset)
 150                 reset_terminal(tty_path(context));
 151
 152         if (context->tty_vt_disallocate && context->tty_path)
 153                 vt_disallocate(context->tty_path);
 154 }
 155
 156 static int open_null_as(int flags, int nfd) {
 157         int fd, r;
 158
 159         assert(nfd >= 0);
 160
 161         if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
 162                 return -errno;
 163
 164         if (fd != nfd) {
 165                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 166                 close_nointr_nofail(fd);
 167         } else
 168                 r = nfd;
 169
 170         return r;
 171 }
 172
 173 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
 174         int fd, r;
 175         union {
 176                 struct sockaddr sa;
 177                 struct sockaddr_un un;
 178         } sa;
 179
 180         assert(context);
 181         assert(output < _EXEC_OUTPUT_MAX);
 182         assert(ident);
 183         assert(nfd >= 0);
 184
 185         if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
 186                 return -errno;
 187
 188         zero(sa);
 189         sa.sa.sa_family = AF_UNIX;
 190         strncpy(sa.un.sun_path, LOGGER_SOCKET, sizeof(sa.un.sun_path));
 191
 192         if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + sizeof(LOGGER_SOCKET) - 1) < 0) {
 193                 close_nointr_nofail(fd);
 194                 return -errno;
 195         }
 196
 197         if (shutdown(fd, SHUT_RD) < 0) {
 198                 close_nointr_nofail(fd);
 199                 return -errno;
 200         }
 201
 202         /* We speak a very simple protocol between log server
 203          * and client: one line for the log destination (kmsg
 204          * or syslog), followed by the priority field,
 205          * followed by the process name. Since we replaced
 206          * stdin/stderr we simple use stdio to write to
 207          * it. Note that we use stderr, to minimize buffer
 208          * flushing issues. */
 209
 210         dprintf(fd,
 211                 "%s\n"
 212                 "%i\n"
 213                 "%s\n"
 214                 "%i\n",
 215                 output == EXEC_OUTPUT_KMSG ?             "kmsg" :
 216                 output == EXEC_OUTPUT_KMSG_AND_CONSOLE ? "kmsg+console" :
 217                 output == EXEC_OUTPUT_SYSLOG ?           "syslog" :
 218                                                          "syslog+console",
 219                 context->syslog_priority,
 220                 context->syslog_identifier ? context->syslog_identifier : ident,
 221                 context->syslog_level_prefix);
 222
 223         if (fd != nfd) {
 224                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 225                 close_nointr_nofail(fd);
 226         } else
 227                 r = nfd;
 228
 229         return r;
 230 }
 231 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
 232         int fd, r;
 233
 234         assert(path);
 235         assert(nfd >= 0);
 236
 237         if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
 238                 return fd;
 239
 240         if (fd != nfd) {
 241                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 242                 close_nointr_nofail(fd);
 243         } else
 244                 r = nfd;
 245
 246         return r;
 247 }
 248
 249 static bool is_terminal_input(ExecInput i) {
 250         return
 251                 i == EXEC_INPUT_TTY ||
 252                 i == EXEC_INPUT_TTY_FORCE ||
 253                 i == EXEC_INPUT_TTY_FAIL;
 254 }
 255
 256 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
 257
 258         if (is_terminal_input(std_input) && !apply_tty_stdin)
 259                 return EXEC_INPUT_NULL;
 260
 261         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 262                 return EXEC_INPUT_NULL;
 263
 264         return std_input;
 265 }
 266
 267 static int fixup_output(ExecOutput std_output, int socket_fd) {
 268
 269         if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 270                 return EXEC_OUTPUT_INHERIT;
 271
 272         return std_output;
 273 }
 274
 275 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
 276         ExecInput i;
 277
 278         assert(context);
 279
 280         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 281
 282         switch (i) {
 283
 284         case EXEC_INPUT_NULL:
 285                 return open_null_as(O_RDONLY, STDIN_FILENO);
 286
 287         case EXEC_INPUT_TTY:
 288         case EXEC_INPUT_TTY_FORCE:
 289         case EXEC_INPUT_TTY_FAIL: {
 290                 int fd, r;
 291
 292                 if ((fd = acquire_terminal(
 293                                      tty_path(context),
 294                                      i == EXEC_INPUT_TTY_FAIL,
 295                                      i == EXEC_INPUT_TTY_FORCE,
 296                                      false)) < 0)
 297                         return fd;
 298
 299                 if (fd != STDIN_FILENO) {
 300                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 301                         close_nointr_nofail(fd);
 302                 } else
 303                         r = STDIN_FILENO;
 304
 305                 return r;
 306         }
 307
 308         case EXEC_INPUT_SOCKET:
 309                 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 310
 311         default:
 312                 assert_not_reached("Unknown input type");
 313         }
 314 }
 315
 316 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 317         ExecOutput o;
 318         ExecInput i;
 319
 320         assert(context);
 321         assert(ident);
 322
 323         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 324         o = fixup_output(context->std_output, socket_fd);
 325
 326         /* This expects the input is already set up */
 327
 328         switch (o) {
 329
 330         case EXEC_OUTPUT_INHERIT:
 331
 332                 /* If input got downgraded, inherit the original value */
 333                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 334                         return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 335
 336                 /* If the input is connected to anything that's not a /dev/null, inherit that... */
 337                 if (i != EXEC_INPUT_NULL)
 338                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 339
 340                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 341                 if (getppid() != 1)
 342                         return STDOUT_FILENO;
 343
 344                 /* We need to open /dev/null here anew, to get the
 345                  * right access mode. So we fall through */
 346
 347         case EXEC_OUTPUT_NULL:
 348                 return open_null_as(O_WRONLY, STDOUT_FILENO);
 349
 350         case EXEC_OUTPUT_TTY:
 351                 if (is_terminal_input(i))
 352                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 353
 354                 /* We don't reset the terminal if this is just about output */
 355                 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 356
 357         case EXEC_OUTPUT_SYSLOG:
 358         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 359         case EXEC_OUTPUT_KMSG:
 360         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 361                 return connect_logger_as(context, o, ident, STDOUT_FILENO);
 362
 363         case EXEC_OUTPUT_SOCKET:
 364                 assert(socket_fd >= 0);
 365                 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 366
 367         default:
 368                 assert_not_reached("Unknown output type");
 369         }
 370 }
 371
 372 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 373         ExecOutput o, e;
 374         ExecInput i;
 375
 376         assert(context);
 377         assert(ident);
 378
 379         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 380         o = fixup_output(context->std_output, socket_fd);
 381         e = fixup_output(context->std_error, socket_fd);
 382
 383         /* This expects the input and output are already set up */
 384
 385         /* Don't change the stderr file descriptor if we inherit all
 386          * the way and are not on a tty */
 387         if (e == EXEC_OUTPUT_INHERIT &&
 388             o == EXEC_OUTPUT_INHERIT &&
 389             i == EXEC_INPUT_NULL &&
 390             !is_terminal_input(context->std_input) &&
 391             getppid () != 1)
 392                 return STDERR_FILENO;
 393
