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 ((pam_code = pam_setcred(handle, PAM_ESTABLISH_CRED | PAM_SILENT)) != PAM_SUCCESS)
 821                 goto fail;
 822
 823         if ((!(e = pam_getenvlist(handle)))) {
 824                 pam_code = PAM_BUF_ERR;
 825                 goto fail;
 826         }
 827
 828         /* Block SIGTERM, so that we know that it won't get lost in
 829          * the child */
 830         if (sigemptyset(&ss) < 0 ||
 831             sigaddset(&ss, SIGTERM) < 0 ||
 832             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 833                 goto fail;
 834
 835         parent_pid = getpid();
 836
 837         if ((pam_pid = fork()) < 0)
 838                 goto fail;
 839
 840         if (pam_pid == 0) {
 841                 int sig;
 842                 int r = EXIT_PAM;
 843
 844                 /* The child's job is to reset the PAM session on
 845                  * termination */
 846
 847                 /* This string must fit in 10 chars (i.e. the length
 848                  * of "/sbin/init") */
 849                 rename_process("sd:pam");
 850
 851                 /* Make sure we don't keep open the passed fds in this
 852                 child. We assume that otherwise only those fds are
 853                 open here that have been opened by PAM. */
 854                 close_many(fds, n_fds);
 855
 856                 /* Wait until our parent died. This will most likely
 857                  * not work since the kernel does not allow
 858                  * unprivileged parents kill their privileged children
 859                  * this way. We rely on the control groups kill logic
 860                  * to do the rest for us. */
 861                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 862                         goto child_finish;
 863
 864                 /* Check if our parent process might already have
 865                  * died? */
 866                 if (getppid() == parent_pid) {
 867                         if (sigwait(&ss, &sig) < 0)
 868                                 goto child_finish;
 869
 870                         assert(sig == SIGTERM);
 871                 }
 872
 873                 /* Only if our parent died we'll end the session */
 874                 if (getppid() != parent_pid)
 875                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
 876                                 goto child_finish;
 877
 878                 r = 0;
 879
 880         child_finish:
 881                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 882                 _exit(r);
 883         }
 884
 885         /* If the child was forked off successfully it will do all the
 886          * cleanups, so forget about the handle here. */
 887         handle = NULL;
 888
 889         /* Unblock SIGSUR1 again in the parent */
 890         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 891                 goto fail;
 892
 893         /* We close the log explicitly here, since the PAM modules
 894          * might have opened it, but we don't want this fd around. */
 895         closelog();
 896
 897         return 0;
 898
 899 fail:
 900         if (handle) {
 901                 if (close_session)
 902                         pam_code = pam_close_session(handle, PAM_DATA_SILENT);
 903
 904                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 905         }
 906
 907         strv_free(e);
 908
 909         closelog();
 910
 911         if (pam_pid > 1) {
 912                 kill(pam_pid, SIGTERM);
 913                 kill(pam_pid, SIGCONT);
 914         }
 915
 916         return EXIT_PAM;
 917 }
 918 #endif
 919
 920 static int do_capability_bounding_set_drop(uint64_t drop) {
 921         unsigned long i;
 922         cap_t old_cap = NULL, new_cap = NULL;
 923         cap_flag_value_t fv;
 924         int r;
 925
 926         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 927          * in the effective set (yes, the kernel drops that when
 928          * executing init!), so get it back temporarily so that we can
 929          * call PR_CAPBSET_DROP. */
 930
 931         old_cap = cap_get_proc();
 932         if (!old_cap)
 933                 return -errno;
 934
 935         if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
 936                 r = -errno;
 937                 goto finish;
 938         }
 939
 940         if (fv != CAP_SET) {
 941                 static const cap_value_t v = CAP_SETPCAP;
 942
 943                 new_cap = cap_dup(old_cap);
 944                 if (!new_cap) {
 945                         r = -errno;
 946                         goto finish;
 947                 }
 948
 949                 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
