src/core/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 Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 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   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser 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 #include <sys/poll.h>
  41 #include <glob.h>
  42 #include <sys/personality.h>
  43 #include <libgen.h>
  44 #undef basename
  45
  46 #ifdef HAVE_PAM
  47 #include <security/pam_appl.h>
  48 #endif
  49
  50 #ifdef HAVE_SELINUX
  51 #include <selinux/selinux.h>
  52 #endif
  53
  54 #ifdef HAVE_SECCOMP
  55 #include <seccomp.h>
  56 #endif
  57
  58 #ifdef HAVE_APPARMOR
  59 #include <sys/apparmor.h>
  60 #endif
  61
  62 #include "execute.h"
  63 #include "strv.h"
  64 #include "macro.h"
  65 #include "capability.h"
  66 #include "util.h"
  67 #include "log.h"
  68 #include "sd-messages.h"
  69 #include "ioprio.h"
  70 #include "securebits.h"
  71 #include "namespace.h"
  72 #include "exit-status.h"
  73 #include "missing.h"
  74 #include "utmp-wtmp.h"
  75 #include "def.h"
  76 #include "path-util.h"
  77 #include "env-util.h"
  78 #include "fileio.h"
  79 #include "unit.h"
  80 #include "async.h"
  81 #include "selinux-util.h"
  82 #include "errno-list.h"
  83 #include "af-list.h"
  84 #include "mkdir.h"
  85 #include "apparmor-util.h"
  86 #include "bus-kernel.h"
  87 #include "label.h"
  88
  89 #ifdef HAVE_SECCOMP
  90 #include "seccomp-util.h"
  91 #endif
  92
  93 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
  94 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
  95
  96 /* This assumes there is a 'tty' group */
  97 #define TTY_MODE 0620
  98
  99 #define SNDBUF_SIZE (8*1024*1024)
 100
 101 static int shift_fds(int fds[], unsigned n_fds) {
 102         int start, restart_from;
 103
 104         if (n_fds <= 0)
 105                 return 0;
 106
 107         /* Modifies the fds array! (sorts it) */
 108
 109         assert(fds);
 110
 111         start = 0;
 112         for (;;) {
 113                 int i;
 114
 115                 restart_from = -1;
 116
 117                 for (i = start; i < (int) n_fds; i++) {
 118                         int nfd;
 119
 120                         /* Already at right index? */
 121                         if (fds[i] == i+3)
 122                                 continue;
 123
 124                         if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
 125                                 return -errno;
 126
 127                         safe_close(fds[i]);
 128                         fds[i] = nfd;
 129
 130                         /* Hmm, the fd we wanted isn't free? Then
 131                          * let's remember that and try again from here*/
 132                         if (nfd != i+3 && restart_from < 0)
 133                                 restart_from = i;
 134                 }
 135
 136                 if (restart_from < 0)
 137                         break;
 138
 139                 start = restart_from;
 140         }
 141
 142         return 0;
 143 }
 144
 145 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
 146         unsigned i;
 147         int r;
 148
 149         if (n_fds <= 0)
 150                 return 0;
 151
 152         assert(fds);
 153
 154         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
 155
 156         for (i = 0; i < n_fds; i++) {
 157
 158                 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
 159                         return r;
 160
 161                 /* We unconditionally drop FD_CLOEXEC from the fds,
 162                  * since after all we want to pass these fds to our
 163                  * children */
 164
 165                 if ((r = fd_cloexec(fds[i], false)) < 0)
 166                         return r;
 167         }
 168
 169         return 0;
 170 }
 171
 172 _pure_ static const char *tty_path(const ExecContext *context) {
 173         assert(context);
 174
 175         if (context->tty_path)
 176                 return context->tty_path;
 177
 178         return "/dev/console";
 179 }
 180
 181 static void exec_context_tty_reset(const ExecContext *context) {
 182         assert(context);
 183
 184         if (context->tty_vhangup)
 185                 terminal_vhangup(tty_path(context));
 186
 187         if (context->tty_reset)
 188                 reset_terminal(tty_path(context));
 189
 190         if (context->tty_vt_disallocate && context->tty_path)
 191                 vt_disallocate(context->tty_path);
 192 }
 193
 194 static bool is_terminal_output(ExecOutput o) {
 195         return
 196                 o == EXEC_OUTPUT_TTY ||
 197                 o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
 198                 o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
 199                 o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
 200 }
 201
 202 static int open_null_as(int flags, int nfd) {
 203         int fd, r;
 204
 205         assert(nfd >= 0);
 206
 207         fd = open("/dev/null", flags|O_NOCTTY);
 208         if (fd < 0)
 209                 return -errno;
 210
 211         if (fd != nfd) {
 212                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 213                 safe_close(fd);
 214         } else
 215                 r = nfd;
 216
 217         return r;
 218 }
 219
 220 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
 221         int fd, r;
 222         union sockaddr_union sa = {
 223                 .un.sun_family = AF_UNIX,
 224                 .un.sun_path = "/run/systemd/journal/stdout",
 225         };
 226
 227         assert(context);
 228         assert(output < _EXEC_OUTPUT_MAX);
 229         assert(ident);
 230         assert(nfd >= 0);
 231
 232         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 233         if (fd < 0)
 234                 return -errno;
 235
 236         r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
 237         if (r < 0) {
 238                 safe_close(fd);
 239                 return -errno;
 240         }
 241
 242         if (shutdown(fd, SHUT_RD) < 0) {
 243                 safe_close(fd);
 244                 return -errno;
 245         }
 246
 247         fd_inc_sndbuf(fd, SNDBUF_SIZE);
 248
 249         dprintf(fd,
 250                 "%s\n"
 251                 "%s\n"
 252                 "%i\n"
 253                 "%i\n"
 254                 "%i\n"
 255                 "%i\n"
 256                 "%i\n",
 257                 context->syslog_identifier ? context->syslog_identifier : ident,
 258                 unit_id,
 259                 context->syslog_priority,
 260                 !!context->syslog_level_prefix,
 261                 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
 262                 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
 263                 is_terminal_output(output));
 264
 265         if (fd != nfd) {
 266                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 267                 safe_close(fd);
 268         } else
 269                 r = nfd;
 270
 271         return r;
 272 }
 273 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
 274         int fd, r;
 275
 276         assert(path);
 277         assert(nfd >= 0);
 278
 279         if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
 280                 return fd;
 281
 282         if (fd != nfd) {
 283                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 284                 safe_close(fd);
 285         } else
 286                 r = nfd;
 287
 288         return r;
 289 }
 290
 291 static bool is_terminal_input(ExecInput i) {
 292         return
 293                 i == EXEC_INPUT_TTY ||
 294                 i == EXEC_INPUT_TTY_FORCE ||
 295                 i == EXEC_INPUT_TTY_FAIL;
 296 }
 297
 298 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
 299
 300         if (is_terminal_input(std_input) && !apply_tty_stdin)
 301                 return EXEC_INPUT_NULL;
 302
 303         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 304                 return EXEC_INPUT_NULL;
 305
 306         return std_input;
 307 }
 308
 309 static int fixup_output(ExecOutput std_output, int socket_fd) {
 310
 311         if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 312                 return EXEC_OUTPUT_INHERIT;
 313
 314         return std_output;
 315 }
 316
 317 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
 318         ExecInput i;
 319
 320         assert(context);
 321
 322         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 323
 324         switch (i) {
 325
 326         case EXEC_INPUT_NULL:
 327                 return open_null_as(O_RDONLY, STDIN_FILENO);
 328
 329         case EXEC_INPUT_TTY:
 330         case EXEC_INPUT_TTY_FORCE:
 331         case EXEC_INPUT_TTY_FAIL: {
 332                 int fd, r;
 333
 334                 fd = acquire_terminal(tty_path(context),
 335                                       i == EXEC_INPUT_TTY_FAIL,
 336                                       i == EXEC_INPUT_TTY_FORCE,
 337                                       false,
 338                                       USEC_INFINITY);
 339                 if (fd < 0)
 340                         return fd;
 341
 342                 if (fd != STDIN_FILENO) {
 343                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 344                         safe_close(fd);
 345                 } else
 346                         r = STDIN_FILENO;
 347
 348                 return r;
 349         }
 350
 351         case EXEC_INPUT_SOCKET:
 352                 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 353
 354         default:
 355                 assert_not_reached("Unknown input type");
 356         }
 357 }
 358
 359 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
 360         ExecOutput o;
 361         ExecInput i;
 362         int r;
 363
 364         assert(context);
 365         assert(ident);
 366
 367         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 368         o = fixup_output(context->std_output, socket_fd);
 369
 370         if (fileno == STDERR_FILENO) {
 371                 ExecOutput e;
 372                 e = fixup_output(context->std_error, socket_fd);
 373
 374                 /* This expects the input and output are already set up */
 375
 376                 /* Don't change the stderr file descriptor if we inherit all
 377                  * the way and are not on a tty */
 378                 if (e == EXEC_OUTPUT_INHERIT &&
 379                     o == EXEC_OUTPUT_INHERIT &&
 380                     i == EXEC_INPUT_NULL &&
 381                     !is_terminal_input(context->std_input) &&
 382                     getppid () != 1)
 383                         return fileno;
 384
 385                 /* Duplicate from stdout if possible */
 386                 if (e == o || e == EXEC_OUTPUT_INHERIT)
 387                         return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
 388
 389                 o = e;
 390
 391         } else if (o == EXEC_OUTPUT_INHERIT) {
 392                 /* If input got downgraded, inherit the original value */
 393                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 394                         return open_terminal_as(tty_path(context), O_WRONLY, fileno);
 395
 396                 /* If the input is connected to anything that's not a /dev/null, inherit that... */
 397                 if (i != EXEC_INPUT_NULL)
 398                         return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
 399
 400                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 401                 if (getppid() != 1)
 402                         return fileno;
 403
 404                 /* We need to open /dev/null here anew, to get the right access mode. */
 405                 return open_null_as(O_WRONLY, fileno);
 406         }
 407
 408         switch (o) {
 409
 410         case EXEC_OUTPUT_NULL:
 411                 return open_null_as(O_WRONLY, fileno);
 412
 413         case EXEC_OUTPUT_TTY:
 414                 if (is_terminal_input(i))
 415                         return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
 416
 417                 /* We don't reset the terminal if this is just about output */
 418                 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
 419
 420         case EXEC_OUTPUT_SYSLOG:
 421         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 422         case EXEC_OUTPUT_KMSG:
 423         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 424         case EXEC_OUTPUT_JOURNAL:
 425         case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
 426                 r = connect_logger_as(context, o, ident, unit_id, fileno);
 427                 if (r < 0) {
 428                         log_struct_unit(LOG_CRIT, unit_id,
 429                                 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
 430                                 fileno == STDOUT_FILENO ? "out" : "err",
 431                                 unit_id, strerror(-r),
