src/core/execute.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <errno.h>
   4 #include <fcntl.h>
   5 #include <poll.h>
   6 #include <sys/eventfd.h>
   7 #include <sys/ioctl.h>
   8 #include <sys/mman.h>
   9 #include <sys/mount.h>
  10 #include <sys/personality.h>
  11 #include <sys/prctl.h>
  12 #include <sys/shm.h>
  13 #include <sys/types.h>
  14 #include <sys/un.h>
  15 #include <unistd.h>
  16 #include <utmpx.h>
  17
  18 #if HAVE_PAM
  19 #include <security/pam_appl.h>
  20 #endif
  21
  22 #if HAVE_SELINUX
  23 #include <selinux/selinux.h>
  24 #endif
  25
  26 #if HAVE_SECCOMP
  27 #include <seccomp.h>
  28 #endif
  29
  30 #if HAVE_APPARMOR
  31 #include <sys/apparmor.h>
  32 #endif
  33
  34 #include "sd-messages.h"
  35
  36 #include "acl-util.h"
  37 #include "af-list.h"
  38 #include "alloc-util.h"
  39 #if HAVE_APPARMOR
  40 #include "apparmor-util.h"
  41 #endif
  42 #include "argv-util.h"
  43 #include "async.h"
  44 #include "barrier.h"
  45 #include "bpf-lsm.h"
  46 #include "cap-list.h"
  47 #include "capability-util.h"
  48 #include "cgroup-setup.h"
  49 #include "chase-symlinks.h"
  50 #include "chown-recursive.h"
  51 #include "constants.h"
  52 #include "cpu-set-util.h"
  53 #include "creds-util.h"
  54 #include "data-fd-util.h"
  55 #include "env-file.h"
  56 #include "env-util.h"
  57 #include "errno-list.h"
  58 #include "escape.h"
  59 #include "execute.h"
  60 #include "exit-status.h"
  61 #include "fd-util.h"
  62 #include "fileio.h"
  63 #include "format-util.h"
  64 #include "glob-util.h"
  65 #include "hexdecoct.h"
  66 #include "io-util.h"
  67 #include "ioprio-util.h"
  68 #include "label.h"
  69 #include "log.h"
  70 #include "macro.h"
  71 #include "manager.h"
  72 #include "manager-dump.h"
  73 #include "memory-util.h"
  74 #include "missing_fs.h"
  75 #include "missing_ioprio.h"
  76 #include "mkdir-label.h"
  77 #include "mount-util.h"
  78 #include "mountpoint-util.h"
  79 #include "namespace.h"
  80 #include "parse-util.h"
  81 #include "path-util.h"
  82 #include "process-util.h"
  83 #include "random-util.h"
  84 #include "recurse-dir.h"
  85 #include "rlimit-util.h"
  86 #include "rm-rf.h"
  87 #if HAVE_SECCOMP
  88 #include "seccomp-util.h"
  89 #endif
  90 #include "securebits-util.h"
  91 #include "selinux-util.h"
  92 #include "signal-util.h"
  93 #include "smack-util.h"
  94 #include "socket-util.h"
  95 #include "sort-util.h"
  96 #include "special.h"
  97 #include "stat-util.h"
  98 #include "string-table.h"
  99 #include "string-util.h"
 100 #include "strv.h"
 101 #include "syslog-util.h"
 102 #include "terminal-util.h"
 103 #include "tmpfile-util.h"
 104 #include "umask-util.h"
 105 #include "unit-serialize.h"
 106 #include "user-util.h"
 107 #include "utmp-wtmp.h"
 108
 109 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
 110 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
 111
 112 #define SNDBUF_SIZE (8*1024*1024)
 113
 114 static int shift_fds(int fds[], size_t n_fds) {
 115         if (n_fds <= 0)
 116                 return 0;
 117
 118         /* Modifies the fds array! (sorts it) */
 119
 120         assert(fds);
 121
 122         for (int start = 0;;) {
 123                 int restart_from = -1;
 124
 125                 for (int i = start; i < (int) n_fds; i++) {
 126                         int nfd;
 127
 128                         /* Already at right index? */
 129                         if (fds[i] == i+3)
 130                                 continue;
 131
 132                         nfd = fcntl(fds[i], F_DUPFD, i + 3);
 133                         if (nfd < 0)
 134                                 return -errno;
 135
 136                         safe_close(fds[i]);
 137                         fds[i] = nfd;
 138
 139                         /* Hmm, the fd we wanted isn't free? Then
 140                          * let's remember that and try again from here */
 141                         if (nfd != i+3 && restart_from < 0)
 142                                 restart_from = i;
 143                 }
 144
 145                 if (restart_from < 0)
 146                         break;
 147
 148                 start = restart_from;
 149         }
 150
 151         return 0;
 152 }
 153
 154 static int flags_fds(
 155                 const int fds[],
 156                 size_t n_socket_fds,
 157                 size_t n_fds,
 158                 bool nonblock) {
 159
 160         int r;
 161
 162         if (n_fds <= 0)
 163                 return 0;
 164
 165         assert(fds);
 166
 167         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
 168          * O_NONBLOCK only applies to socket activation though. */
 169
 170         for (size_t i = 0; i < n_fds; i++) {
 171
 172                 if (i < n_socket_fds) {
 173                         r = fd_nonblock(fds[i], nonblock);
 174                         if (r < 0)
 175                                 return r;
 176                 }
 177
 178                 /* We unconditionally drop FD_CLOEXEC from the fds,
 179                  * since after all we want to pass these fds to our
 180                  * children */
 181
 182                 r = fd_cloexec(fds[i], false);
 183                 if (r < 0)
 184                         return r;
 185         }
 186
 187         return 0;
 188 }
 189
 190 static const char *exec_context_tty_path(const ExecContext *context) {
 191         assert(context);
 192
 193         if (context->stdio_as_fds)
 194                 return NULL;
 195
 196         if (context->tty_path)
 197                 return context->tty_path;
 198
 199         return "/dev/console";
 200 }
 201
 202 static void exec_context_tty_reset(const ExecContext *context, const ExecParameters *p) {
 203         const char *path;
 204
 205         assert(context);
 206
 207         path = exec_context_tty_path(context);
 208
 209         if (context->tty_vhangup) {
 210                 if (p && p->stdin_fd >= 0)
 211                         (void) terminal_vhangup_fd(p->stdin_fd);
 212                 else if (path)
 213                         (void) terminal_vhangup(path);
 214         }
 215
 216         if (context->tty_reset) {
 217                 if (p && p->stdin_fd >= 0)
 218                         (void) reset_terminal_fd(p->stdin_fd, true);
 219                 else if (path)
 220                         (void) reset_terminal(path);
 221         }
 222
 223         if (p && p->stdin_fd >= 0)
 224                 (void) terminal_set_size_fd(p->stdin_fd, path, context->tty_rows, context->tty_cols);
 225
 226         if (context->tty_vt_disallocate && path)
 227                 (void) vt_disallocate(path);
 228 }
 229
 230 static bool is_terminal_input(ExecInput i) {
 231         return IN_SET(i,
 232                       EXEC_INPUT_TTY,
 233                       EXEC_INPUT_TTY_FORCE,
 234                       EXEC_INPUT_TTY_FAIL);
 235 }
 236
 237 static bool is_terminal_output(ExecOutput o) {
 238         return IN_SET(o,
 239                       EXEC_OUTPUT_TTY,
 240                       EXEC_OUTPUT_KMSG_AND_CONSOLE,
 241                       EXEC_OUTPUT_JOURNAL_AND_CONSOLE);
 242 }
 243
 244 static bool is_kmsg_output(ExecOutput o) {
 245         return IN_SET(o,
 246                       EXEC_OUTPUT_KMSG,
 247                       EXEC_OUTPUT_KMSG_AND_CONSOLE);
 248 }
 249
 250 static bool exec_context_needs_term(const ExecContext *c) {
 251         assert(c);
 252
 253         /* Return true if the execution context suggests we should set $TERM to something useful. */
 254
 255         if (is_terminal_input(c->std_input))
 256                 return true;
 257
 258         if (is_terminal_output(c->std_output))
 259                 return true;
 260
 261         if (is_terminal_output(c->std_error))
 262                 return true;
 263
 264         return !!c->tty_path;
 265 }
 266
 267 static int open_null_as(int flags, int nfd) {
 268         int fd;
 269
 270         assert(nfd >= 0);
 271
 272         fd = open("/dev/null", flags|O_NOCTTY);
 273         if (fd < 0)
 274                 return -errno;
 275
 276         return move_fd(fd, nfd, false);
 277 }
 278
 279 static int connect_journal_socket(
 280                 int fd,
 281                 const char *log_namespace,
 282                 uid_t uid,
 283                 gid_t gid) {
 284
 285         uid_t olduid = UID_INVALID;
 286         gid_t oldgid = GID_INVALID;
 287         const char *j;
 288         int r;
 289
 290         j = log_namespace ?