 950                         r = -errno;
 951                         goto finish;
 952                 }
 953
 954                 if (cap_set_proc(new_cap) < 0) {
 955                         r = -errno;
 956                         goto finish;
 957                 }
 958         }
 959
 960         for (i = 0; i <= CAP_LAST_CAP; i++)
 961                 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
 962                         if (prctl(PR_CAPBSET_DROP, i) < 0) {
 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
1248 #ifdef HAVE_PAM
1249                 if (context->pam_name && username) {
1250                         if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) < 0) {
1251                                 r = EXIT_PAM;
1252                                 goto fail_child;
1253                         }
1254                 }
1255 #endif
1256
1257                 if (apply_permissions)
1258                         if (enforce_groups(context, username, uid) < 0) {
1259                                 r = EXIT_GROUP;
1260                                 goto fail_child;
1261                         }
1262
1263                 umask(context->umask);
1264
1265                 if (strv_length(context->read_write_dirs) > 0 ||
1266                     strv_length(context->read_only_dirs) > 0 ||
1267                     strv_length(context->inaccessible_dirs) > 0 ||
1268                     context->mount_flags != MS_SHARED ||
1269                     context->private_tmp)
1270                         if ((r = setup_namespace(
1271                                              context->read_write_dirs,
1272                                              context->read_only_dirs,
1273                                              context->inaccessible_dirs,
1274                                              context->private_tmp,
1275                                              context->mount_flags)) < 0)
1276                                 goto fail_child;
1277
1278                 if (apply_chroot) {
1279                         if (context->root_directory)
1280                                 if (chroot(context->root_directory) < 0) {
1281                                         r = EXIT_CHROOT;
1282                                         goto fail_child;
1283                                 }
1284
1285                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1286                                 r = EXIT_CHDIR;
1287                                 goto fail_child;
1288                         }
1289                 } else {
1290
1291                         char *d;
1292
1293                         if (asprintf(&d, "%s/%s",
1294                                      context->root_directory ? context->root_directory : "",
1295                                      context->working_directory ? context->working_directory : "") < 0) {
1296                                 r = EXIT_MEMORY;
1297                                 goto fail_child;
1298                         }
1299
1300                         if (chdir(d) < 0) {
1301                                 free(d);
1302                                 r = EXIT_CHDIR;
1303                                 goto fail_child;
1304                         }
1305
1306                         free(d);
1307                 }
1308
1309                 /* We repeat the fd closing here, to make sure that
1310                  * nothing is leaked from the PAM modules */
1311                 if (close_all_fds(fds, n_fds) < 0 ||
1312                     shift_fds(fds, n_fds) < 0 ||
1313                     flags_fds(fds, n_fds, context->non_blocking) < 0) {
1314                         r = EXIT_FDS;
1315                         goto fail_child;
1316                 }
1317
1318                 if (apply_permissions) {
1319
1320                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1321                                 if (!context->rlimit[i])
1322                                         continue;
1323
1324                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1325                                         r = EXIT_LIMITS;
1326                                         goto fail_child;
1327                                 }
1328                         }
1329
1330                         if (context->capability_bounding_set_drop)
1331                                 if (do_capability_bounding_set_drop(context->capability_bounding_set_drop) < 0) {
1332                                         r = EXIT_CAPABILITIES;
1333                                         goto fail_child;
1334                                 }
1335
1336                         if (context->user)