 432                                 "ERRNO=%d", -r,
 433                                 NULL);
 434                         r = open_null_as(O_WRONLY, fileno);
 435                 }
 436                 return r;
 437
 438         case EXEC_OUTPUT_SOCKET:
 439                 assert(socket_fd >= 0);
 440                 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
 441
 442         default:
 443                 assert_not_reached("Unknown error type");
 444         }
 445 }
 446
 447 static int chown_terminal(int fd, uid_t uid) {
 448         struct stat st;
 449
 450         assert(fd >= 0);
 451
 452         /* This might fail. What matters are the results. */
 453         (void) fchown(fd, uid, -1);
 454         (void) fchmod(fd, TTY_MODE);
 455
 456         if (fstat(fd, &st) < 0)
 457                 return -errno;
 458
 459         if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
 460                 return -EPERM;
 461
 462         return 0;
 463 }
 464
 465 static int setup_confirm_stdio(int *_saved_stdin,
 466                                int *_saved_stdout) {
 467         int fd = -1, saved_stdin, saved_stdout = -1, r;
 468
 469         assert(_saved_stdin);
 470         assert(_saved_stdout);
 471
 472         saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
 473         if (saved_stdin < 0)
 474                 return -errno;
 475
 476         saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
 477         if (saved_stdout < 0) {
 478                 r = errno;
 479                 goto fail;
 480         }
 481
 482         fd = acquire_terminal(
 483                         "/dev/console",
 484                         false,
 485                         false,
 486                         false,
 487                         DEFAULT_CONFIRM_USEC);
 488         if (fd < 0) {
 489                 r = fd;
 490                 goto fail;
 491         }
 492
 493         r = chown_terminal(fd, getuid());
 494         if (r < 0)
 495                 goto fail;
 496
 497         if (dup2(fd, STDIN_FILENO) < 0) {
 498                 r = -errno;
 499                 goto fail;
 500         }
 501
 502         if (dup2(fd, STDOUT_FILENO) < 0) {
 503                 r = -errno;
 504                 goto fail;
 505         }
 506
 507         if (fd >= 2)
 508                 safe_close(fd);
 509
 510         *_saved_stdin = saved_stdin;
 511         *_saved_stdout = saved_stdout;
 512
 513         return 0;
 514
 515 fail:
 516         safe_close(saved_stdout);
 517         safe_close(saved_stdin);
 518         safe_close(fd);
 519
 520         return r;
 521 }
 522
 523 _printf_(1, 2) static int write_confirm_message(const char *format, ...) {
 524         _cleanup_close_ int fd = -1;
 525         va_list ap;
 526
 527         assert(format);
 528
 529         fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
 530         if (fd < 0)
 531                 return fd;
 532
 533         va_start(ap, format);
 534         vdprintf(fd, format, ap);
 535         va_end(ap);
 536
 537         return 0;
 538 }
 539
 540 static int restore_confirm_stdio(int *saved_stdin,
 541                                  int *saved_stdout) {
 542
 543         int r = 0;
 544
 545         assert(saved_stdin);
 546         assert(saved_stdout);
 547
 548         release_terminal();
 549
 550         if (*saved_stdin >= 0)
 551                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 552                         r = -errno;
 553
 554         if (*saved_stdout >= 0)
 555                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 556                         r = -errno;
 557
 558         safe_close(*saved_stdin);
 559         safe_close(*saved_stdout);
 560
 561         return r;
 562 }
 563
 564 static int ask_for_confirmation(char *response, char **argv) {
 565         int saved_stdout = -1, saved_stdin = -1, r;
 566         _cleanup_free_ char *line = NULL;
 567
 568         r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
 569         if (r < 0)
 570                 return r;
 571
 572         line = exec_command_line(argv);
 573         if (!line)
 574                 return -ENOMEM;
 575
 576         r = ask_char(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
 577
 578         restore_confirm_stdio(&saved_stdin, &saved_stdout);
 579
 580         return r;
 581 }
 582
 583 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
 584         bool keep_groups = false;
 585         int r;
 586
 587         assert(context);
 588
 589         /* Lookup and set GID and supplementary group list. Here too
 590          * we avoid NSS lookups for gid=0. */
 591
 592         if (context->group || username) {
 593
 594                 if (context->group) {
 595                         const char *g = context->group;
 596
 597                         if ((r = get_group_creds(&g, &gid)) < 0)
 598                                 return r;
 599                 }
 600
 601                 /* First step, initialize groups from /etc/groups */
 602                 if (username && gid != 0) {
 603                         if (initgroups(username, gid) < 0)
 604                                 return -errno;
 605
 606                         keep_groups = true;
 607                 }
 608
 609                 /* Second step, set our gids */
 610                 if (setresgid(gid, gid, gid) < 0)
 611                         return -errno;
 612         }
 613
 614         if (context->supplementary_groups) {
 615                 int ngroups_max, k;
 616                 gid_t *gids;
 617                 char **i;
 618
 619                 /* Final step, initialize any manually set supplementary groups */
 620                 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
 621
 622                 if (!(gids = new(gid_t, ngroups_max)))
 623                         return -ENOMEM;
 624
 625                 if (keep_groups) {
 626                         if ((k = getgroups(ngroups_max, gids)) < 0) {
 627                                 free(gids);
 628                                 return -errno;
 629                         }
 630                 } else
 631                         k = 0;
 632
 633                 STRV_FOREACH(i, context->supplementary_groups) {
 634                         const char *g;
 635
 636                         if (k >= ngroups_max) {
 637                                 free(gids);
 638                                 return -E2BIG;
 639                         }
 640
 641                         g = *i;
 642                         r = get_group_creds(&g, gids+k);
 643                         if (r < 0) {
 644                                 free(gids);
 645                                 return r;
 646                         }
 647
 648                         k++;
 649                 }
 650
 651                 if (setgroups(k, gids) < 0) {
 652                         free(gids);
 653                         return -errno;
 654                 }
 655
 656                 free(gids);
 657         }
 658
 659         return 0;
 660 }
 661
 662 static int enforce_user(const ExecContext *context, uid_t uid) {
 663         assert(context);
 664
 665         /* Sets (but doesn't lookup) the uid and make sure we keep the
 666          * capabilities while doing so. */
 667
 668         if (context->capabilities) {
 669                 _cleanup_cap_free_ cap_t d = NULL;
 670                 static const cap_value_t bits[] = {
 671                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
 672                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
 673                 };
 674
 675                 /* First step: If we need to keep capabilities but
 676                  * drop privileges we need to make sure we keep our
 677                  * caps, while we drop privileges. */
 678                 if (uid != 0) {
 679                         int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
 680
 681                         if (prctl(PR_GET_SECUREBITS) != sb)
 682                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 683                                         return -errno;
 684                 }
 685
 686                 /* Second step: set the capabilities. This will reduce
 687                  * the capabilities to the minimum we need. */
 688
 689                 d = cap_dup(context->capabilities);
 690                 if (!d)
 691                         return -errno;
 692
 693                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
 694                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
 695                         return -errno;
 696
 697                 if (cap_set_proc(d) < 0)
 698                         return -errno;
 699         }
 700
 701         /* Third step: actually set the uids */
 702         if (setresuid(uid, uid, uid) < 0)
 703                 return -errno;
 704
 705         /* At this point we should have all necessary capabilities but
 706            are otherwise a normal user. However, the caps might got
 707            corrupted due to the setresuid() so we need clean them up
 708            later. This is done outside of this call. */
 709
 710         return 0;
 711 }
 712
 713 #ifdef HAVE_PAM
 714
 715 static int null_conv(
 716                 int num_msg,
 717                 const struct pam_message **msg,
 718                 struct pam_response **resp,
 719                 void *appdata_ptr) {
 720
 721         /* We don't support conversations */
 722
 723         return PAM_CONV_ERR;
 724 }
 725
 726 static int setup_pam(
 727                 const char *name,
 728                 const char *user,
 729                 uid_t uid,
 730                 const char *tty,
 731                 char ***pam_env,
 732                 int fds[], unsigned n_fds) {
 733
 734         static const struct pam_conv conv = {
 735                 .conv = null_conv,
 736                 .appdata_ptr = NULL
 737         };
 738
 739         pam_handle_t *handle = NULL;
 740         sigset_t ss, old_ss;
 741         int pam_code = PAM_SUCCESS;
 742         int err;
 743         char **e = NULL;
 744         bool close_session = false;
 745         pid_t pam_pid = 0, parent_pid;
 746         int flags = 0;
 747
 748         assert(name);
 749         assert(user);
 750         assert(pam_env);
 751
 752         /* We set up PAM in the parent process, then fork. The child
 753          * will then stay around until killed via PR_GET_PDEATHSIG or
 754          * systemd via the cgroup logic. It will then remove the PAM
 755          * session again. The parent process will exec() the actual
 756          * daemon. We do things this way to ensure that the main PID
 757          * of the daemon is the one we initially fork()ed. */
 758
 759         if (log_get_max_level() < LOG_PRI(LOG_DEBUG))
 760                 flags |= PAM_SILENT;
 761
 762         pam_code = pam_start(name, user, &conv, &handle);
 763         if (pam_code != PAM_SUCCESS) {
 764                 handle = NULL;
 765                 goto fail;
 766         }
 767
 768         if (tty) {
 769                 pam_code = pam_set_item(handle, PAM_TTY, tty);
 770                 if (pam_code != PAM_SUCCESS)
 771                         goto fail;
 772         }
 773
 774         pam_code = pam_acct_mgmt(handle, flags);
 775         if (pam_code != PAM_SUCCESS)
 776                 goto fail;
 777
 778         pam_code = pam_open_session(handle, flags);
 779         if (pam_code != PAM_SUCCESS)
 780                 goto fail;
 781
 782         close_session = true;
 783
 784         e = pam_getenvlist(handle);
 785         if (!e) {
 786                 pam_code = PAM_BUF_ERR;
 787                 goto fail;
 788         }
 789
 790         /* Block SIGTERM, so that we know that it won't get lost in
 791          * the child */
 792         if (sigemptyset(&ss) < 0 ||
 793             sigaddset(&ss, SIGTERM) < 0 ||
 794             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 795                 goto fail;
 796
 797         parent_pid = getpid();
 798
 799         pam_pid = fork();
 800         if (pam_pid < 0)
 801                 goto fail;
 802
 803         if (pam_pid == 0) {
 804                 int sig;
 805                 int r = EXIT_PAM;
 806
 807                 /* The child's job is to reset the PAM session on
 808                  * termination */
 809
 810                 /* This string must fit in 10 chars (i.e. the length
 811                  * of "/sbin/init"), to look pretty in /bin/ps */
 812                 rename_process("(sd-pam)");
 813
 814                 /* Make sure we don't keep open the passed fds in this
 815                 child. We assume that otherwise only those fds are
 816                 open here that have been opened by PAM. */
 817                 close_many(fds, n_fds);
 818
 819                 /* Drop privileges - we don't need any to pam_close_session
 820                  * and this will make PR_SET_PDEATHSIG work in most cases.