 291                 strjoina("/run/systemd/journal.", log_namespace, "/stdout") :
 292                 "/run/systemd/journal/stdout";
 293
 294         if (gid_is_valid(gid)) {
 295                 oldgid = getgid();
 296
 297                 if (setegid(gid) < 0)
 298                         return -errno;
 299         }
 300
 301         if (uid_is_valid(uid)) {
 302                 olduid = getuid();
 303
 304                 if (seteuid(uid) < 0) {
 305                         r = -errno;
 306                         goto restore_gid;
 307                 }
 308         }
 309
 310         r = connect_unix_path(fd, AT_FDCWD, j);
 311
 312         /* If we fail to restore the uid or gid, things will likely fail later on. This should only happen if
 313            an LSM interferes. */
 314
 315         if (uid_is_valid(uid))
 316                 (void) seteuid(olduid);
 317
 318  restore_gid:
 319         if (gid_is_valid(gid))
 320                 (void) setegid(oldgid);
 321
 322         return r;
 323 }
 324
 325 static int connect_logger_as(
 326                 const Unit *unit,
 327                 const ExecContext *context,
 328                 const ExecParameters *params,
 329                 ExecOutput output,
 330                 const char *ident,
 331                 int nfd,
 332                 uid_t uid,
 333                 gid_t gid) {
 334
 335         _cleanup_close_ int fd = -EBADF;
 336         int r;
 337
 338         assert(context);
 339         assert(params);
 340         assert(output < _EXEC_OUTPUT_MAX);
 341         assert(ident);
 342         assert(nfd >= 0);
 343
 344         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 345         if (fd < 0)
 346                 return -errno;
 347
 348         r = connect_journal_socket(fd, context->log_namespace, uid, gid);
 349         if (r < 0)
 350                 return r;
 351
 352         if (shutdown(fd, SHUT_RD) < 0)
 353                 return -errno;
 354
 355         (void) fd_inc_sndbuf(fd, SNDBUF_SIZE);
 356
 357         if (dprintf(fd,
 358                 "%s\n"
 359                 "%s\n"
 360                 "%i\n"
 361                 "%i\n"
 362                 "%i\n"
 363                 "%i\n"
 364                 "%i\n",
 365                 context->syslog_identifier ?: ident,
 366                 params->flags & EXEC_PASS_LOG_UNIT ? unit->id : "",
 367                 context->syslog_priority,
 368                 !!context->syslog_level_prefix,
 369                 false,
 370                 is_kmsg_output(output),
 371                 is_terminal_output(output)) < 0)
 372                 return -errno;
 373
 374         return move_fd(TAKE_FD(fd), nfd, false);
 375 }
 376
 377 static int open_terminal_as(const char *path, int flags, int nfd) {
 378         int fd;
 379
 380         assert(path);
 381         assert(nfd >= 0);
 382
 383         fd = open_terminal(path, flags | O_NOCTTY);
 384         if (fd < 0)
 385                 return fd;
 386
 387         return move_fd(fd, nfd, false);
 388 }
 389
 390 static int acquire_path(const char *path, int flags, mode_t mode) {
 391         _cleanup_close_ int fd = -EBADF;
 392         int r;
 393
 394         assert(path);
 395
 396         if (IN_SET(flags & O_ACCMODE, O_WRONLY, O_RDWR))
 397                 flags |= O_CREAT;
 398
 399         fd = open(path, flags|O_NOCTTY, mode);
 400         if (fd >= 0)
 401                 return TAKE_FD(fd);
 402
 403         if (errno != ENXIO) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
 404                 return -errno;
 405
 406         /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
 407
 408         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 409         if (fd < 0)
 410                 return -errno;
 411
 412         r = connect_unix_path(fd, AT_FDCWD, path);
 413         if (IN_SET(r, -ENOTSOCK, -EINVAL))
 414                 /* Propagate initial error if we get ENOTSOCK or EINVAL, i.e. we have indication that this
 415                  * wasn't an AF_UNIX socket after all */
 416                 return -ENXIO;
 417         if (r < 0)
 418                 return r;
 419
 420         if ((flags & O_ACCMODE) == O_RDONLY)
 421                 r = shutdown(fd, SHUT_WR);
 422         else if ((flags & O_ACCMODE) == O_WRONLY)
 423                 r = shutdown(fd, SHUT_RD);
 424         else
 425                 r = 0;
 426         if (r < 0)
 427                 return -errno;
 428
 429         return TAKE_FD(fd);
 430 }
 431
 432 static int fixup_input(
 433                 const ExecContext *context,
 434                 int socket_fd,
 435                 bool apply_tty_stdin) {
 436
 437         ExecInput std_input;
 438
 439         assert(context);
 440
 441         std_input = context->std_input;
 442
 443         if (is_terminal_input(std_input) && !apply_tty_stdin)
 444                 return EXEC_INPUT_NULL;
 445
 446         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 447                 return EXEC_INPUT_NULL;
 448
 449         if (std_input == EXEC_INPUT_DATA && context->stdin_data_size == 0)
 450                 return EXEC_INPUT_NULL;
 451
 452         return std_input;
 453 }
 454
 455 static int fixup_output(ExecOutput output, int socket_fd) {
 456
 457         if (output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 458                 return EXEC_OUTPUT_INHERIT;
 459
 460         return output;
 461 }
 462
 463 static int setup_input(
 464                 const ExecContext *context,
 465                 const ExecParameters *params,
 466                 int socket_fd,
 467                 const int named_iofds[static 3]) {
 468
 469         ExecInput i;
 470         int r;
 471
 472         assert(context);
 473         assert(params);
 474         assert(named_iofds);
 475
 476         if (params->stdin_fd >= 0) {
 477                 if (dup2(params->stdin_fd, STDIN_FILENO) < 0)
 478                         return -errno;
 479
 480                 /* Try to make this the controlling tty, if it is a tty, and reset it */
 481                 if (isatty(STDIN_FILENO)) {
 482                         (void) ioctl(STDIN_FILENO, TIOCSCTTY, context->std_input == EXEC_INPUT_TTY_FORCE);
 483                         (void) reset_terminal_fd(STDIN_FILENO, true);
 484                         (void) terminal_set_size_fd(STDIN_FILENO, NULL, context->tty_rows, context->tty_cols);
 485                 }
 486
 487                 return STDIN_FILENO;
 488         }
 489
 490         i = fixup_input(context, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN);
 491
 492         switch (i) {
 493
 494         case EXEC_INPUT_NULL:
 495                 return open_null_as(O_RDONLY, STDIN_FILENO);
 496
 497         case EXEC_INPUT_TTY:
 498         case EXEC_INPUT_TTY_FORCE:
 499         case EXEC_INPUT_TTY_FAIL: {
 500                 int fd;
 501
 502                 fd = acquire_terminal(exec_context_tty_path(context),
 503                                       i == EXEC_INPUT_TTY_FAIL  ? ACQUIRE_TERMINAL_TRY :
 504                                       i == EXEC_INPUT_TTY_FORCE ? ACQUIRE_TERMINAL_FORCE :
 505                                                                   ACQUIRE_TERMINAL_WAIT,
 506                                       USEC_INFINITY);
 507                 if (fd < 0)
 508                         return fd;
 509