 394         /* Duplicate from stdout if possible */
 395         if (e == o || e == EXEC_OUTPUT_INHERIT)
 396                 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 397
 398         switch (e) {
 399
 400         case EXEC_OUTPUT_NULL:
 401                 return open_null_as(O_WRONLY, STDERR_FILENO);
 402
 403         case EXEC_OUTPUT_TTY:
 404                 if (is_terminal_input(i))
 405                         return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 406
 407                 /* We don't reset the terminal if this is just about output */
 408                 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
 409
 410         case EXEC_OUTPUT_SYSLOG:
 411         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 412         case EXEC_OUTPUT_KMSG:
 413         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 414                 return connect_logger_as(context, e, ident, STDERR_FILENO);
 415
 416         case EXEC_OUTPUT_SOCKET:
 417                 assert(socket_fd >= 0);
 418                 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 419
 420         default:
 421                 assert_not_reached("Unknown error type");
 422         }
 423 }
 424
 425 static int chown_terminal(int fd, uid_t uid) {
 426         struct stat st;
 427
 428         assert(fd >= 0);
 429
 430         /* This might fail. What matters are the results. */
 431         (void) fchown(fd, uid, -1);
 432         (void) fchmod(fd, TTY_MODE);
 433
 434         if (fstat(fd, &st) < 0)
 435                 return -errno;
 436
 437         if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
 438                 return -EPERM;
 439
 440         return 0;
 441 }
 442
 443 static int setup_confirm_stdio(const ExecContext *context,
 444                                int *_saved_stdin,
 445                                int *_saved_stdout) {
 446         int fd = -1, saved_stdin, saved_stdout = -1, r;
 447
 448         assert(context);
 449         assert(_saved_stdin);
 450         assert(_saved_stdout);
 451
 452         /* This returns positive EXIT_xxx return values instead of
 453          * negative errno style values! */
 454
 455         if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
 456                 return EXIT_STDIN;
 457
 458         if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
 459                 r = EXIT_STDOUT;
 460                 goto fail;
 461         }
 462
 463         if ((fd = acquire_terminal(
 464                              tty_path(context),
 465                              context->std_input == EXEC_INPUT_TTY_FAIL,
 466                              context->std_input == EXEC_INPUT_TTY_FORCE,
 467                              false)) < 0) {
 468                 r = EXIT_STDIN;
 469                 goto fail;
 470         }
 471
 472         if (chown_terminal(fd, getuid()) < 0) {
 473                 r = EXIT_STDIN;
 474                 goto fail;
 475         }
 476
 477         if (dup2(fd, STDIN_FILENO) < 0) {
 478                 r = EXIT_STDIN;
 479                 goto fail;
 480         }
 481
 482         if (dup2(fd, STDOUT_FILENO) < 0) {
 483                 r = EXIT_STDOUT;
 484                 goto fail;
 485         }
 486
 487         if (fd >= 2)
 488                 close_nointr_nofail(fd);
 489
 490         *_saved_stdin = saved_stdin;
 491         *_saved_stdout = saved_stdout;
 492
 493         return 0;
 494
 495 fail:
 496         if (saved_stdout >= 0)
 497                 close_nointr_nofail(saved_stdout);
 498
 499         if (saved_stdin >= 0)
 500                 close_nointr_nofail(saved_stdin);
 501
 502         if (fd >= 0)
 503                 close_nointr_nofail(fd);
 504
 505         return r;
 506 }
 507
 508 static int restore_confirm_stdio(const ExecContext *context,
 509                                  int *saved_stdin,
 510                                  int *saved_stdout,
 511                                  bool *keep_stdin,
 512                                  bool *keep_stdout) {
 513
 514         assert(context);
 515         assert(saved_stdin);
 516         assert(*saved_stdin >= 0);
 517         assert(saved_stdout);
 518         assert(*saved_stdout >= 0);
 519
 520         /* This returns positive EXIT_xxx return values instead of
 521          * negative errno style values! */
 522
 523         if (is_terminal_input(context->std_input)) {
 524
 525                 /* The service wants terminal input. */
 526
 527                 *keep_stdin = true;
 528                 *keep_stdout =
 529                         context->std_output == EXEC_OUTPUT_INHERIT ||
 530                         context->std_output == EXEC_OUTPUT_TTY;
 531
 532         } else {
 533                 /* If the service doesn't want a controlling terminal,
 534                  * then we need to get rid entirely of what we have
 535                  * already. */
 536
 537                 if (release_terminal() < 0)
 538                         return EXIT_STDIN;
 539
 540                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 541                         return EXIT_STDIN;
 542
 543                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 544                         return EXIT_STDOUT;
 545
 546                 *keep_stdout = *keep_stdin = false;
 547         }
 548
 549         return 0;
 550 }
 551
 552 static int get_group_creds(const char *groupname, gid_t *gid) {
 553         struct group *g;
 554         unsigned long lu;
 555
 556         assert(groupname);
 557         assert(gid);
 558
 559         /* We enforce some special rules for gid=0: in order to avoid
 560          * NSS lookups for root we hardcode its data. */
 561
 562         if (streq(groupname, "root") || streq(groupname, "0")) {
 563                 *gid = 0;
 564                 return 0;
 565         }
 566
 567         if (safe_atolu(groupname, &lu) >= 0) {
 568                 errno = 0;
 569                 g = getgrgid((gid_t) lu);
 570         } else {
 571                 errno = 0;
 572                 g = getgrnam(groupname);
 573         }
 574
 575         if (!g)
 576                 return errno != 0 ? -errno : -ESRCH;
 577
 578         *gid = g->gr_gid;
 579         return 0;
 580 }
 581
 582 static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
 583         struct passwd *p;
 584         unsigned long lu;
 585
 586         assert(username);
 587         assert(*username);
 588         assert(uid);
 589         assert(gid);
 590         assert(home);
 591
 592         /* We enforce some special rules for uid=0: in order to avoid
 593          * NSS lookups for root we hardcode its data. */
 594
 595         if (streq(*username, "root") || streq(*username, "0")) {
 596                 *username = "root";
 597                 *uid = 0;
 598                 *gid = 0;
 599                 *home = "/root";
 600                 return 0;
 601         }
 602
 603         if (safe_atolu(*username, &lu) >= 0) {
 604                 errno = 0;
 605                 p = getpwuid((uid_t) lu);
 606
 607                 /* If there are multiple users with the same id, make
 608                  * sure to leave $USER to the configured value instead
 609                  * of the first occurrence in the database. However if
 610                  * the uid was configured by a numeric uid, then let's
 611                  * pick the real username from /etc/passwd. */
 612                 if (*username && p)
 613                         *username = p->pw_name;
 614         } else {
 615                 errno = 0;
 616                 p = getpwnam(*username);
 617         }
 618
 619         if (!p)
 620                 return errno != 0 ? -errno : -ESRCH;
 621
 622         *uid = p->pw_uid;
 623         *gid = p->pw_gid;
 624         *home = p->pw_dir;
 625         return 0;
 626 }
 627
 628 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
 629         bool keep_groups = false;
 630         int r;
 631
 632         assert(context);
 633
 634         /* Lookup and set GID and supplementary group list. Here too
 635          * we avoid NSS lookups for gid=0. */
 636
 637         if (context->group || username) {
 638
 639                 if (context->group)
 640                         if ((r = get_group_creds(context->group, &gid)) < 0)
 641                                 return r;
 642
 643                 /* First step, initialize groups from /etc/groups */
 644                 if (username && gid != 0) {
 645                         if (initgroups(username, gid) < 0)
 646                                 return -errno;
 647