 821                  * If this fails, ignore the error - but expect sd-pam threads
 822                  * to fail to exit normally */
 823                 if (setresuid(uid, uid, uid) < 0)
 824                         log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r));
 825
 826                 /* Wait until our parent died. This will only work if
 827                  * the above setresuid() succeeds, otherwise the kernel
 828                  * will not allow unprivileged parents kill their privileged
 829                  * children this way. We rely on the control groups kill logic
 830                  * to do the rest for us. */
 831                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 832                         goto child_finish;
 833
 834                 /* Check if our parent process might already have
 835                  * died? */
 836                 if (getppid() == parent_pid) {
 837                         for (;;) {
 838                                 if (sigwait(&ss, &sig) < 0) {
 839                                         if (errno == EINTR)
 840                                                 continue;
 841
 842                                         goto child_finish;
 843                                 }
 844
 845                                 assert(sig == SIGTERM);
 846                                 break;
 847                         }
 848                 }
 849
 850                 /* If our parent died we'll end the session */
 851                 if (getppid() != parent_pid) {
 852                         pam_code = pam_close_session(handle, flags);
 853                         if (pam_code != PAM_SUCCESS)
 854                                 goto child_finish;
 855                 }
 856
 857                 r = 0;
 858
 859         child_finish:
 860                 pam_end(handle, pam_code | flags);
 861                 _exit(r);
 862         }
 863
 864         /* If the child was forked off successfully it will do all the
 865          * cleanups, so forget about the handle here. */
 866         handle = NULL;
 867
 868         /* Unblock SIGTERM again in the parent */
 869         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 870                 goto fail;
 871
 872         /* We close the log explicitly here, since the PAM modules
 873          * might have opened it, but we don't want this fd around. */
 874         closelog();
 875
 876         *pam_env = e;
 877         e = NULL;
 878
 879         return 0;
 880
 881 fail:
 882         if (pam_code != PAM_SUCCESS) {
 883                 log_error("PAM failed: %s", pam_strerror(handle, pam_code));
 884                 err = -EPERM;  /* PAM errors do not map to errno */
 885         } else {
 886                 log_error("PAM failed: %m");
 887                 err = -errno;
 888         }
 889
 890         if (handle) {
 891                 if (close_session)
 892                         pam_code = pam_close_session(handle, flags);
 893
 894                 pam_end(handle, pam_code | flags);
 895         }
 896
 897         strv_free(e);
 898
 899         closelog();
 900
 901         if (pam_pid > 1) {
 902                 kill(pam_pid, SIGTERM);
 903                 kill(pam_pid, SIGCONT);
 904         }
 905
 906         return err;
 907 }
 908 #endif
 909
 910 static void rename_process_from_path(const char *path) {
 911         char process_name[11];
 912         const char *p;
 913         size_t l;
 914
 915         /* This resulting string must fit in 10 chars (i.e. the length
 916          * of "/sbin/init") to look pretty in /bin/ps */
 917
 918         p = basename(path);
 919         if (isempty(p)) {
 920                 rename_process("(...)");
 921                 return;
 922         }
 923
 924         l = strlen(p);
 925         if (l > 8) {
 926                 /* The end of the process name is usually more
 927                  * interesting, since the first bit might just be
 928                  * "systemd-" */
 929                 p = p + l - 8;
 930                 l = 8;
 931         }
 932
 933         process_name[0] = '(';
 934         memcpy(process_name+1, p, l);
 935         process_name[1+l] = ')';
 936         process_name[1+l+1] = 0;
 937
 938         rename_process(process_name);
 939 }
 940
 941 #ifdef HAVE_SECCOMP
 942
 943 static int apply_seccomp(const ExecContext *c) {
 944         uint32_t negative_action, action;
 945         scmp_filter_ctx *seccomp;
 946         Iterator i;
 947         void *id;
 948         int r;
 949
 950         assert(c);
 951
 952         negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
 953
 954         seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW);
 955         if (!seccomp)
 956                 return -ENOMEM;
 957
 958         if (c->syscall_archs) {
 959
 960                 SET_FOREACH(id, c->syscall_archs, i) {
 961                         r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
 962                         if (r == -EEXIST)
 963                                 continue;
 964                         if (r < 0)
 965                                 goto finish;
 966                 }
 967
 968         } else {
 969                 r = seccomp_add_secondary_archs(seccomp);
 970                 if (r < 0)
 971                         goto finish;
 972         }
 973
 974         action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action;
 975         SET_FOREACH(id, c->syscall_filter, i) {
 976                 r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0);
 977                 if (r < 0)
 978                         goto finish;
 979         }
 980
 981         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
 982         if (r < 0)
 983                 goto finish;
 984
 985         r = seccomp_load(seccomp);
 986
 987 finish:
 988         seccomp_release(seccomp);
 989         return r;
 990 }
 991
 992 static int apply_address_families(const ExecContext *c) {
 993         scmp_filter_ctx *seccomp;
 994         Iterator i;
 995         int r;
 996
 997         assert(c);
 998
 999         seccomp = seccomp_init(SCMP_ACT_ALLOW);
1000         if (!seccomp)
1001                 return -ENOMEM;
1002
1003         r = seccomp_add_secondary_archs(seccomp);
1004         if (r < 0)
1005                 goto finish;
1006
1007         if (c->address_families_whitelist) {
1008                 int af, first = 0, last = 0;
1009                 void *afp;
1010
1011                 /* If this is a whitelist, we first block the address
1012                  * families that are out of range and then everything
1013                  * that is not in the set. First, we find the lowest
1014                  * and highest address family in the set. */
1015
1016                 SET_FOREACH(afp, c->address_families, i) {
1017                         af = PTR_TO_INT(afp);
1018
1019                         if (af <= 0 || af >= af_max())
1020                                 continue;
1021
1022                         if (first == 0 || af < first)
1023                                 first = af;
1024
1025                         if (last == 0 || af > last)
1026                                 last = af;
1027                 }
1028
1029                 assert((first == 0) == (last == 0));
1030
1031                 if (first == 0) {
1032
1033                         /* No entries in the valid range, block everything */
1034                         r = seccomp_rule_add(
1035                                         seccomp,
1036                                         SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1037                                         SCMP_SYS(socket),
1038                                         0);
1039                         if (r < 0)
1040                                 goto finish;
1041
1042                 } else {
1043
1044                         /* Block everything below the first entry */
1045                         r = seccomp_rule_add(
1046                                         seccomp,
1047                                         SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1048                                         SCMP_SYS(socket),
1049                                         1,
1050                                         SCMP_A0(SCMP_CMP_LT, first));
1051                         if (r < 0)
1052                                 goto finish;
1053
1054                         /* Block everything above the last entry */
1055                         r = seccomp_rule_add(
1056                                         seccomp,
1057                                         SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1058                                         SCMP_SYS(socket),
1059                                         1,
1060                                         SCMP_A0(SCMP_CMP_GT, last));
1061                         if (r < 0)
1062                                 goto finish;
1063
1064                         /* Block everything between the first and last
1065                          * entry */
1066                         for (af = 1; af < af_max(); af++) {
1067
1068                                 if (set_contains(c->address_families, INT_TO_PTR(af)))
1069                                         continue;
1070
1071                                 r = seccomp_rule_add(
1072                                                 seccomp,
1073                                                 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1074                                                 SCMP_SYS(socket),
1075                                                 1,
1076                                                 SCMP_A0(SCMP_CMP_EQ, af));
1077                                 if (r < 0)
1078                                         goto finish;
1079                         }
1080                 }
1081
1082         } else {
1083                 void *af;
1084
1085                 /* If this is a blacklist, then generate one rule for
1086                  * each address family that are then combined in OR
1087                  * checks. */
1088
1089                 SET_FOREACH(af, c->address_families, i) {
1090
1091                         r = seccomp_rule_add(
1092                                         seccomp,