 510                 r = terminal_set_size_fd(fd, exec_context_tty_path(context), context->tty_rows, context->tty_cols);
 511                 if (r < 0)
 512                         return r;
 513
 514                 return move_fd(fd, STDIN_FILENO, false);
 515         }
 516
 517         case EXEC_INPUT_SOCKET:
 518                 assert(socket_fd >= 0);
 519
 520                 return RET_NERRNO(dup2(socket_fd, STDIN_FILENO));
 521
 522         case EXEC_INPUT_NAMED_FD:
 523                 assert(named_iofds[STDIN_FILENO] >= 0);
 524
 525                 (void) fd_nonblock(named_iofds[STDIN_FILENO], false);
 526                 return RET_NERRNO(dup2(named_iofds[STDIN_FILENO], STDIN_FILENO));
 527
 528         case EXEC_INPUT_DATA: {
 529                 int fd;
 530
 531                 fd = acquire_data_fd(context->stdin_data, context->stdin_data_size, 0);
 532                 if (fd < 0)
 533                         return fd;
 534
 535                 return move_fd(fd, STDIN_FILENO, false);
 536         }
 537
 538         case EXEC_INPUT_FILE: {
 539                 bool rw;
 540                 int fd;
 541
 542                 assert(context->stdio_file[STDIN_FILENO]);
 543
 544                 rw = (context->std_output == EXEC_OUTPUT_FILE && streq_ptr(context->stdio_file[STDIN_FILENO], context->stdio_file[STDOUT_FILENO])) ||
 545                         (context->std_error == EXEC_OUTPUT_FILE && streq_ptr(context->stdio_file[STDIN_FILENO], context->stdio_file[STDERR_FILENO]));
 546
 547                 fd = acquire_path(context->stdio_file[STDIN_FILENO], rw ? O_RDWR : O_RDONLY, 0666 & ~context->umask);
 548                 if (fd < 0)
 549                         return fd;
 550
 551                 return move_fd(fd, STDIN_FILENO, false);
 552         }
 553
 554         default:
 555                 assert_not_reached();
 556         }
 557 }
 558
 559 static bool can_inherit_stderr_from_stdout(
 560                 const ExecContext *context,
 561                 ExecOutput o,
 562                 ExecOutput e) {
 563
 564         assert(context);
 565
 566         /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
 567          * stderr fd */
 568
 569         if (e == EXEC_OUTPUT_INHERIT)
 570                 return true;
 571         if (e != o)
 572                 return false;
 573
 574         if (e == EXEC_OUTPUT_NAMED_FD)
 575                 return streq_ptr(context->stdio_fdname[STDOUT_FILENO], context->stdio_fdname[STDERR_FILENO]);
 576
 577         if (IN_SET(e, EXEC_OUTPUT_FILE, EXEC_OUTPUT_FILE_APPEND, EXEC_OUTPUT_FILE_TRUNCATE))
 578                 return streq_ptr(context->stdio_file[STDOUT_FILENO], context->stdio_file[STDERR_FILENO]);
 579
 580         return true;
 581 }
 582
 583 static int setup_output(
 584                 const Unit *unit,
 585                 const ExecContext *context,
 586                 const ExecParameters *params,
 587                 int fileno,
 588                 int socket_fd,
 589                 const int named_iofds[static 3],
 590                 const char *ident,
 591                 uid_t uid,
 592                 gid_t gid,
 593                 dev_t *journal_stream_dev,
 594                 ino_t *journal_stream_ino) {
 595
 596         ExecOutput o;
 597         ExecInput i;
 598         int r;
 599
 600         assert(unit);
 601         assert(context);
 602         assert(params);
 603         assert(ident);
 604         assert(journal_stream_dev);
 605         assert(journal_stream_ino);
 606
 607         if (fileno == STDOUT_FILENO && params->stdout_fd >= 0) {
 608
 609                 if (dup2(params->stdout_fd, STDOUT_FILENO) < 0)
 610                         return -errno;
 611
 612                 return STDOUT_FILENO;
 613         }
 614
 615         if (fileno == STDERR_FILENO && params->stderr_fd >= 0) {
 616                 if (dup2(params->stderr_fd, STDERR_FILENO) < 0)
 617                         return -errno;
 618
 619                 return STDERR_FILENO;
 620         }
 621
 622         i = fixup_input(context, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN);
 623         o = fixup_output(context->std_output, socket_fd);
 624
 625         if (fileno == STDERR_FILENO) {
 626                 ExecOutput e;
 627                 e = fixup_output(context->std_error, socket_fd);
 628
 629                 /* This expects the input and output are already set up */
 630
 631                 /* Don't change the stderr file descriptor if we inherit all
 632                  * the way and are not on a tty */
 633                 if (e == EXEC_OUTPUT_INHERIT &&
 634                     o == EXEC_OUTPUT_INHERIT &&
 635                     i == EXEC_INPUT_NULL &&
 636                     !is_terminal_input(context->std_input) &&
 637                     getppid() != 1)
 638                         return fileno;
 639
 640                 /* Duplicate from stdout if possible */
 641                 if (can_inherit_stderr_from_stdout(context, o, e))
 642                         return RET_NERRNO(dup2(STDOUT_FILENO, fileno));
 643
 644                 o = e;
 645
 646         } else if (o == EXEC_OUTPUT_INHERIT) {
 647                 /* If input got downgraded, inherit the original value */
 648                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 649                         return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno);
 650
 651                 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
 652                 if (!IN_SET(i, EXEC_INPUT_NULL, EXEC_INPUT_DATA))
 653                         return RET_NERRNO(dup2(STDIN_FILENO, fileno));
 654
 655                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 656                 if (getppid() != 1)
 657                         return fileno;
 658
 659                 /* We need to open /dev/null here anew, to get the right access mode. */
 660                 return open_null_as(O_WRONLY, fileno);
 661         }
 662
 663         switch (o) {
 664
 665         case EXEC_OUTPUT_NULL:
 666                 return open_null_as(O_WRONLY, fileno);
 667
 668         case EXEC_OUTPUT_TTY:
 669                 if (is_terminal_input(i))
 670                         return RET_NERRNO(dup2(STDIN_FILENO, fileno));
 671
 672                 /* We don't reset the terminal if this is just about output */
 673                 return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno);
 674
 675         case EXEC_OUTPUT_KMSG:
 676         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 677         case EXEC_OUTPUT_JOURNAL:
 678         case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
 679                 r = connect_logger_as(unit, context, params, o, ident, fileno, uid, gid);
 680                 if (r < 0) {
 681                         log_unit_warning_errno(unit, r, "Failed to connect %s to the journal socket, ignoring: %m",
 682                                                fileno == STDOUT_FILENO ? "stdout" : "stderr");
 683                         r = open_null_as(O_WRONLY, fileno);
 684                 } else {
 685                         struct stat st;
 686
 687                         /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
 688                          * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
 689                          * services to detect whether they are connected to the journal or not.