1337                                 if (enforce_user(context, uid) < 0) {
1338                                         r = EXIT_USER;
1339                                         goto fail_child;
1340                                 }
1341
1342                         /* PR_GET_SECUREBITS is not privileged, while
1343                          * PR_SET_SECUREBITS is. So to suppress
1344                          * potential EPERMs we'll try not to call
1345                          * PR_SET_SECUREBITS unless necessary. */
1346                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1347                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1348                                         r = EXIT_SECUREBITS;
1349                                         goto fail_child;
1350                                 }
1351
1352                         if (context->capabilities)
1353                                 if (cap_set_proc(context->capabilities) < 0) {
1354                                         r = EXIT_CAPABILITIES;
1355                                         goto fail_child;
1356                                 }
1357                 }
1358
1359                 if (!(our_env = new0(char*, 7))) {
1360                         r = EXIT_MEMORY;
1361                         goto fail_child;
1362                 }
1363
1364                 if (n_fds > 0)
1365                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1366                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1367                                 r = EXIT_MEMORY;
1368                                 goto fail_child;
1369                         }
1370
1371                 if (home)
1372                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1373                                 r = EXIT_MEMORY;
1374                                 goto fail_child;
1375                         }
1376
1377                 if (username)
1378                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1379                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1380                                 r = EXIT_MEMORY;
1381                                 goto fail_child;
1382                         }
1383
1384                 if (is_terminal_input(context->std_input) ||
1385                     context->std_output == EXEC_OUTPUT_TTY ||
1386                     context->std_error == EXEC_OUTPUT_TTY)
1387                         if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1388                                 r = EXIT_MEMORY;
1389                                 goto fail_child;
1390                         }
1391
1392                 assert(n_env <= 7);
1393
1394                 if (!(final_env = strv_env_merge(
1395                                       5,
1396                                       environment,
1397                                       our_env,
1398                                       context->environment,
1399                                       files_env,
1400                                       pam_env,
1401                                       NULL))) {
1402                         r = EXIT_MEMORY;
1403                         goto fail_child;
1404                 }
1405
1406                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1407                         r = EXIT_MEMORY;
1408                         goto fail_child;
1409                 }
1410
1411                 final_env = strv_env_clean(final_env);
1412
1413                 execve(command->path, final_argv, final_env);
1414                 r = EXIT_EXEC;
1415
1416         fail_child:
1417                 strv_free(our_env);
1418                 strv_free(final_env);
1419                 strv_free(pam_env);
1420                 strv_free(files_env);
1421                 strv_free(final_argv);
1422
1423                 if (saved_stdin >= 0)
1424                         close_nointr_nofail(saved_stdin);
1425
1426                 if (saved_stdout >= 0)
1427                         close_nointr_nofail(saved_stdout);
1428
1429                 _exit(r);
1430         }
1431
1432         strv_free(files_env);
1433
1434         /* We add the new process to the cgroup both in the child (so
1435          * that we can be sure that no user code is ever executed
1436          * outside of the cgroup) and in the parent (so that we can be
1437          * sure that when we kill the cgroup the process will be
1438          * killed too). */
1439         if (cgroup_bondings)
1440                 cgroup_bonding_install_list(cgroup_bondings, pid);
1441
1442         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1443
1444         exec_status_start(&command->exec_status, pid);
1445
1446         *ret = pid;