1093                                         SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1094                                         SCMP_SYS(socket),
1095                                         1,
1096                                         SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af)));
1097                         if (r < 0)
1098                                 goto finish;
1099                 }
1100         }
1101
1102         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1103         if (r < 0)
1104                 goto finish;
1105
1106         r = seccomp_load(seccomp);
1107
1108 finish:
1109         seccomp_release(seccomp);
1110         return r;
1111 }
1112
1113 #endif
1114
1115 static void do_idle_pipe_dance(int idle_pipe[4]) {
1116         assert(idle_pipe);
1117
1118
1119         safe_close(idle_pipe[1]);
1120         safe_close(idle_pipe[2]);
1121
1122         if (idle_pipe[0] >= 0) {
1123                 int r;
1124
1125                 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1126
1127                 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
1128                         /* Signal systemd that we are bored and want to continue. */
1129                         write(idle_pipe[3], "x", 1);
1130
1131                         /* Wait for systemd to react to the signal above. */
1132                         fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
1133                 }
1134
1135                 safe_close(idle_pipe[0]);
1136
1137         }
1138
1139         safe_close(idle_pipe[3]);
1140 }
1141
1142 static int build_environment(
1143                 const ExecContext *c,
1144                 unsigned n_fds,
1145                 usec_t watchdog_usec,
1146                 const char *home,
1147                 const char *username,
1148                 const char *shell,
1149                 char ***ret) {
1150
1151         _cleanup_strv_free_ char **our_env = NULL;
1152         unsigned n_env = 0;
1153         char *x;
1154
1155         assert(c);
1156         assert(ret);
1157
1158         our_env = new0(char*, 10);
1159         if (!our_env)
1160                 return -ENOMEM;
1161
1162         if (n_fds > 0) {
1163                 if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
1164                         return -ENOMEM;
1165                 our_env[n_env++] = x;
1166
1167                 if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
1168                         return -ENOMEM;
1169                 our_env[n_env++] = x;
1170         }
1171
1172         if (watchdog_usec > 0) {
1173                 if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
1174                         return -ENOMEM;
1175                 our_env[n_env++] = x;
1176
1177                 if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, watchdog_usec) < 0)
1178                         return -ENOMEM;
1179                 our_env[n_env++] = x;
1180         }
1181
1182         if (home) {
1183                 x = strappend("HOME=", home);
1184                 if (!x)
1185                         return -ENOMEM;
1186                 our_env[n_env++] = x;
1187         }
1188
1189         if (username) {
1190                 x = strappend("LOGNAME=", username);
1191                 if (!x)
1192                         return -ENOMEM;
1193                 our_env[n_env++] = x;
1194
1195                 x = strappend("USER=", username);
1196                 if (!x)
1197                         return -ENOMEM;
1198                 our_env[n_env++] = x;
1199         }
1200
1201         if (shell) {
1202                 x = strappend("SHELL=", shell);
1203                 if (!x)
1204                         return -ENOMEM;
1205                 our_env[n_env++] = x;
1206         }
1207
1208         if (is_terminal_input(c->std_input) ||
1209             c->std_output == EXEC_OUTPUT_TTY ||
1210             c->std_error == EXEC_OUTPUT_TTY ||
1211             c->tty_path) {
1212
1213                 x = strdup(default_term_for_tty(tty_path(c)));
1214                 if (!x)
1215                         return -ENOMEM;
1216                 our_env[n_env++] = x;
1217         }
1218
1219         our_env[n_env++] = NULL;
1220         assert(n_env <= 10);
1221
1222         *ret = our_env;
1223         our_env = NULL;
1224
1225         return 0;
1226 }
1227
1228 static int exec_child(ExecCommand *command,
1229                       const ExecContext *context,
1230                       const ExecParameters *params,
1231                       ExecRuntime *runtime,
1232                       char **argv,
1233                       int socket_fd,
1234                       int *fds, unsigned n_fds,
1235                       char **files_env,
1236                       int *error) {
1237
1238         _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1239         const char *username = NULL, *home = NULL, *shell = NULL;
1240         unsigned n_dont_close = 0;
1241         int dont_close[n_fds + 4];
1242         uid_t uid = (uid_t) -1;
1243         gid_t gid = (gid_t) -1;
1244         int i, err;
1245
1246         assert(command);
1247         assert(context);
1248         assert(params);
1249         assert(error);
1250
1251         rename_process_from_path(command->path);
1252
1253         /* We reset exactly these signals, since they are the
1254          * only ones we set to SIG_IGN in the main daemon. All
1255          * others we leave untouched because we set them to
1256          * SIG_DFL or a valid handler initially, both of which
1257          * will be demoted to SIG_DFL. */
1258         default_signals(SIGNALS_CRASH_HANDLER,
1259                         SIGNALS_IGNORE, -1);
1260
1261         if (context->ignore_sigpipe)
1262                 ignore_signals(SIGPIPE, -1);
1263
1264         err = reset_signal_mask();
1265         if (err < 0) {
1266                 *error = EXIT_SIGNAL_MASK;
1267                 return err;
1268         }
1269
1270         if (params->idle_pipe)
1271                 do_idle_pipe_dance(params->idle_pipe);
1272
1273         /* Close sockets very early to make sure we don't
1274          * block init reexecution because it cannot bind its
1275          * sockets */
1276         log_forget_fds();
1277
1278         if (socket_fd >= 0)
1279                 dont_close[n_dont_close++] = socket_fd;
1280         if (n_fds > 0) {
1281                 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
1282                 n_dont_close += n_fds;
1283         }
1284         if (params->bus_endpoint_fd >= 0)
1285                 dont_close[n_dont_close++] = params->bus_endpoint_fd;
1286         if (runtime) {
1287                 if (runtime->netns_storage_socket[0] >= 0)
1288                         dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
1289                 if (runtime->netns_storage_socket[1] >= 0)
1290                         dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
1291         }
1292
1293         err = close_all_fds(dont_close, n_dont_close);
1294         if (err < 0) {
1295                 *error = EXIT_FDS;
1296                 return err;
1297         }
1298
1299         if (!context->same_pgrp)
1300                 if (setsid() < 0) {
1301                         *error = EXIT_SETSID;
1302                         return -errno;
1303                 }
1304
1305         exec_context_tty_reset(context);
1306
1307         if (params->confirm_spawn) {
1308                 char response;
1309
1310                 err = ask_for_confirmation(&response, argv);
1311                 if (err == -ETIMEDOUT)
1312                         write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1313                 else if (err < 0)
1314                         write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1315                 else if (response == 's') {
1316                         write_confirm_message("Skipping execution.\n");
1317                         *error = EXIT_CONFIRM;
1318                         return -ECANCELED;
1319                 } else if (response == 'n') {
1320                         write_confirm_message("Failing execution.\n");
1321                         *error = 0;
1322                         return 0;
1323                 }
1324         }
1325
1326         /* If a socket is connected to STDIN/STDOUT/STDERR, we
1327          * must sure to drop O_NONBLOCK */
1328         if (socket_fd >= 0)
1329                 fd_nonblock(socket_fd, false);
1330
1331         err = setup_input(context, socket_fd, params->apply_tty_stdin);
1332         if (err < 0) {
1333                 *error = EXIT_STDIN;
1334                 return err;
1335         }
1336
1337         err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin);
1338         if (err < 0) {
1339                 *error = EXIT_STDOUT;
1340                 return err;
1341         }
1342
1343         err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin);
1344         if (err < 0) {
1345                 *error = EXIT_STDERR;
1346                 return err;
1347         }
1348
1349         if (params->cgroup_path) {
1350                 err = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0);
1351                 if (err < 0) {
1352                         *error = EXIT_CGROUP;
1353                         return err;
1354                 }
1355         }
1356
1357         if (context->oom_score_adjust_set) {
1358                 char t[16];
1359
1360                 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1361                 char_array_0(t);
1362
1363                 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1364                         *error = EXIT_OOM_ADJUST;
1365                         return -errno;
1366                 }
1367         }
1368
1369         if (context->nice_set)
1370                 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1371                         *error = EXIT_NICE;
1372                         return -errno;
1373                 }
1374
1375         if (context->cpu_sched_set) {
1376                 struct sched_param param = {
1377                         .sched_priority = context->cpu_sched_priority,
1378                 };
1379
1380                 err = sched_setscheduler(0,
1381                                          context->cpu_sched_policy |
1382                                          (context->cpu_sched_reset_on_fork ?