 690                          *
 691                          * If both stdout and stderr are connected to a stream then let's make sure to store the data
 692                          * about STDERR as that's usually the best way to do logging. */
 693
 694                         if (fstat(fileno, &st) >= 0 &&
 695                             (*journal_stream_ino == 0 || fileno == STDERR_FILENO)) {
 696                                 *journal_stream_dev = st.st_dev;
 697                                 *journal_stream_ino = st.st_ino;
 698                         }
 699                 }
 700                 return r;
 701
 702         case EXEC_OUTPUT_SOCKET:
 703                 assert(socket_fd >= 0);
 704
 705                 return RET_NERRNO(dup2(socket_fd, fileno));
 706
 707         case EXEC_OUTPUT_NAMED_FD:
 708                 assert(named_iofds[fileno] >= 0);
 709
 710                 (void) fd_nonblock(named_iofds[fileno], false);
 711                 return RET_NERRNO(dup2(named_iofds[fileno], fileno));
 712
 713         case EXEC_OUTPUT_FILE:
 714         case EXEC_OUTPUT_FILE_APPEND:
 715         case EXEC_OUTPUT_FILE_TRUNCATE: {
 716                 bool rw;
 717                 int fd, flags;
 718
 719                 assert(context->stdio_file[fileno]);
 720
 721                 rw = context->std_input == EXEC_INPUT_FILE &&
 722                         streq_ptr(context->stdio_file[fileno], context->stdio_file[STDIN_FILENO]);
 723
 724                 if (rw)
 725                         return RET_NERRNO(dup2(STDIN_FILENO, fileno));
 726
 727                 flags = O_WRONLY;
 728                 if (o == EXEC_OUTPUT_FILE_APPEND)
 729                         flags |= O_APPEND;
 730                 else if (o == EXEC_OUTPUT_FILE_TRUNCATE)
 731                         flags |= O_TRUNC;
 732
 733                 fd = acquire_path(context->stdio_file[fileno], flags, 0666 & ~context->umask);
 734                 if (fd < 0)
 735                         return fd;
 736
 737                 return move_fd(fd, fileno, 0);
 738         }
 739
 740         default:
 741                 assert_not_reached();
 742         }
 743 }
 744
 745 static int chown_terminal(int fd, uid_t uid) {
 746         int r;
 747
 748         assert(fd >= 0);
 749
 750         /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
 751         if (isatty(fd) < 1) {
 752                 if (IN_SET(errno, EINVAL, ENOTTY))
 753                         return 0; /* not a tty */
 754
 755                 return -errno;
 756         }
 757
 758         /* This might fail. What matters are the results. */
 759         r = fchmod_and_chown(fd, TTY_MODE, uid, GID_INVALID);
 760         if (r < 0)
 761                 return r;
 762
 763         return 1;
 764 }
 765
 766 static int setup_confirm_stdio(
 767                 const ExecContext *context,
 768                 const char *vc,
 769                 int *ret_saved_stdin,
 770                 int *ret_saved_stdout) {
 771
 772         _cleanup_close_ int fd = -EBADF, saved_stdin = -EBADF, saved_stdout = -EBADF;
 773         int r;
 774
 775         assert(ret_saved_stdin);
 776         assert(ret_saved_stdout);
 777
 778         saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
 779         if (saved_stdin < 0)
 780                 return -errno;
 781
 782         saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
 783         if (saved_stdout < 0)
 784                 return -errno;
 785
 786         fd = acquire_terminal(vc, ACQUIRE_TERMINAL_WAIT, DEFAULT_CONFIRM_USEC);
 787         if (fd < 0)
 788                 return fd;
 789
 790         r = chown_terminal(fd, getuid());
 791         if (r < 0)
 792                 return r;
 793
 794         r = reset_terminal_fd(fd, true);
 795         if (r < 0)
 796                 return r;
 797
 798         r = terminal_set_size_fd(fd, vc, context->tty_rows, context->tty_cols);
 799         if (r < 0)
 800                 return r;
 801
 802         r = rearrange_stdio(fd, fd, STDERR_FILENO); /* Invalidates 'fd' also on failure */
 803         TAKE_FD(fd);
 804         if (r < 0)
 805                 return r;
 806
 807         *ret_saved_stdin = TAKE_FD(saved_stdin);
 808         *ret_saved_stdout = TAKE_FD(saved_stdout);
 809         return 0;
 810 }
 811
 812 static void write_confirm_error_fd(int err, int fd, const Unit *u) {
 813         assert(err < 0);
 814
 815         if (err == -ETIMEDOUT)
 816                 dprintf(fd, "Confirmation question timed out for %s, assuming positive response.\n", u->id);
 817         else {
 818                 errno = -err;
 819                 dprintf(fd, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u->id);
 820         }
 821 }
 822
 823 static void write_confirm_error(int err, const char *vc, const Unit *u) {
 824         _cleanup_close_ int fd = -EBADF;
 825
 826         assert(vc);
 827
 828         fd = open_terminal(vc, O_WRONLY|O_NOCTTY|O_CLOEXEC);
 829         if (fd < 0)
 830                 return;
 831
 832         write_confirm_error_fd(err, fd, u);
 833 }
 834
 835 static int restore_confirm_stdio(int *saved_stdin, int *saved_stdout) {
 836         int r = 0;
 837
 838         assert(saved_stdin);
 839         assert(saved_stdout);
 840
 841         release_terminal();
 842
 843         if (*saved_stdin >= 0)
 844                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 845                         r = -errno;
 846
 847         if (*saved_stdout >= 0)
 848                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 849                         r = -errno;
 850
 851         *saved_stdin = safe_close(*saved_stdin);
 852         *saved_stdout = safe_close(*saved_stdout);
 853
 854         return r;
 855 }
 856
 857 enum {
 858         CONFIRM_PRETEND_FAILURE = -1,
 859         CONFIRM_PRETEND_SUCCESS =  0,
 860         CONFIRM_EXECUTE = 1,
 861 };
 862
 863 static int ask_for_confirmation(const ExecContext *context, const char *vc, Unit *u, const char *cmdline) {
 864         int saved_stdout = -1, saved_stdin = -1, r;
 865         _cleanup_free_ char *e = NULL;
 866         char c;
 867
 868         /* For any internal errors, assume a positive response. */
 869         r = setup_confirm_stdio(context, vc, &saved_stdin, &saved_stdout);
 870         if (r < 0) {
 871                 write_confirm_error(r, vc, u);
 872                 return CONFIRM_EXECUTE;
 873         }
 874
 875         /* confirm_spawn might have been disabled while we were sleeping. */
 876         if (manager_is_confirm_spawn_disabled(u->manager)) {
 877                 r = 1;
 878                 goto restore_stdio;
 879         }
 880
 881         e = ellipsize(cmdline, 60, 100);
 882         if (!e) {
 883                 log_oom();
 884                 r = CONFIRM_EXECUTE;
 885                 goto restore_stdio;
 886         }
 887
 888         for (;;) {
 889                 r = ask_char(&c, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e);
 890                 if (r < 0) {
 891                         write_confirm_error_fd(r, STDOUT_FILENO, u);
 892                         r = CONFIRM_EXECUTE;
 893                         goto restore_stdio;
 894                 }
 895
 896                 switch (c) {
 897                 case 'c':
 898                         printf("Resuming normal execution.\n");
 899                         manager_disable_confirm_spawn();
 900                         r = 1;
 901                         break;
 902                 case 'D':
 903                         unit_dump(u, stdout, "  ");
 904                         continue; /* ask again */
 905                 case 'f':
 906                         printf("Failing execution.\n");
 907                         r = CONFIRM_PRETEND_FAILURE;
 908                         break;
 909                 case 'h':
 910                         printf("  c - continue, proceed without asking anymore\n"
 911                                "  D - dump, show the state of the unit\n"
 912                                "  f - fail, don't execute the command and pretend it failed\n"
 913                                "  h - help\n"
 914                                "  i - info, show a short summary of the unit\n"