1447         return 0;
1448
1449 fail_parent:
1450         strv_free(files_env);
1451
1452         return r;
1453 }
1454
1455 void exec_context_init(ExecContext *c) {
1456         assert(c);
1457
1458         c->umask = 0002;
1459         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1460         c->cpu_sched_policy = SCHED_OTHER;
1461         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1462         c->syslog_level_prefix = true;
1463         c->mount_flags = MS_SHARED;
1464         c->kill_signal = SIGTERM;
1465         c->send_sigkill = true;
1466 }
1467
1468 void exec_context_done(ExecContext *c) {
1469         unsigned l;
1470
1471         assert(c);
1472
1473         strv_free(c->environment);
1474         c->environment = NULL;
1475
1476         strv_free(c->environment_files);
1477         c->environment_files = NULL;
1478
1479         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1480                 free(c->rlimit[l]);
1481                 c->rlimit[l] = NULL;
1482         }
1483
1484         free(c->working_directory);
1485         c->working_directory = NULL;
1486         free(c->root_directory);
1487         c->root_directory = NULL;
1488
1489         free(c->tty_path);
1490         c->tty_path = NULL;
1491
1492         free(c->tcpwrap_name);
1493         c->tcpwrap_name = NULL;
1494
1495         free(c->syslog_identifier);
1496         c->syslog_identifier = NULL;
1497
1498         free(c->user);
1499         c->user = NULL;
1500
1501         free(c->group);
1502         c->group = NULL;
1503
1504         strv_free(c->supplementary_groups);
1505         c->supplementary_groups = NULL;
1506
1507         free(c->pam_name);
1508         c->pam_name = NULL;
1509
1510         if (c->capabilities) {
1511                 cap_free(c->capabilities);
1512                 c->capabilities = NULL;
1513         }
1514
1515         strv_free(c->read_only_dirs);
1516         c->read_only_dirs = NULL;
1517
1518         strv_free(c->read_write_dirs);
1519         c->read_write_dirs = NULL;
1520
1521         strv_free(c->inaccessible_dirs);
1522         c->inaccessible_dirs = NULL;
1523
1524         if (c->cpuset)
1525                 CPU_FREE(c->cpuset);
1526
1527         free(c->utmp_id);
1528         c->utmp_id = NULL;
1529 }
1530
1531 void exec_command_done(ExecCommand *c) {
1532         assert(c);
1533
1534         free(c->path);
1535         c->path = NULL;
1536
1537         strv_free(c->argv);
1538         c->argv = NULL;
1539 }
1540
1541 void exec_command_done_array(ExecCommand *c, unsigned n) {
1542         unsigned i;
1543
1544         for (i = 0; i < n; i++)
1545                 exec_command_done(c+i);
1546 }
1547
1548 void exec_command_free_list(ExecCommand *c) {
1549         ExecCommand *i;
1550
1551         while ((i = c)) {
1552                 LIST_REMOVE(ExecCommand, command, c, i);
1553                 exec_command_done(i);
1554                 free(i);
1555         }
1556 }
1557
1558 void exec_command_free_array(ExecCommand **c, unsigned n) {
1559         unsigned i;
1560
1561         for (i = 0; i < n; i++) {
1562                 exec_command_free_list(c[i]);
1563                 c[i] = NULL;
1564         }
1565 }
1566
1567 int exec_context_load_environment(const ExecContext *c, char ***l) {
1568         char **i, **r = NULL;
1569
1570         assert(c);
1571         assert(l);
1572
1573         STRV_FOREACH(i, c->environment_files) {
1574                 char *fn;
1575                 int k;
1576                 bool ignore = false;
1577                 char **p;
1578
1579                 fn = *i;
1580
1581                 if (fn[0] == '-') {
1582                         ignore = true;
1583                         fn ++;
1584                 }
1585
1586                 if (!path_is_absolute(fn)) {
1587
1588                         if (ignore)
1589                                 continue;
1590
1591                         strv_free(r);
1592                         return -EINVAL;
1593                 }
1594
1595                 if ((k = load_env_file(fn, &p)) < 0) {
1596
1597                         if (ignore)
1598                                 continue;
1599
1600                         strv_free(r);
1601                         return k;
1602                 }
1603
1604                 if (r == NULL)
1605                         r = p;
1606                 else {
1607                         char **m;
1608
1609                         m = strv_env_merge(2, r, p);
1610                         strv_free(r);
1611                         strv_free(p);
1612
1613                         if (!m)
1614                                 return -ENOMEM;
1615
1616                         r = m;
1617                 }
1618         }
1619
1620         *l = r;