1383                                           SCHED_RESET_ON_FORK : 0),
1384                                          &param);
1385                 if (err < 0) {
1386                         *error = EXIT_SETSCHEDULER;
1387                         return -errno;
1388                 }
1389         }
1390
1391         if (context->cpuset)
1392                 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1393                         *error = EXIT_CPUAFFINITY;
1394                         return -errno;
1395                 }
1396
1397         if (context->ioprio_set)
1398                 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1399                         *error = EXIT_IOPRIO;
1400                         return -errno;
1401                 }
1402
1403         if (context->timer_slack_nsec != NSEC_INFINITY)
1404                 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1405                         *error = EXIT_TIMERSLACK;
1406                         return -errno;
1407                 }
1408
1409         if (context->personality != 0xffffffffUL)
1410                 if (personality(context->personality) < 0) {
1411                         *error = EXIT_PERSONALITY;
1412                         return -errno;
1413                 }
1414
1415         if (context->utmp_id)
1416                 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1417
1418         if (context->user) {
1419                 username = context->user;
1420                 err = get_user_creds(&username, &uid, &gid, &home, &shell);
1421                 if (err < 0) {
1422                         *error = EXIT_USER;
1423                         return err;
1424                 }
1425
1426                 if (is_terminal_input(context->std_input)) {
1427                         err = chown_terminal(STDIN_FILENO, uid);
1428                         if (err < 0) {
1429                                 *error = EXIT_STDIN;
1430                                 return err;
1431                         }
1432                 }
1433         }
1434
1435 #ifdef ENABLE_KDBUS
1436         if (params->bus_endpoint_fd >= 0 && context->bus_endpoint) {
1437                 uid_t ep_uid = (uid == (uid_t) -1) ? 0 : uid;
1438
1439                 err = bus_kernel_set_endpoint_policy(params->bus_endpoint_fd, ep_uid, context->bus_endpoint);
1440                 if (err < 0) {
1441                         *error = EXIT_BUS_ENDPOINT;
1442                         return err;
1443                 }
1444         }
1445 #endif
1446
1447 #ifdef HAVE_PAM
1448         if (params->cgroup_path && context->user && context->pam_name) {
1449                 err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid);
1450                 if (err < 0) {
1451                         *error = EXIT_CGROUP;
1452                         return err;
1453                 }
1454
1455
1456                 err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid);
1457                 if (err < 0) {
1458                         *error = EXIT_CGROUP;
1459                         return err;
1460                 }
1461         }
1462 #endif
1463
1464         if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) {
1465                 char **rt;
1466
1467                 STRV_FOREACH(rt, context->runtime_directory) {
1468                         _cleanup_free_ char *p;
1469
1470                         p = strjoin(params->runtime_prefix, "/", *rt, NULL);
1471                         if (!p) {
1472                                 *error = EXIT_RUNTIME_DIRECTORY;
1473                                 return -ENOMEM;
1474                         }
1475
1476                         err = mkdir_safe(p, context->runtime_directory_mode, uid, gid);
1477                         if (err < 0) {
1478                                 *error = EXIT_RUNTIME_DIRECTORY;
1479                                 return err;
1480                         }
1481                 }
1482         }
1483
1484         if (params->apply_permissions) {
1485                 err = enforce_groups(context, username, gid);
1486                 if (err < 0) {
1487                         *error = EXIT_GROUP;
1488                         return err;
1489                 }
1490         }
1491
1492         umask(context->umask);
1493
1494 #ifdef HAVE_PAM
1495         if (params->apply_permissions && context->pam_name && username) {
1496                 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1497                 if (err < 0) {
1498                         *error = EXIT_PAM;
1499                         return err;
1500                 }
1501         }
1502 #endif
1503
1504         if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
1505                 err = setup_netns(runtime->netns_storage_socket);
1506                 if (err < 0) {
1507                         *error = EXIT_NETWORK;
1508                         return err;
1509                 }
1510         }
1511
1512         if (!strv_isempty(context->read_write_dirs) ||
1513             !strv_isempty(context->read_only_dirs) ||
1514             !strv_isempty(context->inaccessible_dirs) ||
1515             context->mount_flags != 0 ||
1516             (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
1517             params->bus_endpoint_path ||
1518             context->private_devices ||
1519             context->protect_system != PROTECT_SYSTEM_NO ||
1520             context->protect_home != PROTECT_HOME_NO) {
1521
1522                 char *tmp = NULL, *var = NULL;
1523
1524                 /* The runtime struct only contains the parent
1525                  * of the private /tmp, which is
1526                  * non-accessible to world users. Inside of it
1527                  * there's a /tmp that is sticky, and that's
1528                  * the one we want to use here. */
1529
1530                 if (context->private_tmp && runtime) {
1531                         if (runtime->tmp_dir)
1532                                 tmp = strappenda(runtime->tmp_dir, "/tmp");
1533                         if (runtime->var_tmp_dir)
1534                                 var = strappenda(runtime->var_tmp_dir, "/tmp");
1535                 }
1536
1537                 err = setup_namespace(
1538                                 context->read_write_dirs,
1539                                 context->read_only_dirs,
1540                                 context->inaccessible_dirs,
1541                                 tmp,
1542                                 var,
1543                                 params->bus_endpoint_path,
1544                                 context->private_devices,
1545                                 context->protect_home,
1546                                 context->protect_system,
1547                                 context->mount_flags);
1548
1549                 if (err == -EPERM)
1550                         log_error_unit(params->unit_id, "Failed to setup namespace, ignoring: %s", strerror(-err));
1551                 else if (err < 0) {
1552                         *error = EXIT_NAMESPACE;
1553                         return err;
1554                 }
1555         }
1556
1557         if (params->apply_chroot) {
1558                 if (context->root_directory)
1559                         if (chroot(context->root_directory) < 0) {
1560                                 *error = EXIT_CHROOT;
1561                                 return -errno;
1562                         }
1563
1564                 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1565                         *error = EXIT_CHDIR;
1566                         return -errno;
1567                 }
1568         } else {
1569                 _cleanup_free_ char *d = NULL;
1570
1571                 if (asprintf(&d, "%s/%s",
1572                              context->root_directory ? context->root_directory : "",
1573                              context->working_directory ? context->working_directory : "") < 0) {
1574                         *error = EXIT_MEMORY;
1575                         return -ENOMEM;
1576                 }
1577
1578                 if (chdir(d) < 0) {
1579                         *error = EXIT_CHDIR;
1580                         return -errno;
1581                 }
1582         }
1583
1584         /* We repeat the fd closing here, to make sure that
1585          * nothing is leaked from the PAM modules. Note that
1586          * we are more aggressive this time since socket_fd
1587          * and the netns fds we don't need anymore. The custom
1588          * endpoint fd was needed to upload the policy and can
1589          * now be closed as well. */
1590         err = close_all_fds(fds, n_fds);
1591         if (err >= 0)
1592                 err = shift_fds(fds, n_fds);
1593         if (err >= 0)
1594                 err = flags_fds(fds, n_fds, context->non_blocking);
1595         if (err < 0) {
1596                 *error = EXIT_FDS;
1597                 return err;
1598         }
1599
1600         if (params->apply_permissions) {
1601
1602                 for (i = 0; i < _RLIMIT_MAX; i++) {
1603                         if (!context->rlimit[i])
1604                                 continue;
1605
1606                         if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1607                                 *error = EXIT_LIMITS;
1608                                 return -errno;
1609                         }
1610                 }
1611
1612                 if (context->capability_bounding_set_drop) {
1613                         err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1614                         if (err < 0) {
1615                                 *error = EXIT_CAPABILITIES;
1616                                 return err;
1617                         }
1618                 }
1619
1620                 if (context->user) {
1621                         err = enforce_user(context, uid);
1622                         if (err < 0) {
1623                                 *error = EXIT_USER;
1624                                 return err;
1625                         }
1626                 }
1627
1628                 /* PR_GET_SECUREBITS is not privileged, while
1629                  * PR_SET_SECUREBITS is. So to suppress
1630                  * potential EPERMs we'll try not to call
1631                  * PR_SET_SECUREBITS unless necessary. */
1632                 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1633                         if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1634                                 *error = EXIT_SECUREBITS;
1635                                 return -errno;
1636                         }
1637
1638                 if (context->capabilities)
1639                         if (cap_set_proc(context->capabilities) < 0) {
1640                                 *error = EXIT_CAPABILITIES;
1641                                 return -errno;
1642                         }
1643
1644                 if (context->no_new_privileges)
1645                         if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1646                                 *error = EXIT_NO_NEW_PRIVILEGES;
1647                                 return -errno;
1648                         }
1649
1650 #ifdef HAVE_SECCOMP
1651                 if (context->address_families_whitelist ||
1652                     !set_isempty(context->address_families)) {
1653                         err = apply_address_families(context);
1654                         if (err < 0) {
1655                                 *error = EXIT_ADDRESS_FAMILIES;
1656                                 return err;
1657                         }
1658                 }
1659
1660                 if (context->syscall_whitelist ||
1661                     !set_isempty(context->syscall_filter) ||
1662                     !set_isempty(context->syscall_archs)) {
1663                         err = apply_seccomp(context);
1664                         if (err < 0) {
1665                                 *error = EXIT_SECCOMP;
1666                                 return err;
1667                         }
1668                 }
1669 #endif
1670
1671 #ifdef HAVE_SELINUX
1672                 if (use_selinux()) {
1673                         if (context->selinux_context) {
1674                                 err = setexeccon(context->selinux_context);
1675                                 if (err < 0 && !context->selinux_context_ignore) {
1676                                         *error = EXIT_SELINUX_CONTEXT;
1677                                         return err;
1678                                 }
1679                         }
1680
1681                         if (params->selinux_context_net && socket_fd >= 0) {
1682                                 _cleanup_free_ char *label = NULL;
1683
1684                                 err = label_get_child_mls_label(socket_fd, command->path, &label);
1685                                 if (err < 0) {
1686                                         *error = EXIT_SELINUX_CONTEXT;