 648                         keep_groups = true;
 649                 }
 650
 651                 /* Second step, set our gids */
 652                 if (setresgid(gid, gid, gid) < 0)
 653                         return -errno;
 654         }
 655
 656         if (context->supplementary_groups) {
 657                 int ngroups_max, k;
 658                 gid_t *gids;
 659                 char **i;
 660
 661                 /* Final step, initialize any manually set supplementary groups */
 662                 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
 663
 664                 if (!(gids = new(gid_t, ngroups_max)))
 665                         return -ENOMEM;
 666
 667                 if (keep_groups) {
 668                         if ((k = getgroups(ngroups_max, gids)) < 0) {
 669                                 free(gids);
 670                                 return -errno;
 671                         }
 672                 } else
 673                         k = 0;
 674
 675                 STRV_FOREACH(i, context->supplementary_groups) {
 676
 677                         if (k >= ngroups_max) {
 678                                 free(gids);
 679                                 return -E2BIG;
 680                         }
 681
 682                         if ((r = get_group_creds(*i, gids+k)) < 0) {
 683                                 free(gids);
 684                                 return r;
 685                         }
 686
 687                         k++;
 688                 }
 689
 690                 if (setgroups(k, gids) < 0) {
 691                         free(gids);
 692                         return -errno;
 693                 }
 694
 695                 free(gids);
 696         }
 697
 698         return 0;
 699 }
 700
 701 static int enforce_user(const ExecContext *context, uid_t uid) {
 702         int r;
 703         assert(context);
 704
 705         /* Sets (but doesn't lookup) the uid and make sure we keep the
 706          * capabilities while doing so. */
 707
 708         if (context->capabilities) {
 709                 cap_t d;
 710                 static const cap_value_t bits[] = {
 711                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
 712                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
 713                 };
 714
 715                 /* First step: If we need to keep capabilities but
 716                  * drop privileges we need to make sure we keep our
 717                  * caps, whiel we drop privileges. */
 718                 if (uid != 0) {
 719                         int sb = context->secure_bits|SECURE_KEEP_CAPS;
 720
 721                         if (prctl(PR_GET_SECUREBITS) != sb)
 722                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 723                                         return -errno;
 724                 }
 725
 726                 /* Second step: set the capabilities. This will reduce
 727                  * the capabilities to the minimum we need. */
 728
 729                 if (!(d = cap_dup(context->capabilities)))
 730                         return -errno;
 731
 732                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
 733                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
 734                         r = -errno;
 735                         cap_free(d);
 736                         return r;
 737                 }
 738
 739                 if (cap_set_proc(d) < 0) {
 740                         r = -errno;
 741                         cap_free(d);
 742                         return r;
 743                 }
 744
 745                 cap_free(d);
 746         }
 747
 748         /* Third step: actually set the uids */
 749         if (setresuid(uid, uid, uid) < 0)
 750                 return -errno;
 751
 752         /* At this point we should have all necessary capabilities but
 753            are otherwise a normal user. However, the caps might got
 754            corrupted due to the setresuid() so we need clean them up
 755            later. This is done outside of this call. */
 756
 757         return 0;
 758 }
 759
 760 #ifdef HAVE_PAM
 761
 762 static int null_conv(
 763                 int num_msg,
 764                 const struct pam_message **msg,
 765                 struct pam_response **resp,
 766                 void *appdata_ptr) {
 767
 768         /* We don't support conversations */
 769
 770         return PAM_CONV_ERR;
 771 }
 772
 773 static int setup_pam(
 774                 const char *name,
 775                 const char *user,
 776                 const char *tty,
 777                 char ***pam_env,
 778                 int fds[], unsigned n_fds) {
 779
 780         static const struct pam_conv conv = {
 781                 .conv = null_conv,
 782                 .appdata_ptr = NULL
 783         };
 784
 785         pam_handle_t *handle = NULL;
 786         sigset_t ss, old_ss;
 787         int pam_code = PAM_SUCCESS;
 788         char **e = NULL;
 789         bool close_session = false;
 790         pid_t pam_pid = 0, parent_pid;
 791
 792         assert(name);
 793         assert(user);
 794         assert(pam_env);
 795
 796         /* We set up PAM in the parent process, then fork. The child
 797          * will then stay around until killed via PR_GET_PDEATHSIG or
 798          * systemd via the cgroup logic. It will then remove the PAM
 799          * session again. The parent process will exec() the actual
 800          * daemon. We do things this way to ensure that the main PID
 801          * of the daemon is the one we initially fork()ed. */
 802
 803         if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
 804                 handle = NULL;
 805                 goto fail;
 806         }
 807
 808         if (tty)
 809                 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
 810                         goto fail;
 811
 812         if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
 813                 goto fail;
 814
 815         if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
 816                 goto fail;
 817
 818         close_session = true;
 819
 820         if ((!(e = pam_getenvlist(handle)))) {
 821                 pam_code = PAM_BUF_ERR;
 822                 goto fail;
 823         }
 824
 825         /* Block SIGTERM, so that we know that it won't get lost in
 826          * the child */
 827         if (sigemptyset(&ss) < 0 ||
 828             sigaddset(&ss, SIGTERM) < 0 ||
 829             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 830                 goto fail;
 831
 832         parent_pid = getpid();
 833
 834         if ((pam_pid = fork()) < 0)
 835                 goto fail;
 836
 837         if (pam_pid == 0) {
 838                 int sig;
 839                 int r = EXIT_PAM;
 840
 841                 /* The child's job is to reset the PAM session on
 842                  * termination */
 843
 844                 /* This string must fit in 10 chars (i.e. the length
 845                  * of "/sbin/init") */
 846                 rename_process("sd:pam");
 847
 848                 /* Make sure we don't keep open the passed fds in this
 849                 child. We assume that otherwise only those fds are
 850                 open here that have been opened by PAM. */
 851                 close_many(fds, n_fds);
 852
 853                 /* Wait until our parent died. This will most likely
 854                  * not work since the kernel does not allow
 855                  * unprivileged parents kill their privileged children
 856                  * this way. We rely on the control groups kill logic
 857                  * to do the rest for us. */
 858                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 859                         goto child_finish;
 860
 861                 /* Check if our parent process might already have
 862                  * died? */
 863                 if (getppid() == parent_pid) {
 864                         if (sigwait(&ss, &sig) < 0)
 865                                 goto child_finish;
 866
 867                         assert(sig == SIGTERM);
 868                 }
 869
 870                 /* Only if our parent died we'll end the session */
 871                 if (getppid() != parent_pid)
 872                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
 873                                 goto child_finish;
 874
 875                 r = 0;
 876
 877         child_finish:
 878                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 879                 _exit(r);