1621
1622         return 0;
1623 }
1624
1625 static void strv_fprintf(FILE *f, char **l) {
1626         char **g;
1627
1628         assert(f);
1629
1630         STRV_FOREACH(g, l)
1631                 fprintf(f, " %s", *g);
1632 }
1633
1634 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1635         char ** e;
1636         unsigned i;
1637
1638         assert(c);
1639         assert(f);
1640
1641         if (!prefix)
1642                 prefix = "";
1643
1644         fprintf(f,
1645                 "%sUMask: %04o\n"
1646                 "%sWorkingDirectory: %s\n"
1647                 "%sRootDirectory: %s\n"
1648                 "%sNonBlocking: %s\n"
1649                 "%sPrivateTmp: %s\n",
1650                 prefix, c->umask,
1651                 prefix, c->working_directory ? c->working_directory : "/",
1652                 prefix, c->root_directory ? c->root_directory : "/",
1653                 prefix, yes_no(c->non_blocking),
1654                 prefix, yes_no(c->private_tmp));
1655
1656         STRV_FOREACH(e, c->environment)
1657                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1658
1659         STRV_FOREACH(e, c->environment_files)
1660                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1661
1662         if (c->tcpwrap_name)
1663                 fprintf(f,
1664                         "%sTCPWrapName: %s\n",
1665                         prefix, c->tcpwrap_name);
1666
1667         if (c->nice_set)
1668                 fprintf(f,
1669                         "%sNice: %i\n",
1670                         prefix, c->nice);
1671
1672         if (c->oom_score_adjust_set)
1673                 fprintf(f,
1674                         "%sOOMScoreAdjust: %i\n",
1675                         prefix, c->oom_score_adjust);
1676
1677         for (i = 0; i < RLIM_NLIMITS; i++)
1678                 if (c->rlimit[i])
1679                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1680
1681         if (c->ioprio_set)
1682                 fprintf(f,
1683                         "%sIOSchedulingClass: %s\n"
1684                         "%sIOPriority: %i\n",
1685                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1686                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1687
1688         if (c->cpu_sched_set)
1689                 fprintf(f,
1690                         "%sCPUSchedulingPolicy: %s\n"
1691                         "%sCPUSchedulingPriority: %i\n"
1692                         "%sCPUSchedulingResetOnFork: %s\n",
1693                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1694                         prefix, c->cpu_sched_priority,
1695                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1696
1697         if (c->cpuset) {
1698                 fprintf(f, "%sCPUAffinity:", prefix);
1699                 for (i = 0; i < c->cpuset_ncpus; i++)
1700                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1701                                 fprintf(f, " %i", i);
1702                 fputs("\n", f);
1703         }
1704
1705         if (c->timer_slack_nsec_set)
1706                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1707
1708         fprintf(f,
1709                 "%sStandardInput: %s\n"
1710                 "%sStandardOutput: %s\n"
1711                 "%sStandardError: %s\n",
1712                 prefix, exec_input_to_string(c->std_input),
1713                 prefix, exec_output_to_string(c->std_output),
1714                 prefix, exec_output_to_string(c->std_error));
1715
1716         if (c->tty_path)
1717                 fprintf(f,
1718                         "%sTTYPath: %s\n"
1719                         "%sTTYReset: %s\n"
1720                         "%sTTYVHangup: %s\n"
1721                         "%sTTYVTDisallocate: %s\n",
1722                         prefix, c->tty_path,
1723                         prefix, yes_no(c->tty_reset),
1724                         prefix, yes_no(c->tty_vhangup),
1725                         prefix, yes_no(c->tty_vt_disallocate));
1726
1727         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1728             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1729             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1730             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1731                 fprintf(f,
1732                         "%sSyslogFacility: %s\n"
1733                         "%sSyslogLevel: %s\n",
1734                         prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1735                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1736
1737         if (c->capabilities) {
1738                 char *t;
1739                 if ((t = cap_to_text(c->capabilities, NULL))) {