1687                                         return err;
1688                                 }
1689
1690                                 err = setexeccon(label);
1691                                 if (err < 0) {
1692                                         *error = EXIT_SELINUX_CONTEXT;
1693                                         return err;
1694                                 }
1695                         }
1696                 }
1697 #endif
1698
1699 #ifdef HAVE_APPARMOR
1700                 if (context->apparmor_profile && use_apparmor()) {
1701                         err = aa_change_onexec(context->apparmor_profile);
1702                         if (err < 0 && !context->apparmor_profile_ignore) {
1703                                 *error = EXIT_APPARMOR_PROFILE;
1704                                 return -errno;
1705                         }
1706                 }
1707 #endif
1708         }
1709
1710         err = build_environment(context, n_fds, params->watchdog_usec, home, username, shell, &our_env);
1711         if (err < 0) {
1712                 *error = EXIT_MEMORY;
1713                 return err;
1714         }
1715
1716         final_env = strv_env_merge(5,
1717                                    params->environment,
1718                                    our_env,
1719                                    context->environment,
1720                                    files_env,
1721                                    pam_env,
1722                                    NULL);
1723         if (!final_env) {
1724                 *error = EXIT_MEMORY;
1725                 return -ENOMEM;
1726         }
1727
1728         final_argv = replace_env_argv(argv, final_env);
1729         if (!final_argv) {
1730                 *error = EXIT_MEMORY;
1731                 return -ENOMEM;
1732         }
1733
1734         final_env = strv_env_clean(final_env);
1735
1736         if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
1737                 _cleanup_free_ char *line;
1738
1739                 line = exec_command_line(final_argv);
1740                 if (line) {
1741                         log_open();
1742                         log_struct_unit(LOG_DEBUG,
1743                                         params->unit_id,
1744                                         "EXECUTABLE=%s", command->path,
1745                                         "MESSAGE=Executing: %s", line,
1746                                         NULL);
1747                         log_close();
1748                 }
1749         }
1750         execve(command->path, final_argv, final_env);
1751         *error = EXIT_EXEC;
1752         return -errno;
1753 }
1754
1755 int exec_spawn(ExecCommand *command,
1756                const ExecContext *context,
1757                const ExecParameters *params,
1758                ExecRuntime *runtime,
1759                pid_t *ret) {
1760
1761         _cleanup_strv_free_ char **files_env = NULL;
1762         int *fds = NULL; unsigned n_fds = 0;
1763         char *line, **argv;
1764         int socket_fd;
1765         pid_t pid;
1766         int err;
1767
1768         assert(command);
1769         assert(context);
1770         assert(ret);
1771         assert(params);
1772         assert(params->fds || params->n_fds <= 0);
1773
1774         if (context->std_input == EXEC_INPUT_SOCKET ||
1775             context->std_output == EXEC_OUTPUT_SOCKET ||
1776             context->std_error == EXEC_OUTPUT_SOCKET) {
1777
1778                 if (params->n_fds != 1)
1779                         return -EINVAL;
1780
1781                 socket_fd = params->fds[0];
1782         } else {
1783                 socket_fd = -1;
1784                 fds = params->fds;
1785                 n_fds = params->n_fds;
1786         }
1787
1788         err = exec_context_load_environment(context, &files_env);
1789         if (err < 0) {
1790                 log_struct_unit(LOG_ERR,
1791                            params->unit_id,
1792                            "MESSAGE=Failed to load environment files: %s", strerror(-err),
1793                            "ERRNO=%d", -err,
1794                            NULL);
1795                 return err;
1796         }
1797
1798         argv = params->argv ?: command->argv;
1799
1800         line = exec_command_line(argv);
1801         if (!line)
1802                 return log_oom();
1803
1804         log_struct_unit(LOG_DEBUG,
1805                         params->unit_id,
1806                         "EXECUTABLE=%s", command->path,
1807                         "MESSAGE=About to execute: %s", line,
1808                         NULL);
1809         free(line);
1810
1811         pid = fork();
1812         if (pid < 0)
1813                 return -errno;
1814
1815         if (pid == 0) {
1816                 int r;
1817
1818                 err = exec_child(command,
1819                                  context,
1820                                  params,
1821                                  runtime,
1822                                  argv,
1823                                  socket_fd,
1824                                  fds, n_fds,
1825                                  files_env,
1826                                  &r);
1827                 if (r != 0) {
1828                         log_open();
1829                         log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1830                                    "EXECUTABLE=%s", command->path,
1831                                    "MESSAGE=Failed at step %s spawning %s: %s",
1832                                           exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1833                                           command->path, strerror(-err),
1834                                    "ERRNO=%d", -err,
1835                                    NULL);
1836                         log_close();
1837                 }
1838
1839                 _exit(r);
1840         }
1841
1842         log_struct_unit(LOG_DEBUG,
1843                         params->unit_id,
1844                         "MESSAGE=Forked %s as "PID_FMT,
1845                         command->path, pid,
1846                         NULL);
1847
1848         /* We add the new process to the cgroup both in the child (so
1849          * that we can be sure that no user code is ever executed
1850          * outside of the cgroup) and in the parent (so that we can be
1851          * sure that when we kill the cgroup the process will be
1852          * killed too). */
1853         if (params->cgroup_path)
1854                 cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid);
1855
1856         exec_status_start(&command->exec_status, pid);
1857
1858         *ret = pid;
1859         return 0;
1860 }
1861
1862 void exec_context_init(ExecContext *c) {
1863         assert(c);
1864
1865         c->umask = 0022;
1866         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1867         c->cpu_sched_policy = SCHED_OTHER;
1868         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1869         c->syslog_level_prefix = true;
1870         c->ignore_sigpipe = true;
1871         c->timer_slack_nsec = NSEC_INFINITY;
1872         c->personality = 0xffffffffUL;
1873         c->runtime_directory_mode = 0755;
1874 }
1875
1876 void exec_context_done(ExecContext *c) {
1877         unsigned l;
1878
1879         assert(c);
1880
1881         strv_free(c->environment);
1882         c->environment = NULL;
1883
1884         strv_free(c->environment_files);
1885         c->environment_files = NULL;
1886
1887         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1888                 free(c->rlimit[l]);
1889                 c->rlimit[l] = NULL;
1890         }
1891
1892         free(c->working_directory);
1893         c->working_directory = NULL;
1894         free(c->root_directory);
1895         c->root_directory = NULL;
1896
1897         free(c->tty_path);
1898         c->tty_path = NULL;
1899
1900         free(c->syslog_identifier);
1901         c->syslog_identifier = NULL;
1902
1903         free(c->user);
1904         c->user = NULL;
1905
1906         free(c->group);
1907         c->group = NULL;
1908
1909         strv_free(c->supplementary_groups);
1910         c->supplementary_groups = NULL;
1911
1912         free(c->pam_name);
1913         c->pam_name = NULL;
1914
1915         if (c->capabilities) {
1916                 cap_free(c->capabilities);
1917                 c->capabilities = NULL;
1918         }
1919
1920         strv_free(c->read_only_dirs);
1921         c->read_only_dirs = NULL;
1922
1923         strv_free(c->read_write_dirs);
1924         c->read_write_dirs = NULL;
1925
1926         strv_free(c->inaccessible_dirs);
1927         c->inaccessible_dirs = NULL;
1928
1929         if (c->cpuset)
1930                 CPU_FREE(c->cpuset);
1931
1932         free(c->utmp_id);
1933         c->utmp_id = NULL;
1934
1935         free(c->selinux_context);
1936         c->selinux_context = NULL;
1937
1938         free(c->apparmor_profile);
1939         c->apparmor_profile = NULL;
1940
1941         set_free(c->syscall_filter);
1942         c->syscall_filter = NULL;
1943
1944         set_free(c->syscall_archs);
1945         c->syscall_archs = NULL;
1946
1947         set_free(c->address_families);
1948         c->address_families = NULL;
1949
1950         strv_free(c->runtime_directory);
1951         c->runtime_directory = NULL;
1952
1953         bus_endpoint_free(c->bus_endpoint);
1954         c->bus_endpoint = NULL;
1955 }
1956
1957 int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
1958         char **i;
1959
1960         assert(c);
1961
1962         if (!runtime_prefix)
1963                 return 0;
1964
1965         STRV_FOREACH(i, c->runtime_directory) {
1966                 _cleanup_free_ char *p;
1967
1968                 p = strjoin(runtime_prefix, "/", *i, NULL);
1969                 if (!p)
1970                         return -ENOMEM;
1971
1972                 /* We execute this synchronously, since we need to be
1973                  * sure this is gone when we start the service
1974                  * next. */
1975                 rm_rf_dangerous(p, false, true, false);
1976         }
1977
1978         return 0;
1979 }
1980
1981 void exec_command_done(ExecCommand *c) {
1982         assert(c);
1983
1984         free(c->path);
1985         c->path = NULL;
1986
1987         strv_free(c->argv);
1988         c->argv = NULL;
1989 }
1990
1991 void exec_command_done_array(ExecCommand *c, unsigned n) {
1992         unsigned i;
1993
1994         for (i = 0; i < n; i++)
1995                 exec_command_done(c+i);
1996 }
1997
1998 void exec_command_free_list(ExecCommand *c) {
1999         ExecCommand *i;
2000
2001         while ((i = c)) {
2002                 LIST_REMOVE(command, c, i);
2003                 exec_command_done(i);
2004                 free(i);
2005         }
2006 }
2007
2008 void exec_command_free_array(ExecCommand **c, unsigned n) {
2009         unsigned i;
2010
2011         for (i = 0; i < n; i++) {
2012                 exec_command_free_list(c[i]);
2013                 c[i] = NULL;
2014         }
2015 }
2016
2017 int exec_context_load_environment(const ExecContext *c, char ***l) {
2018         char **i, **r = NULL;
2019
2020         assert(c);
2021         assert(l);
2022
2023         STRV_FOREACH(i, c->environment_files) {
2024                 char *fn;
2025                 int k;
2026                 bool ignore = false;
2027                 char **p;
2028                 _cleanup_globfree_ glob_t pglob = {};
2029                 int count, n;
2030
2031                 fn = *i;
2032
2033                 if (fn[0] == '-') {
2034                         ignore = true;
2035                         fn ++;
2036                 }
2037
2038                 if (!path_is_absolute(fn)) {
2039                         if (ignore)
2040                                 continue;
2041
2042                         strv_free(r);
2043                         return -EINVAL;
2044                 }
2045
2046                 /* Filename supports globbing, take all matching files */
2047                 errno = 0;
2048                 if (glob(fn, 0, NULL, &pglob) != 0) {
2049                         if (ignore)
2050                                 continue;
2051
2052                         strv_free(r);
2053                         return errno ? -errno : -EINVAL;
2054                 }
2055                 count = pglob.gl_pathc;
2056                 if (count == 0) {
2057                         if (ignore)
2058                                 continue;
2059
2060                         strv_free(r);
2061                         return -EINVAL;
2062                 }
2063                 for (n = 0; n < count; n++) {
2064                         k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p);
2065                         if (k < 0) {
2066                                 if (ignore)
2067                                         continue;
2068
2069                                 strv_free(r);
2070                                 return k;