 880         }
 881
 882         /* If the child was forked off successfully it will do all the
 883          * cleanups, so forget about the handle here. */
 884         handle = NULL;
 885
 886         /* Unblock SIGTERM again in the parent */
 887         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 888                 goto fail;
 889
 890         /* We close the log explicitly here, since the PAM modules
 891          * might have opened it, but we don't want this fd around. */
 892         closelog();
 893
 894         return 0;
 895
 896 fail:
 897         if (handle) {
 898                 if (close_session)
 899                         pam_code = pam_close_session(handle, PAM_DATA_SILENT);
 900
 901                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 902         }
 903
 904         strv_free(e);
 905
 906         closelog();
 907
 908         if (pam_pid > 1) {
 909                 kill(pam_pid, SIGTERM);
 910                 kill(pam_pid, SIGCONT);
 911         }
 912
 913         return EXIT_PAM;
 914 }
 915 #endif
 916
 917 static int do_capability_bounding_set_drop(uint64_t drop) {
 918         unsigned long i;
 919         cap_t old_cap = NULL, new_cap = NULL;
 920         cap_flag_value_t fv;
 921         int r;
 922
 923         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 924          * in the effective set (yes, the kernel drops that when
 925          * executing init!), so get it back temporarily so that we can
 926          * call PR_CAPBSET_DROP. */
 927
 928         old_cap = cap_get_proc();
 929         if (!old_cap)
 930                 return -errno;
 931
 932         if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
 933                 r = -errno;
 934                 goto finish;
 935         }
 936
 937         if (fv != CAP_SET) {
 938                 static const cap_value_t v = CAP_SETPCAP;
 939
 940                 new_cap = cap_dup(old_cap);
 941                 if (!new_cap) {
 942                         r = -errno;
 943                         goto finish;
 944                 }
 945
 946                 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
 947                         r = -errno;
 948                         goto finish;
 949                 }
 950
 951                 if (cap_set_proc(new_cap) < 0) {
 952                         r = -errno;
 953                         goto finish;
 954                 }
 955         }
 956
 957         for (i = 0; i <= MAX(63LU, (unsigned long) CAP_LAST_CAP); i++)
 958                 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
 959                         if (prctl(PR_CAPBSET_DROP, i) < 0) {
 960                                 if (errno == EINVAL)
 961                                         break;
 962
 963                                 r = -errno;
 964                                 goto finish;
 965                         }
 966                 }
 967
 968         r = 0;
 969
 970 finish:
 971         if (new_cap)
 972                 cap_free(new_cap);
 973
 974         if (old_cap) {
 975                 cap_set_proc(old_cap);
 976                 cap_free(old_cap);
 977         }
 978
 979         return r;
 980 }
 981
 982 int exec_spawn(ExecCommand *command,
 983                char **argv,
 984                const ExecContext *context,
 985                int fds[], unsigned n_fds,
 986                char **environment,
 987                bool apply_permissions,
 988                bool apply_chroot,
 989                bool apply_tty_stdin,
 990                bool confirm_spawn,
 991                CGroupBonding *cgroup_bondings,
 992                pid_t *ret) {
 993
 994         pid_t pid;
 995         int r;
 996         char *line;
 997         int socket_fd;
 998         char **files_env = NULL;
 999
1000         assert(command);
1001         assert(context);
1002         assert(ret);
1003         assert(fds || n_fds <= 0);
1004
1005         if (context->std_input == EXEC_INPUT_SOCKET ||
1006             context->std_output == EXEC_OUTPUT_SOCKET ||
1007             context->std_error == EXEC_OUTPUT_SOCKET) {
1008
1009                 if (n_fds != 1)
1010                         return -EINVAL;
1011
1012                 socket_fd = fds[0];
1013
1014                 fds = NULL;
1015                 n_fds = 0;
1016         } else
1017                 socket_fd = -1;
1018
1019         if ((r = exec_context_load_environment(context, &files_env)) < 0) {
1020                 log_error("Failed to load environment files: %s", strerror(-r));
1021                 return r;
1022         }
1023
1024         if (!argv)
1025                 argv = command->argv;
1026
1027         if (!(line = exec_command_line(argv))) {
1028                 r = -ENOMEM;
1029                 goto fail_parent;
1030         }
1031
1032         log_debug("About to execute: %s", line);
1033         free(line);
1034
1035         if (cgroup_bondings)
1036                 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
1037                         goto fail_parent;
1038
1039         if ((pid = fork()) < 0) {
1040                 r = -errno;
1041                 goto fail_parent;
1042         }
1043
1044         if (pid == 0) {
1045                 int i;
1046                 sigset_t ss;
1047                 const char *username = NULL, *home = NULL;
1048                 uid_t uid = (uid_t) -1;
1049                 gid_t gid = (gid_t) -1;
1050                 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1051                 unsigned n_env = 0;
1052                 int saved_stdout = -1, saved_stdin = -1;
1053                 bool keep_stdout = false, keep_stdin = false;
1054
1055                 /* child */
1056
1057                 /* This string must fit in 10 chars (i.e. the length
1058                  * of "/sbin/init") */
1059                 rename_process("sd.exec");
1060
1061                 /* We reset exactly these signals, since they are the
1062                  * only ones we set to SIG_IGN in the main daemon. All
1063                  * others we leave untouched because we set them to
1064                  * SIG_DFL or a valid handler initially, both of which
1065                  * will be demoted to SIG_DFL. */
1066                 default_signals(SIGNALS_CRASH_HANDLER,
1067                                 SIGNALS_IGNORE, -1);
1068
1069                 if (sigemptyset(&ss) < 0 ||
1070                     sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1071                         r = EXIT_SIGNAL_MASK;
1072                         goto fail_child;
1073                 }
1074
1075                 /* Close sockets very early to make sure we don't
1076                  * block init reexecution because it cannot bind its
1077                  * sockets */
1078                 if (close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1079                                   socket_fd >= 0 ? 1 : n_fds) < 0) {
1080                         r = EXIT_FDS;
1081                         goto fail_child;
1082                 }
1083
1084                 if (!context->same_pgrp)
1085                         if (setsid() < 0) {
1086                                 r = EXIT_SETSID;
1087                                 goto fail_child;
1088                         }
1089
1090                 if (context->tcpwrap_name) {
1091                         if (socket_fd >= 0)
1092                                 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1093                                         r = EXIT_TCPWRAP;
1094                                         goto fail_child;
1095                                 }
1096
1097                         for (i = 0; i < (int) n_fds; i++) {
1098                                 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1099                                         r = EXIT_TCPWRAP;
1100                                         goto fail_child;
1101                                 }
1102                         }
1103                 }
1104
1105                 exec_context_tty_reset(context);
1106
1107                 /* We skip the confirmation step if we shall not apply the TTY */
1108                 if (confirm_spawn &&
1109                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1110                         char response;
1111
1112                         /* Set up terminal for the question */
1113                         if ((r = setup_confirm_stdio(context,
1114                                                      &saved_stdin, &saved_stdout)))
1115                                 goto fail_child;
1116