 915                                "  j - jobs, show jobs that are in progress\n"
 916                                "  s - skip, don't execute the command and pretend it succeeded\n"
 917                                "  y - yes, execute the command\n");
 918                         continue; /* ask again */
 919                 case 'i':
 920                         printf("  Description: %s\n"
 921                                "  Unit:        %s\n"
 922                                "  Command:     %s\n",
 923                                u->id, u->description, cmdline);
 924                         continue; /* ask again */
 925                 case 'j':
 926                         manager_dump_jobs(u->manager, stdout, /* patterns= */ NULL, "  ");
 927                         continue; /* ask again */
 928                 case 'n':
 929                         /* 'n' was removed in favor of 'f'. */
 930                         printf("Didn't understand 'n', did you mean 'f'?\n");
 931                         continue; /* ask again */
 932                 case 's':
 933                         printf("Skipping execution.\n");
 934                         r = CONFIRM_PRETEND_SUCCESS;
 935                         break;
 936                 case 'y':
 937                         r = CONFIRM_EXECUTE;
 938                         break;
 939                 default:
 940                         assert_not_reached();
 941                 }
 942                 break;
 943         }
 944
 945 restore_stdio:
 946         restore_confirm_stdio(&saved_stdin, &saved_stdout);
 947         return r;
 948 }
 949
 950 static int get_fixed_user(const ExecContext *c, const char **user,
 951                           uid_t *uid, gid_t *gid,
 952                           const char **home, const char **shell) {
 953         int r;
 954         const char *name;
 955
 956         assert(c);
 957
 958         if (!c->user)
 959                 return 0;
 960
 961         /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
 962          * (i.e. are "/" or "/bin/nologin"). */
 963
 964         name = c->user;
 965         r = get_user_creds(&name, uid, gid, home, shell, USER_CREDS_CLEAN);
 966         if (r < 0)
 967                 return r;
 968
 969         *user = name;
 970         return 0;
 971 }
 972
 973 static int get_fixed_group(const ExecContext *c, const char **group, gid_t *gid) {
 974         int r;
 975         const char *name;
 976
 977         assert(c);
 978
 979         if (!c->group)
 980                 return 0;
 981
 982         name = c->group;
 983         r = get_group_creds(&name, gid, 0);
 984         if (r < 0)
 985                 return r;
 986
 987         *group = name;
 988         return 0;
 989 }
 990
 991 static int get_supplementary_groups(const ExecContext *c, const char *user,
 992                                     const char *group, gid_t gid,
 993                                     gid_t **supplementary_gids, int *ngids) {
 994         int r, k = 0;
 995         int ngroups_max;
 996         bool keep_groups = false;
 997         gid_t *groups = NULL;
 998         _cleanup_free_ gid_t *l_gids = NULL;
 999
1000         assert(c);
1001
1002         /*
1003          * If user is given, then lookup GID and supplementary groups list.
1004          * We avoid NSS lookups for gid=0. Also we have to initialize groups
1005          * here and as early as possible so we keep the list of supplementary
1006          * groups of the caller.
1007          */
1008         if (user && gid_is_valid(gid) && gid != 0) {
1009                 /* First step, initialize groups from /etc/groups */
1010                 if (initgroups(user, gid) < 0)
1011                         return -errno;
1012
1013                 keep_groups = true;
1014         }
1015
1016         if (strv_isempty(c->supplementary_groups))
1017                 return 0;
1018
1019         /*
1020          * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1021          * be positive, otherwise fail.
1022          */
1023         errno = 0;
1024         ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
1025         if (ngroups_max <= 0)
1026                 return errno_or_else(EOPNOTSUPP);
1027
1028         l_gids = new(gid_t, ngroups_max);
1029         if (!l_gids)
1030                 return -ENOMEM;
1031
1032         if (keep_groups) {
1033                 /*
1034                  * Lookup the list of groups that the user belongs to, we
1035                  * avoid NSS lookups here too for gid=0.
1036                  */
1037                 k = ngroups_max;
1038                 if (getgrouplist(user, gid, l_gids, &k) < 0)
1039                         return -EINVAL;
1040         } else
1041                 k = 0;
1042
1043         STRV_FOREACH(i, c->supplementary_groups) {
1044                 const char *g;
1045
1046                 if (k >= ngroups_max)
1047                         return -E2BIG;
1048
1049                 g = *i;
1050                 r = get_group_creds(&g, l_gids+k, 0);
1051                 if (r < 0)
1052                         return r;
1053
1054                 k++;
1055         }
1056
1057         /*
1058          * Sets ngids to zero to drop all supplementary groups, happens
1059          * when we are under root and SupplementaryGroups= is empty.
1060          */
1061         if (k == 0) {
1062                 *ngids = 0;
1063                 return 0;
1064         }
1065
1066         /* Otherwise get the final list of supplementary groups */
1067         groups = memdup(l_gids, sizeof(gid_t) * k);
1068         if (!groups)
1069                 return -ENOMEM;
1070
1071         *supplementary_gids = groups;
1072         *ngids = k;
1073
1074         groups = NULL;
1075
1076         return 0;
1077 }
1078
1079 static int enforce_groups(gid_t gid, const gid_t *supplementary_gids, int ngids) {
1080         int r;
1081
1082         /* Handle SupplementaryGroups= if it is not empty */
1083         if (ngids > 0) {
1084                 r = maybe_setgroups(ngids, supplementary_gids);
1085                 if (r < 0)
1086                         return r;
1087         }
1088
1089         if (gid_is_valid(gid)) {
1090                 /* Then set our gids */
1091                 if (setresgid(gid, gid, gid) < 0)
1092                         return -errno;
1093         }
1094
1095         return 0;
1096 }
1097
1098 static int set_securebits(unsigned bits, unsigned mask) {
1099         unsigned applied;
1100         int current;
1101
1102         current = prctl(PR_GET_SECUREBITS);
1103         if (current < 0)
1104                 return -errno;
1105
1106         /* Clear all securebits defined in mask and set bits */
1107         applied = ((unsigned) current & ~mask) | bits;
1108         if ((unsigned) current == applied)
1109                 return 0;
1110
1111         if (prctl(PR_SET_SECUREBITS, applied) < 0)
1112                 return -errno;
1113
1114         return 1;
1115 }
1116
1117 static int enforce_user(const ExecContext *context, uid_t uid) {
1118         assert(context);
1119         int r;
1120
1121         if (!uid_is_valid(uid))
1122                 return 0;
1123
1124         /* Sets (but doesn't look up) the UIS and makes sure we keep the capabilities while doing so. For
1125          * setting secure bits the capability CAP_SETPCAP is required, so we also need keep-caps in this
1126          * case. */
1127
1128         if ((context->capability_ambient_set != 0 || context->secure_bits != 0) && uid != 0) {
1129
1130                 /* First step: If we need to keep capabilities but drop privileges we need to make sure we
1131                  * keep our caps, while we drop privileges. Add KEEP_CAPS to the securebits */
1132                 r = set_securebits(1U << SECURE_KEEP_CAPS, 0);
1133                 if (r < 0)
1134                         return r;
1135         }
1136
1137         /* Second step: actually set the uids */
1138         if (setresuid(uid, uid, uid) < 0)
1139                 return -errno;
1140
1141         /* At this point we should have all necessary capabilities but are otherwise a normal user. However,
1142          * the caps might got corrupted due to the setresuid() so we need clean them up later. This is done
1143          * outside of this call. */