2071                         }
2072                         /* Log invalid environment variables with filename */
2073                         if (p)
2074                                 p = strv_env_clean_log(p, pglob.gl_pathv[n]);
2075
2076                         if (r == NULL)
2077                                 r = p;
2078                         else {
2079                                 char **m;
2080
2081                                 m = strv_env_merge(2, r, p);
2082                                 strv_free(r);
2083                                 strv_free(p);
2084                                 if (!m)
2085                                         return -ENOMEM;
2086
2087                                 r = m;
2088                         }
2089                 }
2090         }
2091
2092         *l = r;
2093
2094         return 0;
2095 }
2096
2097 static bool tty_may_match_dev_console(const char *tty) {
2098         _cleanup_free_ char *active = NULL;
2099        char *console;
2100
2101         if (startswith(tty, "/dev/"))
2102                 tty += 5;
2103
2104         /* trivial identity? */
2105         if (streq(tty, "console"))
2106                 return true;
2107
2108         console = resolve_dev_console(&active);
2109         /* if we could not resolve, assume it may */
2110         if (!console)
2111                 return true;
2112
2113         /* "tty0" means the active VC, so it may be the same sometimes */
2114         return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
2115 }
2116
2117 bool exec_context_may_touch_console(ExecContext *ec) {
2118         return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
2119                 is_terminal_input(ec->std_input) ||
2120                 is_terminal_output(ec->std_output) ||
2121                 is_terminal_output(ec->std_error)) &&
2122                tty_may_match_dev_console(tty_path(ec));
2123 }
2124
2125 static void strv_fprintf(FILE *f, char **l) {
2126         char **g;
2127
2128         assert(f);
2129
2130         STRV_FOREACH(g, l)
2131                 fprintf(f, " %s", *g);
2132 }
2133
2134 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
2135         char **e;
2136         unsigned i;
2137
2138         assert(c);
2139         assert(f);
2140
2141         prefix = strempty(prefix);
2142
2143         fprintf(f,
2144                 "%sUMask: %04o\n"
2145                 "%sWorkingDirectory: %s\n"
2146                 "%sRootDirectory: %s\n"
2147                 "%sNonBlocking: %s\n"
2148                 "%sPrivateTmp: %s\n"
2149                 "%sPrivateNetwork: %s\n"
2150                 "%sPrivateDevices: %s\n"
2151                 "%sProtectHome: %s\n"
2152                 "%sProtectSystem: %s\n"
2153                 "%sIgnoreSIGPIPE: %s\n",
2154                 prefix, c->umask,
2155                 prefix, c->working_directory ? c->working_directory : "/",
2156                 prefix, c->root_directory ? c->root_directory : "/",
2157                 prefix, yes_no(c->non_blocking),
2158                 prefix, yes_no(c->private_tmp),
2159                 prefix, yes_no(c->private_network),
2160                 prefix, yes_no(c->private_devices),
2161                 prefix, protect_home_to_string(c->protect_home),
2162                 prefix, protect_system_to_string(c->protect_system),
2163                 prefix, yes_no(c->ignore_sigpipe));
2164
2165         STRV_FOREACH(e, c->environment)
2166                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
2167
2168         STRV_FOREACH(e, c->environment_files)
2169                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
2170
2171         if (c->nice_set)
2172                 fprintf(f,
2173                         "%sNice: %i\n",
2174                         prefix, c->nice);
2175
2176         if (c->oom_score_adjust_set)
2177                 fprintf(f,
2178                         "%sOOMScoreAdjust: %i\n",
2179                         prefix, c->oom_score_adjust);
2180
2181         for (i = 0; i < RLIM_NLIMITS; i++)
2182                 if (c->rlimit[i])
2183                         fprintf(f, "%s%s: "RLIM_FMT"\n",
2184                                 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
2185
2186         if (c->ioprio_set) {
2187                 _cleanup_free_ char *class_str = NULL;
2188
2189                 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
2190                 fprintf(f,
2191                         "%sIOSchedulingClass: %s\n"
2192                         "%sIOPriority: %i\n",
2193                         prefix, strna(class_str),
2194                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
2195         }
2196
2197         if (c->cpu_sched_set) {
2198                 _cleanup_free_ char *policy_str = NULL;
2199
2200                 sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
2201                 fprintf(f,
2202                         "%sCPUSchedulingPolicy: %s\n"
2203                         "%sCPUSchedulingPriority: %i\n"
2204                         "%sCPUSchedulingResetOnFork: %s\n",
2205                         prefix, strna(policy_str),
2206                         prefix, c->cpu_sched_priority,
2207                         prefix, yes_no(c->cpu_sched_reset_on_fork));
2208         }
2209
2210         if (c->cpuset) {
2211                 fprintf(f, "%sCPUAffinity:", prefix);
2212                 for (i = 0; i < c->cpuset_ncpus; i++)
2213                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
2214                                 fprintf(f, " %u", i);
2215                 fputs("\n", f);
2216         }
2217
2218         if (c->timer_slack_nsec != NSEC_INFINITY)
2219                 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
2220
2221         fprintf(f,
2222                 "%sStandardInput: %s\n"
2223                 "%sStandardOutput: %s\n"
2224                 "%sStandardError: %s\n",
2225                 prefix, exec_input_to_string(c->std_input),
2226                 prefix, exec_output_to_string(c->std_output),
2227                 prefix, exec_output_to_string(c->std_error));
2228
2229         if (c->tty_path)
2230                 fprintf(f,
2231                         "%sTTYPath: %s\n"
2232                         "%sTTYReset: %s\n"
2233                         "%sTTYVHangup: %s\n"
2234                         "%sTTYVTDisallocate: %s\n",
2235                         prefix, c->tty_path,
2236                         prefix, yes_no(c->tty_reset),
2237                         prefix, yes_no(c->tty_vhangup),
2238                         prefix, yes_no(c->tty_vt_disallocate));
2239
2240         if (c->std_output == EXEC_OUTPUT_SYSLOG ||
2241             c->std_output == EXEC_OUTPUT_KMSG ||
2242             c->std_output == EXEC_OUTPUT_JOURNAL ||
2243             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2244             c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2245             c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
2246             c->std_error == EXEC_OUTPUT_SYSLOG ||
2247             c->std_error == EXEC_OUTPUT_KMSG ||
2248             c->std_error == EXEC_OUTPUT_JOURNAL ||
2249             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2250             c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2251             c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
2252
2253                 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
2254
2255                 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2256                 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2257
2258                 fprintf(f,
2259                         "%sSyslogFacility: %s\n"
2260                         "%sSyslogLevel: %s\n",
2261                         prefix, strna(fac_str),
2262                         prefix, strna(lvl_str));
2263         }
2264
2265         if (c->capabilities) {
2266                 _cleanup_cap_free_charp_ char *t;
2267
2268                 t = cap_to_text(c->capabilities, NULL);
2269                 if (t)
2270                         fprintf(f, "%sCapabilities: %s\n", prefix, t);
2271         }
2272
2273         if (c->secure_bits)
2274                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2275                         prefix,
2276                         (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2277                         (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2278                         (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2279                         (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2280                         (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2281                         (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2282
2283         if (c->capability_bounding_set_drop) {
2284                 unsigned long l;
2285                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2286
2287                 for (l = 0; l <= cap_last_cap(); l++)
2288                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2289                                 _cleanup_cap_free_charp_ char *t;
2290
2291                                 t = cap_to_name(l);
2292                                 if (t)
2293                                         fprintf(f, " %s", t);
2294                         }
2295
2296                 fputs("\n", f);
2297         }
2298
2299         if (c->user)
2300                 fprintf(f, "%sUser: %s\n", prefix, c->user);
2301         if (c->group)
2302                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2303
2304         if (strv_length(c->supplementary_groups) > 0) {
2305                 fprintf(f, "%sSupplementaryGroups:", prefix);
2306                 strv_fprintf(f, c->supplementary_groups);
2307                 fputs("\n", f);
2308         }
2309
2310         if (c->pam_name)
2311                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2312
2313         if (strv_length(c->read_write_dirs) > 0) {
2314                 fprintf(f, "%sReadWriteDirs:", prefix);
2315                 strv_fprintf(f, c->read_write_dirs);
2316                 fputs("\n", f);
2317         }
2318
2319         if (strv_length(c->read_only_dirs) > 0) {
2320                 fprintf(f, "%sReadOnlyDirs:", prefix);
2321                 strv_fprintf(f, c->read_only_dirs);
2322                 fputs("\n", f);
2323         }
2324
2325         if (strv_length(c->inaccessible_dirs) > 0) {
2326                 fprintf(f, "%sInaccessibleDirs:", prefix);
2327                 strv_fprintf(f, c->inaccessible_dirs);
2328                 fputs("\n", f);
2329         }
2330
2331         if (c->utmp_id)
2332                 fprintf(f,
2333                         "%sUtmpIdentifier: %s\n",
2334                         prefix, c->utmp_id);
2335
2336         if (c->selinux_context)
2337                 fprintf(f,
2338                         "%sSELinuxContext: %s%s\n",
2339                         prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
2340
2341         if (c->personality != 0xffffffffUL)
2342                 fprintf(f,
2343                         "%sPersonality: %s\n",
2344                         prefix, strna(personality_to_string(c->personality)));
2345
2346         if (c->syscall_filter) {
2347 #ifdef HAVE_SECCOMP
2348                 Iterator j;
2349                 void *id;
2350                 bool first = true;
2351 #endif
2352
2353                 fprintf(f,
2354                         "%sSystemCallFilter: ",
2355                         prefix);
2356
2357                 if (!c->syscall_whitelist)
2358                         fputc('~', f);
2359
2360 #ifdef HAVE_SECCOMP
2361                 SET_FOREACH(id, c->syscall_filter, j) {
2362                         _cleanup_free_ char *name = NULL;
2363
2364                         if (first)
2365                                 first = false;
2366                         else
2367                                 fputc(' ', f);
2368
2369                         name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
2370                         fputs(strna(name), f);
2371                 }
2372 #endif
2373
2374                 fputc('\n', f);
2375         }
2376
2377         if (c->syscall_archs) {
2378 #ifdef HAVE_SECCOMP
2379                 Iterator j;
2380                 void *id;
2381 #endif
2382
2383                 fprintf(f,
2384                         "%sSystemCallArchitectures:",
2385                         prefix);
2386
2387 #ifdef HAVE_SECCOMP
2388                 SET_FOREACH(id, c->syscall_archs, j)
2389                         fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
2390 #endif
2391                 fputc('\n', f);
2392         }
2393
2394         if (c->syscall_errno != 0)
2395                 fprintf(f,
2396                         "%sSystemCallErrorNumber: %s\n",
2397                         prefix, strna(errno_to_name(c->syscall_errno)));
2398
2399         if (c->apparmor_profile)
2400                 fprintf(f,