1117                         /* Now ask the question. */
1118                         if (!(line = exec_command_line(argv))) {
1119                                 r = EXIT_MEMORY;
1120                                 goto fail_child;
1121                         }
1122
1123                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1124                         free(line);
1125
1126                         if (r < 0 || response == 'n') {
1127                                 r = EXIT_CONFIRM;
1128                                 goto fail_child;
1129                         } else if (response == 's') {
1130                                 r = 0;
1131                                 goto fail_child;
1132                         }
1133
1134                         /* Release terminal for the question */
1135                         if ((r = restore_confirm_stdio(context,
1136                                                        &saved_stdin, &saved_stdout,
1137                                                        &keep_stdin, &keep_stdout)))
1138                                 goto fail_child;
1139                 }
1140
1141                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1142                  * must sure to drop O_NONBLOCK */
1143                 if (socket_fd >= 0)
1144                         fd_nonblock(socket_fd, false);
1145
1146                 if (!keep_stdin)
1147                         if (setup_input(context, socket_fd, apply_tty_stdin) < 0) {
1148                                 r = EXIT_STDIN;
1149                                 goto fail_child;
1150                         }
1151
1152                 if (!keep_stdout)
1153                         if (setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1154                                 r = EXIT_STDOUT;
1155                                 goto fail_child;
1156                         }
1157
1158                 if (setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1159                         r = EXIT_STDERR;
1160                         goto fail_child;
1161                 }
1162
1163                 if (cgroup_bondings)
1164                         if (cgroup_bonding_install_list(cgroup_bondings, 0) < 0) {
1165                                 r = EXIT_CGROUP;
1166                                 goto fail_child;
1167                         }
1168
1169                 if (context->oom_score_adjust_set) {
1170                         char t[16];
1171
1172                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1173                         char_array_0(t);
1174
1175                         if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1176                                 /* Compatibility with Linux <= 2.6.35 */
1177
1178                                 int adj;
1179
1180                                 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1181                                 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1182
1183                                 snprintf(t, sizeof(t), "%i", adj);
1184                                 char_array_0(t);
1185
1186                                 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1187                                     && errno != EACCES) {
1188                                         r = EXIT_OOM_ADJUST;
1189                                         goto fail_child;
1190                                 }
1191                         }
1192                 }
1193
1194                 if (context->nice_set)
1195                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1196                                 r = EXIT_NICE;
1197                                 goto fail_child;
1198                         }
1199
1200                 if (context->cpu_sched_set) {
1201                         struct sched_param param;
1202
1203                         zero(param);
1204                         param.sched_priority = context->cpu_sched_priority;
1205
1206                         if (sched_setscheduler(0, context->cpu_sched_policy |
1207                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
1208                                 r = EXIT_SETSCHEDULER;
1209                                 goto fail_child;
1210                         }
1211                 }
1212
1213                 if (context->cpuset)
1214                         if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1215                                 r = EXIT_CPUAFFINITY;
1216                                 goto fail_child;
1217                         }
1218
1219                 if (context->ioprio_set)
1220                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1221                                 r = EXIT_IOPRIO;
1222                                 goto fail_child;
1223                         }
1224
1225                 if (context->timer_slack_nsec_set)
1226                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1227                                 r = EXIT_TIMERSLACK;
1228                                 goto fail_child;
1229                         }
1230
1231                 if (context->utmp_id)
1232                         utmp_put_init_process(0, context->utmp_id, getpid(), getsid(0), context->tty_path);
1233
1234                 if (context->user) {
1235                         username = context->user;
1236                         if (get_user_creds(&username, &uid, &gid, &home) < 0) {
1237                                 r = EXIT_USER;
1238                                 goto fail_child;
1239                         }
1240
1241                         if (is_terminal_input(context->std_input))
1242                                 if (chown_terminal(STDIN_FILENO, uid) < 0) {
1243                                         r = EXIT_STDIN;
1244                                         goto fail_child;
1245                                 }
1246
1247                         if (cgroup_bondings && context->control_group_modify)
1248                                 if (cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid) < 0 ||
1249                                     cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid) < 0) {
1250                                         r = EXIT_CGROUP;
1251                                         goto fail_child;
1252                                 }
1253                 }
1254
1255                 if (apply_permissions)
1256                         if (enforce_groups(context, username, uid) < 0) {
1257                                 r = EXIT_GROUP;
1258                                 goto fail_child;
1259                         }
1260
1261                 umask(context->umask);
1262
1263 #ifdef HAVE_PAM
1264                 if (context->pam_name && username) {
1265                         if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) < 0) {
1266                                 r = EXIT_PAM;
1267                                 goto fail_child;
1268                         }
1269                 }
1270 #endif
1271
1272                 if (strv_length(context->read_write_dirs) > 0 ||
1273                     strv_length(context->read_only_dirs) > 0 ||
1274                     strv_length(context->inaccessible_dirs) > 0 ||
1275                     context->mount_flags != MS_SHARED ||
1276                     context->private_tmp)
1277                         if ((r = setup_namespace(
1278                                              context->read_write_dirs,
1279                                              context->read_only_dirs,
1280                                              context->inaccessible_dirs,
1281                                              context->private_tmp,
1282                                              context->mount_flags)) < 0)
1283                                 goto fail_child;
1284
1285                 if (apply_chroot) {
1286                         if (context->root_directory)
1287                                 if (chroot(context->root_directory) < 0) {
1288                                         r = EXIT_CHROOT;
1289                                         goto fail_child;
1290                                 }
1291
1292                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1293                                 r = EXIT_CHDIR;
1294                                 goto fail_child;
1295                         }
1296                 } else {
1297
1298                         char *d;
1299
1300                         if (asprintf(&d, "%s/%s",