1740                         fprintf(f, "%sCapabilities: %s\n",
1741                                 prefix, t);
1742                         cap_free(t);
1743                 }
1744         }
1745
1746         if (c->secure_bits)
1747                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1748                         prefix,
1749                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1750                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1751                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1752                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1753                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1754                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1755
1756         if (c->capability_bounding_set_drop) {
1757                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1758
1759                 for (i = 0; i <= CAP_LAST_CAP; i++)
1760                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) i))) {
1761                                 char *t;
1762
1763                                 if ((t = cap_to_name(i))) {
1764                                         fprintf(f, " %s", t);
1765                                         cap_free(t);
1766                                 }
1767                         }
1768
1769                 fputs("\n", f);
1770         }
1771
1772         if (c->user)
1773                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1774         if (c->group)
1775                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1776
1777         if (strv_length(c->supplementary_groups) > 0) {
1778                 fprintf(f, "%sSupplementaryGroups:", prefix);
1779                 strv_fprintf(f, c->supplementary_groups);
1780                 fputs("\n", f);
1781         }
1782
1783         if (c->pam_name)
1784                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1785
1786         if (strv_length(c->read_write_dirs) > 0) {
1787                 fprintf(f, "%sReadWriteDirs:", prefix);
1788                 strv_fprintf(f, c->read_write_dirs);
1789                 fputs("\n", f);
1790         }
1791
1792         if (strv_length(c->read_only_dirs) > 0) {
1793                 fprintf(f, "%sReadOnlyDirs:", prefix);
1794                 strv_fprintf(f, c->read_only_dirs);
1795                 fputs("\n", f);
1796         }
1797
1798         if (strv_length(c->inaccessible_dirs) > 0) {
1799                 fprintf(f, "%sInaccessibleDirs:", prefix);
1800                 strv_fprintf(f, c->inaccessible_dirs);
1801                 fputs("\n", f);
1802         }
1803
1804         fprintf(f,
1805                 "%sKillMode: %s\n"
1806                 "%sKillSignal: SIG%s\n"
1807                 "%sSendSIGKILL: %s\n",
1808                 prefix, kill_mode_to_string(c->kill_mode),
1809                 prefix, signal_to_string(c->kill_signal),
1810                 prefix, yes_no(c->send_sigkill));
1811
1812         if (c->utmp_id)
1813                 fprintf(f,
1814                         "%sUtmpIdentifier: %s\n",
1815                         prefix, c->utmp_id);
1816 }
1817
1818 void exec_status_start(ExecStatus *s, pid_t pid) {
1819         assert(s);
1820
1821         zero(*s);
1822         s->pid = pid;
1823         dual_timestamp_get(&s->start_timestamp);
1824 }
1825
1826 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1827         assert(s);
1828
1829         if ((s->pid && s->pid != pid) ||
1830             !s->start_timestamp.realtime <= 0)
1831                 zero(*s);
1832
1833         s->pid = pid;
1834         dual_timestamp_get(&s->exit_timestamp);
1835
1836         s->code = code;
1837         s->status = status;
1838
1839         if (context) {
1840                 if (context->utmp_id)
1841                         utmp_put_dead_process(context->utmp_id, pid, code, status);
1842
1843                 exec_context_tty_reset(context);
1844         }
1845 }
1846
1847 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1848         char buf[FORMAT_TIMESTAMP_MAX];
1849
1850         assert(s);
1851         assert(f);
1852
1853         if (!prefix)
1854                 prefix = "";
1855
1856         if (s->pid <= 0)
1857                 return;
1858
1859         fprintf(f,
1860                 "%sPID: %lu\n",
1861                 prefix, (unsigned long) s->pid);
1862
1863         if (s->start_timestamp.realtime > 0)
1864                 fprintf(f,
1865                         "%sStart Timestamp: %s\n",
1866                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1867
1868         if (s->exit_timestamp.realtime > 0)
1869                 fprintf(f,
1870                         "%sExit Timestamp: %s\n"