1144         return 0;
1145 }
1146
1147 #if HAVE_PAM
1148
1149 static int null_conv(
1150                 int num_msg,
1151                 const struct pam_message **msg,
1152                 struct pam_response **resp,
1153                 void *appdata_ptr) {
1154
1155         /* We don't support conversations */
1156
1157         return PAM_CONV_ERR;
1158 }
1159
1160 #endif
1161
1162 static int setup_pam(
1163                 const char *name,
1164                 const char *user,
1165                 uid_t uid,
1166                 gid_t gid,
1167                 const char *tty,
1168                 char ***env, /* updated on success */
1169                 const int fds[], size_t n_fds) {
1170
1171 #if HAVE_PAM
1172
1173         static const struct pam_conv conv = {
1174                 .conv = null_conv,
1175                 .appdata_ptr = NULL
1176         };
1177
1178         _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
1179         _cleanup_strv_free_ char **e = NULL;
1180         pam_handle_t *handle = NULL;
1181         sigset_t old_ss;
1182         int pam_code = PAM_SUCCESS, r;
1183         bool close_session = false;
1184         pid_t pam_pid = 0, parent_pid;
1185         int flags = 0;
1186
1187         assert(name);
1188         assert(user);
1189         assert(env);
1190
1191         /* We set up PAM in the parent process, then fork. The child
1192          * will then stay around until killed via PR_GET_PDEATHSIG or
1193          * systemd via the cgroup logic. It will then remove the PAM
1194          * session again. The parent process will exec() the actual
1195          * daemon. We do things this way to ensure that the main PID
1196          * of the daemon is the one we initially fork()ed. */
1197
1198         r = barrier_create(&barrier);
1199         if (r < 0)
1200                 goto fail;
1201
1202         if (log_get_max_level() < LOG_DEBUG)
1203                 flags |= PAM_SILENT;
1204
1205         pam_code = pam_start(name, user, &conv, &handle);
1206         if (pam_code != PAM_SUCCESS) {
1207                 handle = NULL;
1208                 goto fail;
1209         }
1210
1211         if (!tty) {
1212                 _cleanup_free_ char *q = NULL;
1213
1214                 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1215                  * out if that's the case, and read the TTY off it. */
1216
1217                 if (getttyname_malloc(STDIN_FILENO, &q) >= 0)
1218                         tty = strjoina("/dev/", q);
1219         }
1220
1221         if (tty) {
1222                 pam_code = pam_set_item(handle, PAM_TTY, tty);
1223                 if (pam_code != PAM_SUCCESS)
1224                         goto fail;
1225         }
1226
1227         STRV_FOREACH(nv, *env) {
1228                 pam_code = pam_putenv(handle, *nv);
1229                 if (pam_code != PAM_SUCCESS)
1230                         goto fail;
1231         }
1232
1233         pam_code = pam_acct_mgmt(handle, flags);
1234         if (pam_code != PAM_SUCCESS)
1235                 goto fail;
1236
1237         pam_code = pam_setcred(handle, PAM_ESTABLISH_CRED | flags);
1238         if (pam_code != PAM_SUCCESS)
1239                 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle, pam_code));
1240
1241         pam_code = pam_open_session(handle, flags);
1242         if (pam_code != PAM_SUCCESS)
1243                 goto fail;
1244
1245         close_session = true;
1246
1247         e = pam_getenvlist(handle);
1248         if (!e) {
1249                 pam_code = PAM_BUF_ERR;
1250                 goto fail;
1251         }
1252
1253         /* Block SIGTERM, so that we know that it won't get lost in the child */
1254
1255         assert_se(sigprocmask_many(SIG_BLOCK, &old_ss, SIGTERM, -1) >= 0);
1256
1257         parent_pid = getpid_cached();
1258
1259         r = safe_fork("(sd-pam)", 0, &pam_pid);
1260         if (r < 0)
1261                 goto fail;
1262         if (r == 0) {
1263                 int sig, ret = EXIT_PAM;
1264
1265                 /* The child's job is to reset the PAM session on termination */
1266                 barrier_set_role(&barrier, BARRIER_CHILD);
1267
1268                 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1269                  * those fds are open here that have been opened by PAM. */
1270                 (void) close_many(fds, n_fds);
1271
1272                 /* Drop privileges - we don't need any to pam_close_session and this will make
1273                  * PR_SET_PDEATHSIG work in most cases.  If this fails, ignore the error - but expect sd-pam
1274                  * threads to fail to exit normally */
1275
1276                 r = maybe_setgroups(0, NULL);
1277                 if (r < 0)
1278                         log_warning_errno(r, "Failed to setgroups() in sd-pam: %m");
1279                 if (setresgid(gid, gid, gid) < 0)
1280                         log_warning_errno(errno, "Failed to setresgid() in sd-pam: %m");
1281                 if (setresuid(uid, uid, uid) < 0)
1282                         log_warning_errno(errno, "Failed to setresuid() in sd-pam: %m");
1283
1284                 (void) ignore_signals(SIGPIPE);
1285
1286                 /* Wait until our parent died. This will only work if the above setresuid() succeeds,
1287                  * otherwise the kernel will not allow unprivileged parents kill their privileged children
1288                  * this way. We rely on the control groups kill logic to do the rest for us. */
1289                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
1290                         goto child_finish;
1291
1292                 /* Tell the parent that our setup is done. This is especially important regarding dropping
1293                  * privileges. Otherwise, unit setup might race against our setresuid(2) call.
1294                  *
1295                  * If the parent aborted, we'll detect this below, hence ignore return failure here. */
1296                 (void) barrier_place(&barrier);
1297
1298                 /* Check if our parent process might already have died? */
1299                 if (getppid() == parent_pid) {
1300                         sigset_t ss;
1301
1302                         assert_se(sigemptyset(&ss) >= 0);
1303                         assert_se(sigaddset(&ss, SIGTERM) >= 0);
1304
1305                         for (;;) {
1306                                 if (sigwait(&ss, &sig) < 0) {
1307                                         if (errno == EINTR)
1308                                                 continue;
1309
1310                                         goto child_finish;
1311                                 }
1312
1313                                 assert(sig == SIGTERM);
1314                                 break;
1315                         }
1316                 }
1317
1318                 pam_code = pam_setcred(handle, PAM_DELETE_CRED | flags);
1319                 if (pam_code != PAM_SUCCESS)
1320                         goto child_finish;
1321
1322                 /* If our parent died we'll end the session */
1323                 if (getppid() != parent_pid) {
1324                         pam_code = pam_close_session(handle, flags);
1325                         if (pam_code != PAM_SUCCESS)
1326                                 goto child_finish;
1327                 }
1328
1329                 ret = 0;
1330
1331         child_finish:
1332                 /* NB: pam_end() when called in child processes should set PAM_DATA_SILENT to let the module
1333                  * know about this. See pam_end(3) */
1334                 (void) pam_end(handle, pam_code | flags | PAM_DATA_SILENT);
1335                 _exit(ret);
1336         }
1337
1338         barrier_set_role(&barrier, BARRIER_PARENT);
1339
1340         /* If the child was forked off successfully it will do all the cleanups, so forget about the handle
1341          * here. */
1342         handle = NULL;
1343
1344         /* Unblock SIGTERM again in the parent */
1345         assert_se(sigprocmask(SIG_SETMASK, &old_ss, NULL) >= 0);
1346
1347         /* We close the log explicitly here, since the PAM modules might have opened it, but we don't want
1348          * this fd around. */
1349         closelog();
1350