1301                                      context->root_directory ? context->root_directory : "",
1302                                      context->working_directory ? context->working_directory : "") < 0) {
1303                                 r = EXIT_MEMORY;
1304                                 goto fail_child;
1305                         }
1306
1307                         if (chdir(d) < 0) {
1308                                 free(d);
1309                                 r = EXIT_CHDIR;
1310                                 goto fail_child;
1311                         }
1312
1313                         free(d);
1314                 }
1315
1316                 /* We repeat the fd closing here, to make sure that
1317                  * nothing is leaked from the PAM modules */
1318                 if (close_all_fds(fds, n_fds) < 0 ||
1319                     shift_fds(fds, n_fds) < 0 ||
1320                     flags_fds(fds, n_fds, context->non_blocking) < 0) {
1321                         r = EXIT_FDS;
1322                         goto fail_child;
1323                 }
1324
1325                 if (apply_permissions) {
1326
1327                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1328                                 if (!context->rlimit[i])
1329                                         continue;
1330
1331                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1332                                         r = EXIT_LIMITS;
1333                                         goto fail_child;
1334                                 }
1335                         }
1336
1337                         if (context->capability_bounding_set_drop)
1338                                 if (do_capability_bounding_set_drop(context->capability_bounding_set_drop) < 0) {
1339                                         r = EXIT_CAPABILITIES;
1340                                         goto fail_child;
1341                                 }
1342
1343                         if (context->user)
1344                                 if (enforce_user(context, uid) < 0) {
1345                                         r = EXIT_USER;
1346                                         goto fail_child;
1347                                 }
1348
1349                         /* PR_GET_SECUREBITS is not privileged, while
1350                          * PR_SET_SECUREBITS is. So to suppress
1351                          * potential EPERMs we'll try not to call
1352                          * PR_SET_SECUREBITS unless necessary. */
1353                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1354                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1355                                         r = EXIT_SECUREBITS;
1356                                         goto fail_child;
1357                                 }
1358
1359                         if (context->capabilities)
1360                                 if (cap_set_proc(context->capabilities) < 0) {
1361                                         r = EXIT_CAPABILITIES;
1362                                         goto fail_child;
1363                                 }
1364                 }
1365
1366                 if (!(our_env = new0(char*, 7))) {
1367                         r = EXIT_MEMORY;
1368                         goto fail_child;
1369                 }
1370
1371                 if (n_fds > 0)
1372                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1373                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1374                                 r = EXIT_MEMORY;
1375                                 goto fail_child;
1376                         }
1377
1378                 if (home)
1379                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1380                                 r = EXIT_MEMORY;
1381                                 goto fail_child;
1382                         }
1383
1384                 if (username)
1385                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1386                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1387                                 r = EXIT_MEMORY;
1388                                 goto fail_child;
1389                         }
1390
1391                 if (is_terminal_input(context->std_input) ||
1392                     context->std_output == EXEC_OUTPUT_TTY ||
1393                     context->std_error == EXEC_OUTPUT_TTY)
1394                         if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1395                                 r = EXIT_MEMORY;
1396                                 goto fail_child;
1397                         }
1398
1399                 assert(n_env <= 7);
1400
1401                 if (!(final_env = strv_env_merge(
1402                                       5,
1403                                       environment,
1404                                       our_env,
1405                                       context->environment,
1406                                       files_env,
1407                                       pam_env,
1408                                       NULL))) {
1409                         r = EXIT_MEMORY;
1410                         goto fail_child;
1411                 }
1412
1413                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1414                         r = EXIT_MEMORY;
1415                         goto fail_child;
1416                 }
1417
1418                 final_env = strv_env_clean(final_env);
1419
1420                 execve(command->path, final_argv, final_env);
1421                 r = EXIT_EXEC;
1422
1423         fail_child:
1424                 strv_free(our_env);
1425                 strv_free(final_env);
1426                 strv_free(pam_env);
1427                 strv_free(files_env);
1428                 strv_free(final_argv);
1429
1430                 if (saved_stdin >= 0)
1431                         close_nointr_nofail(saved_stdin);
1432
1433                 if (saved_stdout >= 0)
1434                         close_nointr_nofail(saved_stdout);
1435
1436                 _exit(r);
1437         }
1438
1439         strv_free(files_env);
1440
1441         /* We add the new process to the cgroup both in the child (so
1442          * that we can be sure that no user code is ever executed
1443          * outside of the cgroup) and in the parent (so that we can be
1444          * sure that when we kill the cgroup the process will be
1445          * killed too). */
1446         if (cgroup_bondings)
1447                 cgroup_bonding_install_list(cgroup_bondings, pid);
1448
1449         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1450
1451         exec_status_start(&command->exec_status, pid);
1452
1453         *ret = pid;
1454         return 0;
1455
1456 fail_parent:
1457         strv_free(files_env);
1458
1459         return r;
1460 }
1461
1462 void exec_context_init(ExecContext *c) {
1463         assert(c);
1464
1465         c->umask = 0002;
1466         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1467         c->cpu_sched_policy = SCHED_OTHER;
1468         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1469         c->syslog_level_prefix = true;
1470         c->mount_flags = MS_SHARED;
1471         c->kill_signal = SIGTERM;
1472         c->send_sigkill = true;
1473 }
1474
1475 void exec_context_done(ExecContext *c) {
1476         unsigned l;
1477
1478         assert(c);
1479
1480         strv_free(c->environment);
1481         c->environment = NULL;
1482
1483         strv_free(c->environment_files);
1484         c->environment_files = NULL;
1485
1486         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1487                 free(c->rlimit[l]);
1488                 c->rlimit[l] = NULL;
1489         }
1490
1491         free(c->working_directory);
1492         c->working_directory = NULL;
1493         free(c->root_directory);
1494         c->root_directory = NULL;
1495
1496         free(c->tty_path);
1497         c->tty_path = NULL;
1498
1499         free(c->tcpwrap_name);
1500         c->tcpwrap_name = NULL;
1501
1502         free(c->syslog_identifier);
1503         c->syslog_identifier = NULL;
1504
1505         free(c->user);
1506         c->user = NULL;
1507
1508         free(c->group);
1509         c->group = NULL;
1510
1511         strv_free(c->supplementary_groups);
1512         c->supplementary_groups = NULL;
1513
1514         free(c->pam_name);
1515         c->pam_name = NULL;
1516
1517         if (c->capabilities) {
1518                 cap_free(c->capabilities);
1519                 c->capabilities = NULL;
1520         }
1521
1522         strv_free(c->read_only_dirs);