1871                         "%sExit Code: %s\n"
1872                         "%sExit Status: %i\n",
1873                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1874                         prefix, sigchld_code_to_string(s->code),
1875                         prefix, s->status);
1876 }
1877
1878 char *exec_command_line(char **argv) {
1879         size_t k;
1880         char *n, *p, **a;
1881         bool first = true;
1882
1883         assert(argv);
1884
1885         k = 1;
1886         STRV_FOREACH(a, argv)
1887                 k += strlen(*a)+3;
1888
1889         if (!(n = new(char, k)))
1890                 return NULL;
1891
1892         p = n;
1893         STRV_FOREACH(a, argv) {
1894
1895                 if (!first)
1896                         *(p++) = ' ';
1897                 else
1898                         first = false;
1899
1900                 if (strpbrk(*a, WHITESPACE)) {
1901                         *(p++) = '\'';
1902                         p = stpcpy(p, *a);
1903                         *(p++) = '\'';
1904                 } else
1905                         p = stpcpy(p, *a);
1906
1907         }
1908
1909         *p = 0;
1910
1911         /* FIXME: this doesn't really handle arguments that have
1912          * spaces and ticks in them */
1913
1914         return n;
1915 }
1916
1917 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1918         char *p2;
1919         const char *prefix2;
1920
1921         char *cmd;
1922
1923         assert(c);
1924         assert(f);
1925
1926         if (!prefix)
1927                 prefix = "";
1928         p2 = strappend(prefix, "\t");
1929         prefix2 = p2 ? p2 : prefix;
1930
1931         cmd = exec_command_line(c->argv);
1932
1933         fprintf(f,
1934                 "%sCommand Line: %s\n",
1935                 prefix, cmd ? cmd : strerror(ENOMEM));
1936
1937         free(cmd);
1938
1939         exec_status_dump(&c->exec_status, f, prefix2);
1940
1941         free(p2);
1942 }
1943
1944 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1945         assert(f);
1946
1947         if (!prefix)
1948                 prefix = "";
1949
1950         LIST_FOREACH(command, c, c)
1951                 exec_command_dump(c, f, prefix);
1952 }
1953
1954 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1955         ExecCommand *end;
1956
1957         assert(l);
1958         assert(e);
1959
1960         if (*l) {
1961                 /* It's kind of important, that we keep the order here */
1962                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1963                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1964         } else
1965               *l = e;
1966 }
1967
1968 int exec_command_set(ExecCommand *c, const char *path, ...) {
1969         va_list ap;
1970         char **l, *p;
1971
1972         assert(c);
1973         assert(path);
1974
1975         va_start(ap, path);
1976         l = strv_new_ap(path, ap);
1977         va_end(ap);
1978
1979         if (!l)
1980                 return -ENOMEM;
1981
1982         if (!(p = strdup(path))) {
1983                 strv_free(l);
1984                 return -ENOMEM;
1985         }
1986
1987         free(c->path);
1988         c->path = p;
1989
1990         strv_free(c->argv);
1991         c->argv = l;
1992
1993         return 0;
1994 }
1995
1996 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1997         [EXEC_INPUT_NULL] = "null",
1998         [EXEC_INPUT_TTY] = "tty",
1999         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2000         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2001         [EXEC_INPUT_SOCKET] = "socket"
2002 };
2003
2004 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2005
2006 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2007         [EXEC_OUTPUT_INHERIT] = "inherit",
2008         [EXEC_OUTPUT_NULL] = "null",
2009         [EXEC_OUTPUT_TTY] = "tty",
2010         [EXEC_OUTPUT_SYSLOG] = "syslog",
2011         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2012         [EXEC_OUTPUT_KMSG] = "kmsg",
2013         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2014         [EXEC_OUTPUT_SOCKET] = "socket"
2015 };
2016
2017 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2018
2019 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2020         [KILL_CONTROL_GROUP] = "control-group",
2021         [KILL_PROCESS] = "process",
2022         [KILL_NONE] = "none"
2023 };
2024
2025 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2026
2027 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2028         [KILL_MAIN] = "main",
2029         [KILL_CONTROL] = "control",
2030         [KILL_ALL] = "all"
2031 };
2032
2033 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);