1351         /* Synchronously wait for the child to initialize. We don't care for errors as we cannot
1352          * recover. However, warn loudly if it happens. */
1353         if (!barrier_place_and_sync(&barrier))
1354                 log_error("PAM initialization failed");
1355
1356         return strv_free_and_replace(*env, e);
1357
1358 fail:
1359         if (pam_code != PAM_SUCCESS) {
1360                 log_error("PAM failed: %s", pam_strerror(handle, pam_code));
1361                 r = -EPERM;  /* PAM errors do not map to errno */
1362         } else
1363                 log_error_errno(r, "PAM failed: %m");
1364
1365         if (handle) {
1366                 if (close_session)
1367                         pam_code = pam_close_session(handle, flags);
1368
1369                 (void) pam_end(handle, pam_code | flags);
1370         }
1371
1372         closelog();
1373         return r;
1374 #else
1375         return 0;
1376 #endif
1377 }
1378
1379 static void rename_process_from_path(const char *path) {
1380         _cleanup_free_ char *buf = NULL;
1381         const char *p;
1382
1383         assert(path);
1384
1385         /* This resulting string must fit in 10 chars (i.e. the length of "/sbin/init") to look pretty in
1386          * /bin/ps */
1387
1388         if (path_extract_filename(path, &buf) < 0) {
1389                 rename_process("(...)");
1390                 return;
1391         }
1392
1393         size_t l = strlen(buf);
1394         if (l > 8) {
1395                 /* The end of the process name is usually more interesting, since the first bit might just be
1396                  * "systemd-" */
1397                 p = buf + l - 8;
1398                 l = 8;
1399         } else
1400                 p = buf;
1401
1402         char process_name[11];
1403         process_name[0] = '(';
1404         memcpy(process_name+1, p, l);
1405         process_name[1+l] = ')';
1406         process_name[1+l+1] = 0;
1407
1408         rename_process(process_name);
1409 }
1410
1411 static bool context_has_address_families(const ExecContext *c) {
1412         assert(c);
1413
1414         return c->address_families_allow_list ||
1415                 !set_isempty(c->address_families);
1416 }
1417
1418 static bool context_has_syscall_filters(const ExecContext *c) {
1419         assert(c);
1420
1421         return c->syscall_allow_list ||
1422                 !hashmap_isempty(c->syscall_filter);
1423 }
1424
1425 static bool context_has_syscall_logs(const ExecContext *c) {
1426         assert(c);
1427
1428         return c->syscall_log_allow_list ||
1429                 !hashmap_isempty(c->syscall_log);
1430 }
1431
1432 static bool context_has_no_new_privileges(const ExecContext *c) {
1433         assert(c);
1434
1435         if (c->no_new_privileges)
1436                 return true;
1437
1438         if (have_effective_cap(CAP_SYS_ADMIN) > 0) /* if we are privileged, we don't need NNP */
1439                 return false;
1440
1441         /* We need NNP if we have any form of seccomp and are unprivileged */
1442         return c->lock_personality ||
1443                 c->memory_deny_write_execute ||
1444                 c->private_devices ||
1445                 c->protect_clock ||
1446                 c->protect_hostname ||
1447                 c->protect_kernel_tunables ||
1448                 c->protect_kernel_modules ||
1449                 c->protect_kernel_logs ||
1450                 context_has_address_families(c) ||
1451                 exec_context_restrict_namespaces_set(c) ||
1452                 c->restrict_realtime ||
1453                 c->restrict_suid_sgid ||
1454                 !set_isempty(c->syscall_archs) ||
1455                 context_has_syscall_filters(c) ||
1456                 context_has_syscall_logs(c);
1457 }
1458
1459 static bool exec_context_has_credentials(const ExecContext *context) {
1460
1461         assert(context);
1462
1463         return !hashmap_isempty(context->set_credentials) ||
1464                 !hashmap_isempty(context->load_credentials);
1465 }
1466
1467 #if HAVE_SECCOMP
1468
1469 static bool skip_seccomp_unavailable(const Unit* u, const char* msg) {
1470
1471         if (is_seccomp_available())
1472                 return false;
1473
1474         log_unit_debug(u, "SECCOMP features not detected in the kernel, skipping %s", msg);
1475         return true;
1476 }
1477
1478 static int apply_syscall_filter(const Unit* u, const ExecContext *c, bool needs_ambient_hack) {
1479         uint32_t negative_action, default_action, action;
1480         int r;
1481
1482         assert(u);
1483         assert(c);
1484
1485         if (!context_has_syscall_filters(c))
1486                 return 0;
1487
1488         if (skip_seccomp_unavailable(u, "SystemCallFilter="))
1489                 return 0;
1490
1491         negative_action = c->syscall_errno == SECCOMP_ERROR_NUMBER_KILL ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c->syscall_errno);
1492
1493         if (c->syscall_allow_list) {
1494                 default_action = negative_action;
1495                 action = SCMP_ACT_ALLOW;
1496         } else {
1497                 default_action = SCMP_ACT_ALLOW;
1498                 action = negative_action;
1499         }
1500
1501         if (needs_ambient_hack) {
1502                 r = seccomp_filter_set_add(c->syscall_filter, c->syscall_allow_list, syscall_filter_sets + SYSCALL_FILTER_SET_SETUID);
1503                 if (r < 0)
1504                         return r;
1505         }
1506
1507         return seccomp_load_syscall_filter_set_raw(default_action, c->syscall_filter, action, false);
1508 }
1509
1510 static int apply_syscall_log(const Unit* u, const ExecContext *c) {
1511 #ifdef SCMP_ACT_LOG
1512         uint32_t default_action, action;
1513 #endif
1514
1515         assert(u);
1516         assert(c);
1517
1518         if (!context_has_syscall_logs(c))
1519                 return 0;
1520
1521 #ifdef SCMP_ACT_LOG
1522         if (skip_seccomp_unavailable(u, "SystemCallLog="))
1523                 return 0;
1524
1525         if (c->syscall_log_allow_list) {
1526                 /* Log nothing but the ones listed */
1527                 default_action = SCMP_ACT_ALLOW;
1528                 action = SCMP_ACT_LOG;
1529         } else {
1530                 /* Log everything but the ones listed */
1531                 default_action = SCMP_ACT_LOG;
1532                 action = SCMP_ACT_ALLOW;
1533         }
1534
1535         return seccomp_load_syscall_filter_set_raw(default_action, c->syscall_log, action, false);
1536 #else
1537         /* old libseccomp */
1538         log_unit_debug(u, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1539         return 0;
1540 #endif
1541 }
1542
1543 static int apply_syscall_archs(const Unit *u, const ExecContext *c) {
1544         assert(u);
1545         assert(c);
1546
1547         if (set_isempty(c->syscall_archs))
1548                 return 0;
1549
1550         if (skip_seccomp_unavailable(u, "SystemCallArchitectures="))
1551                 return 0;
1552
1553         return seccomp_restrict_archs(c->syscall_archs);
1554 }
1555
1556 static int apply_address_families(const Unit* u, const ExecContext *c) {
1557         assert(u);
1558         assert(c);
1559
1560         if (!context_has_address_families(c))
1561                 return 0;
1562
1563         if (skip_seccomp_unavailable(u, "RestrictAddressFamilies="))
1564                 return 0;
1565
1566         return seccomp_restrict_address_families(c->address_families, c->address_families_allow_list);
1567 }
1568
1569 static int apply_memory_deny_write_execute(const Unit* u, const ExecContext *c) {
1570         assert(u);
1571         assert(c);
1572
1573         if (!c->memory_deny_write_execute)
1574                 return 0;
1575
1576         if (skip_seccomp_unavailable(u, "MemoryDenyWriteExecute="))
1577                 return 0;
1578
1579         return seccomp_memory_deny_write_execute();
1580 }
1581
1582 static int apply_restrict_realtime(const Unit* u, const ExecContext *c) {
1583         assert(u);
1584         assert(c);
1585
1586         if (!c->restrict_realtime)
1587                 return 0;
1588
1589         if (skip_seccomp_unavailable(u, "RestrictRealtime="))
1590                 return 0;
1591
1592         return seccomp_restrict_realtime();