1523         c->read_only_dirs = NULL;
1524
1525         strv_free(c->read_write_dirs);
1526         c->read_write_dirs = NULL;
1527
1528         strv_free(c->inaccessible_dirs);
1529         c->inaccessible_dirs = NULL;
1530
1531         if (c->cpuset)
1532                 CPU_FREE(c->cpuset);
1533
1534         free(c->utmp_id);
1535         c->utmp_id = NULL;
1536 }
1537
1538 void exec_command_done(ExecCommand *c) {
1539         assert(c);
1540
1541         free(c->path);
1542         c->path = NULL;
1543
1544         strv_free(c->argv);
1545         c->argv = NULL;
1546 }
1547
1548 void exec_command_done_array(ExecCommand *c, unsigned n) {
1549         unsigned i;
1550
1551         for (i = 0; i < n; i++)
1552                 exec_command_done(c+i);
1553 }
1554
1555 void exec_command_free_list(ExecCommand *c) {
1556         ExecCommand *i;
1557
1558         while ((i = c)) {
1559                 LIST_REMOVE(ExecCommand, command, c, i);
1560                 exec_command_done(i);
1561                 free(i);
1562         }
1563 }
1564
1565 void exec_command_free_array(ExecCommand **c, unsigned n) {
1566         unsigned i;
1567
1568         for (i = 0; i < n; i++) {
1569                 exec_command_free_list(c[i]);
1570                 c[i] = NULL;
1571         }
1572 }
1573
1574 int exec_context_load_environment(const ExecContext *c, char ***l) {
1575         char **i, **r = NULL;
1576
1577         assert(c);
1578         assert(l);
1579
1580         STRV_FOREACH(i, c->environment_files) {
1581                 char *fn;
1582                 int k;
1583                 bool ignore = false;
1584                 char **p;
1585
1586                 fn = *i;
1587
1588                 if (fn[0] == '-') {
1589                         ignore = true;
1590                         fn ++;
1591                 }
1592
1593                 if (!path_is_absolute(fn)) {
1594
1595                         if (ignore)
1596                                 continue;
1597
1598                         strv_free(r);
1599                         return -EINVAL;
1600                 }
1601
1602                 if ((k = load_env_file(fn, &p)) < 0) {
1603
1604                         if (ignore)
1605                                 continue;
1606
1607                         strv_free(r);
1608                         return k;
1609                 }
1610
1611                 if (r == NULL)
1612                         r = p;
1613                 else {
1614                         char **m;
1615
1616                         m = strv_env_merge(2, r, p);
1617                         strv_free(r);
1618                         strv_free(p);
1619
1620                         if (!m)
1621                                 return -ENOMEM;
1622
1623                         r = m;
1624                 }
1625         }
1626
1627         *l = r;
1628
1629         return 0;
1630 }
1631
1632 static void strv_fprintf(FILE *f, char **l) {
1633         char **g;
1634
1635         assert(f);
1636
1637         STRV_FOREACH(g, l)
1638                 fprintf(f, " %s", *g);
1639 }
1640
1641 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1642         char ** e;
1643         unsigned i;
1644
1645         assert(c);
1646         assert(f);
1647
1648         if (!prefix)
1649                 prefix = "";
1650
1651         fprintf(f,
1652                 "%sUMask: %04o\n"
1653                 "%sWorkingDirectory: %s\n"
1654                 "%sRootDirectory: %s\n"
1655                 "%sNonBlocking: %s\n"
1656                 "%sPrivateTmp: %s\n"
1657                 "%sControlGroupModify: %s\n",
1658                 prefix, c->umask,
1659                 prefix, c->working_directory ? c->working_directory : "/",
1660                 prefix, c->root_directory ? c->root_directory : "/",
1661                 prefix, yes_no(c->non_blocking),
1662                 prefix, yes_no(c->private_tmp),
1663                 prefix, yes_no(c->control_group_modify));
1664
1665         STRV_FOREACH(e, c->environment)
1666                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1667
1668         STRV_FOREACH(e, c->environment_files)
1669                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1670
1671         if (c->tcpwrap_name)
1672                 fprintf(f,
1673                         "%sTCPWrapName: %s\n",
1674                         prefix, c->tcpwrap_name);
1675
1676         if (c->nice_set)
1677                 fprintf(f,
1678                         "%sNice: %i\n",
1679                         prefix, c->nice);
1680
1681         if (c->oom_score_adjust_set)
1682                 fprintf(f,
1683                         "%sOOMScoreAdjust: %i\n",
1684                         prefix, c->oom_score_adjust);
1685
1686         for (i = 0; i < RLIM_NLIMITS; i++)
1687                 if (c->rlimit[i])
1688                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1689
1690         if (c->ioprio_set)
1691                 fprintf(f,
1692                         "%sIOSchedulingClass: %s\n"
1693                         "%sIOPriority: %i\n",
1694                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1695                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1696
1697         if (c->cpu_sched_set)
1698                 fprintf(f,
1699                         "%sCPUSchedulingPolicy: %s\n"
1700                         "%sCPUSchedulingPriority: %i\n"
1701                         "%sCPUSchedulingResetOnFork: %s\n",
1702                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1703                         prefix, c->cpu_sched_priority,
1704                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1705
1706         if (c->cpuset) {
1707                 fprintf(f, "%sCPUAffinity:", prefix);
1708                 for (i = 0; i < c->cpuset_ncpus; i++)
1709                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1710                                 fprintf(f, " %i", i);
1711                 fputs("\n", f);
1712         }
1713
1714         if (c->timer_slack_nsec_set)
1715                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1716
1717         fprintf(f,
1718                 "%sStandardInput: %s\n"
1719                 "%sStandardOutput: %s\n"
1720                 "%sStandardError: %s\n",
1721                 prefix, exec_input_to_string(c->std_input),
1722                 prefix, exec_output_to_string(c->std_output),
1723                 prefix, exec_output_to_string(c->std_error));
1724
1725         if (c->tty_path)
1726                 fprintf(f,
1727                         "%sTTYPath: %s\n"
1728                         "%sTTYReset: %s\n"
1729                         "%sTTYVHangup: %s\n"
1730                         "%sTTYVTDisallocate: %s\n",
1731                         prefix, c->tty_path,
1732                         prefix, yes_no(c->tty_reset),
1733                         prefix, yes_no(c->tty_vhangup),
1734                         prefix, yes_no(c->tty_vt_disallocate));
1735
1736         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1737             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1738             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1739             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1740                 fprintf(f,
1741                         "%sSyslogFacility: %s\n"
1742                         "%sSyslogLevel: %s\n",
1743                         prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1744                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1745
1746         if (c->capabilities) {
1747                 char *t;
1748                 if ((t = cap_to_text(c->capabilities, NULL))) {
1749                         fprintf(f, "%sCapabilities: %s\n",
1750                                 prefix, t);
1751                         cap_free(t);
1752                 }
1753         }
1754
1755         if (c->secure_bits)
1756                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1757                         prefix,
1758                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1759                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1760                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1761                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1762                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1763                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1764
1765         if (c->capability_bounding_set_drop) {
1766                 unsigned long l;