2401                         "%sAppArmorProfile: %s%s\n",
2402                         prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
2403 }
2404
2405 void exec_status_start(ExecStatus *s, pid_t pid) {
2406         assert(s);
2407
2408         zero(*s);
2409         s->pid = pid;
2410         dual_timestamp_get(&s->start_timestamp);
2411 }
2412
2413 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2414         assert(s);
2415
2416         if (s->pid && s->pid != pid)
2417                 zero(*s);
2418
2419         s->pid = pid;
2420         dual_timestamp_get(&s->exit_timestamp);
2421
2422         s->code = code;
2423         s->status = status;
2424
2425         if (context) {
2426                 if (context->utmp_id)
2427                         utmp_put_dead_process(context->utmp_id, pid, code, status);
2428
2429                 exec_context_tty_reset(context);
2430         }
2431 }
2432
2433 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2434         char buf[FORMAT_TIMESTAMP_MAX];
2435
2436         assert(s);
2437         assert(f);
2438
2439         if (s->pid <= 0)
2440                 return;
2441
2442         prefix = strempty(prefix);
2443
2444         fprintf(f,
2445                 "%sPID: "PID_FMT"\n",
2446                 prefix, s->pid);
2447
2448         if (s->start_timestamp.realtime > 0)
2449                 fprintf(f,
2450                         "%sStart Timestamp: %s\n",
2451                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2452
2453         if (s->exit_timestamp.realtime > 0)
2454                 fprintf(f,
2455                         "%sExit Timestamp: %s\n"
2456                         "%sExit Code: %s\n"
2457                         "%sExit Status: %i\n",
2458                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2459                         prefix, sigchld_code_to_string(s->code),
2460                         prefix, s->status);
2461 }
2462
2463 char *exec_command_line(char **argv) {
2464         size_t k;
2465         char *n, *p, **a;
2466         bool first = true;
2467
2468         assert(argv);
2469
2470         k = 1;
2471         STRV_FOREACH(a, argv)
2472                 k += strlen(*a)+3;
2473
2474         if (!(n = new(char, k)))
2475                 return NULL;
2476
2477         p = n;
2478         STRV_FOREACH(a, argv) {
2479
2480                 if (!first)
2481                         *(p++) = ' ';
2482                 else
2483                         first = false;
2484
2485                 if (strpbrk(*a, WHITESPACE)) {
2486                         *(p++) = '\'';
2487                         p = stpcpy(p, *a);
2488                         *(p++) = '\'';
2489                 } else
2490                         p = stpcpy(p, *a);
2491
2492         }
2493
2494         *p = 0;
2495
2496         /* FIXME: this doesn't really handle arguments that have
2497          * spaces and ticks in them */
2498
2499         return n;
2500 }
2501
2502 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2503         _cleanup_free_ char *cmd = NULL;
2504         const char *prefix2;
2505
2506         assert(c);
2507         assert(f);
2508
2509         prefix = strempty(prefix);
2510         prefix2 = strappenda(prefix, "\t");
2511
2512         cmd = exec_command_line(c->argv);
2513         fprintf(f,
2514                 "%sCommand Line: %s\n",
2515                 prefix, cmd ? cmd : strerror(ENOMEM));
2516
2517         exec_status_dump(&c->exec_status, f, prefix2);
2518 }
2519
2520 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2521         assert(f);
2522
2523         prefix = strempty(prefix);
2524
2525         LIST_FOREACH(command, c, c)
2526                 exec_command_dump(c, f, prefix);
2527 }
2528
2529 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2530         ExecCommand *end;
2531
2532         assert(l);
2533         assert(e);
2534
2535         if (*l) {
2536                 /* It's kind of important, that we keep the order here */
2537                 LIST_FIND_TAIL(command, *l, end);
2538                 LIST_INSERT_AFTER(command, *l, end, e);
2539         } else
2540               *l = e;
2541 }
2542
2543 int exec_command_set(ExecCommand *c, const char *path, ...) {
2544         va_list ap;
2545         char **l, *p;
2546
2547         assert(c);
2548         assert(path);
2549
2550         va_start(ap, path);
2551         l = strv_new_ap(path, ap);
2552         va_end(ap);
2553
2554         if (!l)
2555                 return -ENOMEM;
2556
2557         p = strdup(path);
2558         if (!p) {
2559                 strv_free(l);
2560                 return -ENOMEM;
2561         }
2562
2563         free(c->path);
2564         c->path = p;
2565
2566         strv_free(c->argv);
2567         c->argv = l;
2568
2569         return 0;
2570 }
2571
2572 int exec_command_append(ExecCommand *c, const char *path, ...) {
2573         _cleanup_strv_free_ char **l = NULL;
2574         va_list ap;
2575         int r;
2576
2577         assert(c);
2578         assert(path);
2579
2580         va_start(ap, path);
2581         l = strv_new_ap(path, ap);
2582         va_end(ap);
2583
2584         if (!l)
2585                 return -ENOMEM;
2586
2587         r = strv_extend_strv(&c->argv, l);
2588         if (r < 0)
2589                 return r;
2590
2591         return 0;
2592 }
2593
2594
2595 static int exec_runtime_allocate(ExecRuntime **rt) {
2596
2597         if (*rt)
2598                 return 0;
2599
2600         *rt = new0(ExecRuntime, 1);
2601         if (!*rt)
2602                 return -ENOMEM;
2603
2604         (*rt)->n_ref = 1;
2605         (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2606
2607         return 0;
2608 }
2609
2610 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2611         int r;
2612
2613         assert(rt);
2614         assert(c);
2615         assert(id);
2616
2617         if (*rt)
2618                 return 1;
2619
2620         if (!c->private_network && !c->private_tmp)
2621                 return 0;
2622
2623         r = exec_runtime_allocate(rt);
2624         if (r < 0)
2625                 return r;
2626
2627         if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2628                 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2629                         return -errno;
2630         }
2631
2632         if (c->private_tmp && !(*rt)->tmp_dir) {
2633                 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2634                 if (r < 0)
2635                         return r;
2636         }
2637
2638         return 1;
2639 }
2640
2641 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2642         assert(r);
2643         assert(r->n_ref > 0);
2644
2645         r->n_ref++;
2646         return r;
2647 }
2648
2649 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2650
2651         if (!r)
2652                 return NULL;
2653
2654         assert(r->n_ref > 0);
2655
2656         r->n_ref--;
2657         if (r->n_ref <= 0) {
2658                 free(r->tmp_dir);
2659                 free(r->var_tmp_dir);
2660                 safe_close_pair(r->netns_storage_socket);
2661                 free(r);
2662         }
2663
2664         return NULL;
2665 }
2666
2667 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2668         assert(u);
2669         assert(f);
2670         assert(fds);
2671
2672         if (!rt)
2673                 return 0;
2674
2675         if (rt->tmp_dir)
2676                 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2677
2678         if (rt->var_tmp_dir)
2679                 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2680
2681         if (rt->netns_storage_socket[0] >= 0) {
2682                 int copy;
2683
2684                 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2685                 if (copy < 0)
2686                         return copy;
2687
2688                 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2689         }
2690
2691         if (rt->netns_storage_socket[1] >= 0) {
2692                 int copy;
2693
2694                 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2695                 if (copy < 0)
2696                         return copy;
2697
2698                 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2699         }
2700
2701         return 0;
2702 }
2703
2704 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2705         int r;
2706
2707         assert(rt);
2708         assert(key);
2709         assert(value);
2710
2711         if (streq(key, "tmp-dir")) {
2712                 char *copy;
2713
2714                 r = exec_runtime_allocate(rt);
2715                 if (r < 0)
2716                         return r;
2717
2718                 copy = strdup(value);
2719                 if (!copy)
2720                         return log_oom();
2721
2722                 free((*rt)->tmp_dir);
2723                 (*rt)->tmp_dir = copy;
2724
2725         } else if (streq(key, "var-tmp-dir")) {
2726                 char *copy;
2727
2728                 r = exec_runtime_allocate(rt);
2729                 if (r < 0)
2730                         return r;
2731
2732                 copy = strdup(value);
2733                 if (!copy)
2734                         return log_oom();
2735
2736                 free((*rt)->var_tmp_dir);
2737                 (*rt)->var_tmp_dir = copy;
2738
2739         } else if (streq(key, "netns-socket-0")) {
2740                 int fd;
2741
2742                 r = exec_runtime_allocate(rt);
2743                 if (r < 0)
2744                         return r;
2745
2746                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2747                         log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2748                 else {
2749                         safe_close((*rt)->netns_storage_socket[0]);
2750                         (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2751                 }
2752         } else if (streq(key, "netns-socket-1")) {
2753                 int fd;
2754
2755                 r = exec_runtime_allocate(rt);
2756                 if (r < 0)
2757                         return r;
2758
2759                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2760                         log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2761                 else {
2762                         safe_close((*rt)->netns_storage_socket[1]);
2763                         (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2764                 }
2765         } else
2766                 return 0;
2767
2768         return 1;
2769 }
2770
2771 static void *remove_tmpdir_thread(void *p) {
2772         _cleanup_free_ char *path = p;
2773
2774         rm_rf_dangerous(path, false, true, false);
2775         return NULL;
2776 }
2777
2778 void exec_runtime_destroy(ExecRuntime *rt) {
2779         int r;
2780
2781         if (!rt)
2782                 return;
2783
2784         /* If there are multiple users of this, let's leave the stuff around */
2785         if (rt->n_ref > 1)
2786                 return;
2787
2788         if (rt->tmp_dir) {
2789                 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2790
2791                 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2792                 if (r < 0) {
2793                         log_warning("Failed to nuke %s: %s", rt->tmp_dir, strerror(-r));
2794                         free(rt->tmp_dir);
2795                 }
2796
2797                 rt->tmp_dir = NULL;
2798         }
2799
2800         if (rt->var_tmp_dir) {
2801                 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2802
2803                 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2804                 if (r < 0) {
2805                         log_warning("Failed to nuke %s: %s", rt->var_tmp_dir, strerror(-r));
2806                         free(rt->var_tmp_dir);
2807                 }
2808
2809                 rt->var_tmp_dir = NULL;
2810         }
2811
2812         safe_close_pair(rt->netns_storage_socket);
2813 }
2814
2815 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2816         [EXEC_INPUT_NULL] = "null",
2817         [EXEC_INPUT_TTY] = "tty",
2818         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2819         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2820         [EXEC_INPUT_SOCKET] = "socket"
2821 };
2822
2823 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2824
2825 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2826         [EXEC_OUTPUT_INHERIT] = "inherit",
2827         [EXEC_OUTPUT_NULL] = "null",
2828         [EXEC_OUTPUT_TTY] = "tty",
2829         [EXEC_OUTPUT_SYSLOG] = "syslog",
2830         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2831         [EXEC_OUTPUT_KMSG] = "kmsg",
2832         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2833         [EXEC_OUTPUT_JOURNAL] = "journal",
2834         [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2835         [EXEC_OUTPUT_SOCKET] = "socket"
2836 };
2837
2838 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);