1767                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1768
1769                 for (l = 0; l <= (unsigned long) CAP_LAST_CAP; l++)
1770                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1771                                 char *t;
1772
1773                                 if ((t = cap_to_name(l))) {
1774                                         fprintf(f, " %s", t);
1775                                         cap_free(t);
1776                                 }
1777                         }
1778
1779                 fputs("\n", f);
1780         }
1781
1782         if (c->user)
1783                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1784         if (c->group)
1785                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1786
1787         if (strv_length(c->supplementary_groups) > 0) {
1788                 fprintf(f, "%sSupplementaryGroups:", prefix);
1789                 strv_fprintf(f, c->supplementary_groups);
1790                 fputs("\n", f);
1791         }
1792
1793         if (c->pam_name)
1794                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1795
1796         if (strv_length(c->read_write_dirs) > 0) {
1797                 fprintf(f, "%sReadWriteDirs:", prefix);
1798                 strv_fprintf(f, c->read_write_dirs);
1799                 fputs("\n", f);
1800         }
1801
1802         if (strv_length(c->read_only_dirs) > 0) {
1803                 fprintf(f, "%sReadOnlyDirs:", prefix);
1804                 strv_fprintf(f, c->read_only_dirs);
1805                 fputs("\n", f);
1806         }
1807
1808         if (strv_length(c->inaccessible_dirs) > 0) {
1809                 fprintf(f, "%sInaccessibleDirs:", prefix);
1810                 strv_fprintf(f, c->inaccessible_dirs);
1811                 fputs("\n", f);
1812         }
1813
1814         fprintf(f,
1815                 "%sKillMode: %s\n"
1816                 "%sKillSignal: SIG%s\n"
1817                 "%sSendSIGKILL: %s\n",
1818                 prefix, kill_mode_to_string(c->kill_mode),
1819                 prefix, signal_to_string(c->kill_signal),
1820                 prefix, yes_no(c->send_sigkill));
1821
1822         if (c->utmp_id)
1823                 fprintf(f,
1824                         "%sUtmpIdentifier: %s\n",
1825                         prefix, c->utmp_id);
1826 }
1827
1828 void exec_status_start(ExecStatus *s, pid_t pid) {
1829         assert(s);
1830
1831         zero(*s);
1832         s->pid = pid;
1833         dual_timestamp_get(&s->start_timestamp);
1834 }
1835
1836 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1837         assert(s);
1838
1839         if ((s->pid && s->pid != pid) ||
1840             !s->start_timestamp.realtime <= 0)
1841                 zero(*s);
1842
1843         s->pid = pid;
1844         dual_timestamp_get(&s->exit_timestamp);
1845
1846         s->code = code;
1847         s->status = status;
1848
1849         if (context) {
1850                 if (context->utmp_id)
1851                         utmp_put_dead_process(context->utmp_id, pid, code, status);
1852
1853                 exec_context_tty_reset(context);
1854         }
1855 }
1856
1857 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1858         char buf[FORMAT_TIMESTAMP_MAX];
1859
1860         assert(s);
1861         assert(f);
1862
1863         if (!prefix)
1864                 prefix = "";
1865
1866         if (s->pid <= 0)
1867                 return;
1868
1869         fprintf(f,
1870                 "%sPID: %lu\n",
1871                 prefix, (unsigned long) s->pid);
1872
1873         if (s->start_timestamp.realtime > 0)
1874                 fprintf(f,
1875                         "%sStart Timestamp: %s\n",
1876                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1877
1878         if (s->exit_timestamp.realtime > 0)
1879                 fprintf(f,
1880                         "%sExit Timestamp: %s\n"
1881                         "%sExit Code: %s\n"
1882                         "%sExit Status: %i\n",
1883                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1884                         prefix, sigchld_code_to_string(s->code),
1885                         prefix, s->status);
1886 }
1887
1888 char *exec_command_line(char **argv) {
1889         size_t k;
1890         char *n, *p, **a;
1891         bool first = true;
1892
1893         assert(argv);
1894
1895         k = 1;
1896         STRV_FOREACH(a, argv)
1897                 k += strlen(*a)+3;
1898
1899         if (!(n = new(char, k)))
1900                 return NULL;
1901
1902         p = n;
1903         STRV_FOREACH(a, argv) {
1904
1905                 if (!first)
1906                         *(p++) = ' ';
1907                 else
1908                         first = false;
1909
1910                 if (strpbrk(*a, WHITESPACE)) {
1911                         *(p++) = '\'';
1912                         p = stpcpy(p, *a);
1913                         *(p++) = '\'';
1914                 } else
1915                         p = stpcpy(p, *a);
1916
1917         }
1918
1919         *p = 0;
1920
1921         /* FIXME: this doesn't really handle arguments that have
1922          * spaces and ticks in them */
1923
1924         return n;
1925 }
1926
1927 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1928         char *p2;
1929         const char *prefix2;
1930
1931         char *cmd;
1932
1933         assert(c);
1934         assert(f);
1935
1936         if (!prefix)
1937                 prefix = "";
1938         p2 = strappend(prefix, "\t");
1939         prefix2 = p2 ? p2 : prefix;
1940
1941         cmd = exec_command_line(c->argv);
1942
1943         fprintf(f,
1944                 "%sCommand Line: %s\n",
1945                 prefix, cmd ? cmd : strerror(ENOMEM));
1946
1947         free(cmd);
1948
1949         exec_status_dump(&c->exec_status, f, prefix2);
1950
1951         free(p2);
1952 }
1953
1954 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1955         assert(f);
1956
1957         if (!prefix)
1958                 prefix = "";
1959
1960         LIST_FOREACH(command, c, c)
1961                 exec_command_dump(c, f, prefix);
1962 }
1963
1964 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1965         ExecCommand *end;
1966
1967         assert(l);
1968         assert(e);
1969
1970         if (*l) {
1971                 /* It's kind of important, that we keep the order here */
1972                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1973                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1974         } else
1975               *l = e;
1976 }
1977
1978 int exec_command_set(ExecCommand *c, const char *path, ...) {
1979         va_list ap;
1980         char **l, *p;
1981
1982         assert(c);
1983         assert(path);
1984
1985         va_start(ap, path);
1986         l = strv_new_ap(path, ap);
1987         va_end(ap);
1988
1989         if (!l)
1990                 return -ENOMEM;
1991
1992         if (!(p = strdup(path))) {
1993                 strv_free(l);
1994                 return -ENOMEM;
1995         }
1996
1997         free(c->path);
1998         c->path = p;
1999
2000         strv_free(c->argv);
2001         c->argv = l;
2002
2003         return 0;
2004 }
2005
2006 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2007         [EXEC_INPUT_NULL] = "null",
2008         [EXEC_INPUT_TTY] = "tty",
2009         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2010         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2011         [EXEC_INPUT_SOCKET] = "socket"
2012 };
2013
2014 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2015
2016 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2017         [EXEC_OUTPUT_INHERIT] = "inherit",
2018         [EXEC_OUTPUT_NULL] = "null",
2019         [EXEC_OUTPUT_TTY] = "tty",
2020         [EXEC_OUTPUT_SYSLOG] = "syslog",
2021         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2022         [EXEC_OUTPUT_KMSG] = "kmsg",
2023         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2024         [EXEC_OUTPUT_SOCKET] = "socket"
2025 };
2026
2027 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2028
2029 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2030         [KILL_CONTROL_GROUP] = "control-group",
2031         [KILL_PROCESS] = "process",
2032         [KILL_NONE] = "none"
2033 };
2034
2035 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2036
2037 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2038         [KILL_MAIN] = "main",
2039         [KILL_CONTROL] = "control",
2040         [KILL_ALL] = "all"
2041 };
2042
2043 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);