1593 }
1594
1595 static int apply_restrict_suid_sgid(const Unit* u, const ExecContext *c) {
1596         assert(u);
1597         assert(c);
1598
1599         if (!c->restrict_suid_sgid)
1600                 return 0;
1601
1602         if (skip_seccomp_unavailable(u, "RestrictSUIDSGID="))
1603                 return 0;
1604
1605         return seccomp_restrict_suid_sgid();
1606 }
1607
1608 static int apply_protect_sysctl(const Unit *u, const ExecContext *c) {
1609         assert(u);
1610         assert(c);
1611
1612         /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1613          * let's protect even those systems where this is left on in the kernel. */
1614
1615         if (!c->protect_kernel_tunables)
1616                 return 0;
1617
1618         if (skip_seccomp_unavailable(u, "ProtectKernelTunables="))
1619                 return 0;
1620
1621         return seccomp_protect_sysctl();
1622 }
1623
1624 static int apply_protect_kernel_modules(const Unit *u, const ExecContext *c) {
1625         assert(u);
1626         assert(c);
1627
1628         /* Turn off module syscalls on ProtectKernelModules=yes */
1629
1630         if (!c->protect_kernel_modules)
1631                 return 0;
1632
1633         if (skip_seccomp_unavailable(u, "ProtectKernelModules="))
1634                 return 0;
1635
1636         return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_MODULE, SCMP_ACT_ERRNO(EPERM), false);
1637 }
1638
1639 static int apply_protect_kernel_logs(const Unit *u, const ExecContext *c) {
1640         assert(u);
1641         assert(c);
1642
1643         if (!c->protect_kernel_logs)
1644                 return 0;
1645
1646         if (skip_seccomp_unavailable(u, "ProtectKernelLogs="))
1647                 return 0;
1648
1649         return seccomp_protect_syslog();
1650 }
1651
1652 static int apply_protect_clock(const Unit *u, const ExecContext *c) {
1653         assert(u);
1654         assert(c);
1655
1656         if (!c->protect_clock)
1657                 return 0;
1658
1659         if (skip_seccomp_unavailable(u, "ProtectClock="))
1660                 return 0;
1661
1662         return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_CLOCK, SCMP_ACT_ERRNO(EPERM), false);
1663 }
1664
1665 static int apply_private_devices(const Unit *u, const ExecContext *c) {
1666         assert(u);
1667         assert(c);
1668
1669         /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1670
1671         if (!c->private_devices)
1672                 return 0;
1673
1674         if (skip_seccomp_unavailable(u, "PrivateDevices="))
1675                 return 0;
1676
1677         return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_RAW_IO, SCMP_ACT_ERRNO(EPERM), false);
1678 }
1679
1680 static int apply_restrict_namespaces(const Unit *u, const ExecContext *c) {
1681         assert(u);
1682         assert(c);
1683
1684         if (!exec_context_restrict_namespaces_set(c))
1685                 return 0;
1686
1687         if (skip_seccomp_unavailable(u, "RestrictNamespaces="))
1688                 return 0;
1689
1690         return seccomp_restrict_namespaces(c->restrict_namespaces);
1691 }
1692
1693 static int apply_lock_personality(const Unit* u, const ExecContext *c) {
1694         unsigned long personality;
1695         int r;
1696
1697         assert(u);
1698         assert(c);
1699
1700         if (!c->lock_personality)
1701                 return 0;
1702
1703         if (skip_seccomp_unavailable(u, "LockPersonality="))
1704                 return 0;
1705
1706         personality = c->personality;
1707
1708         /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1709         if (personality == PERSONALITY_INVALID) {
1710
1711                 r = opinionated_personality(&personality);
1712                 if (r < 0)
1713                         return r;
1714         }
1715
1716         return seccomp_lock_personality(personality);
1717 }
1718
1719 #endif
1720
1721 #if HAVE_LIBBPF
1722 static int apply_restrict_filesystems(Unit *u, const ExecContext *c) {
1723         assert(u);
1724         assert(c);
1725
1726         if (!exec_context_restrict_filesystems_set(c))
1727                 return 0;
1728
1729         if (!u->manager->restrict_fs) {
1730                 /* LSM BPF is unsupported or lsm_bpf_setup failed */
1731                 log_unit_debug(u, "LSM BPF not supported, skipping RestrictFileSystems=");
1732                 return 0;
1733         }
1734
1735         return lsm_bpf_unit_restrict_filesystems(u, c->restrict_filesystems, c->restrict_filesystems_allow_list);
1736 }
1737 #endif
1738
1739 static int apply_protect_hostname(const Unit *u, const ExecContext *c, int *ret_exit_status) {
1740         assert(u);
1741         assert(c);
1742
1743         if (!c->protect_hostname)
1744                 return 0;
1745
1746         if (ns_type_supported(NAMESPACE_UTS)) {
1747                 if (unshare(CLONE_NEWUTS) < 0) {
1748                         if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) {
1749                                 *ret_exit_status = EXIT_NAMESPACE;
1750                                 return log_unit_error_errno(u, errno, "Failed to set up UTS namespacing: %m");
1751                         }
1752
1753                         log_unit_warning(u, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1754                 }
1755         } else
1756                 log_unit_warning(u, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1757
1758 #if HAVE_SECCOMP
1759         int r;
1760
1761         if (skip_seccomp_unavailable(u, "ProtectHostname="))
1762                 return 0;
1763
1764         r = seccomp_protect_hostname();
1765         if (r < 0) {
1766                 *ret_exit_status = EXIT_SECCOMP;
1767                 return log_unit_error_errno(u, r, "Failed to apply hostname restrictions: %m");
1768         }
1769 #endif
1770
1771         return 0;
1772 }
1773
1774 static void do_idle_pipe_dance(int idle_pipe[static 4]) {
1775         assert(idle_pipe);
1776
1777         idle_pipe[1] = safe_close(idle_pipe[1]);
1778         idle_pipe[2] = safe_close(idle_pipe[2]);
1779
1780         if (idle_pipe[0] >= 0) {
1781                 int r;
1782
1783                 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1784
1785                 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
1786                         ssize_t n;
1787
1788                         /* Signal systemd that we are bored and want to continue. */
1789                         n = write(idle_pipe[3], "x", 1);
1790                         if (n > 0)
1791                                 /* Wait for systemd to react to the signal above. */
1792                                 (void) fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
1793                 }
1794
1795                 idle_pipe[0] = safe_close(idle_pipe[0]);
1796
1797         }
1798
1799         idle_pipe[3] = safe_close(idle_pipe[3]);
1800 }
1801
1802 static const char *exec_directory_env_name_to_string(ExecDirectoryType t);
1803
1804 static int build_environment(
1805                 const Unit *u,
1806                 const ExecContext *c,
1807                 const ExecParameters *p,
1808                 size_t n_fds,
1809                 char **fdnames,
1810                 const char *home,
1811                 const char *username,
1812                 const char *shell,
1813                 dev_t journal_stream_dev,
1814                 ino_t journal_stream_ino,
1815                 char ***ret) {
1816
1817         _cleanup_strv_free_ char **our_env = NULL;
1818         size_t n_env = 0;
1819         char *x;
1820
1821         assert(u);
1822         assert(c);
1823         assert(p);
1824         assert(ret);
1825
1826 #define N_ENV_VARS 17
1827         our_env = new0(char*, N_ENV_VARS + _EXEC_DIRECTORY_TYPE_MAX);
1828         if (!our_env)
1829                 return -ENOMEM;
1830
1831         if (n_fds > 0) {
1832                 _cleanup_free_ char *joined = NULL;
1833
1834                 if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid_cached()) < 0)
1835                         return -ENOMEM;
1836                 our_env[n_env++] = x;
1837
1838                 if (asprintf(&x, "LISTEN_FDS=%zu", n_fds) < 0)
1839                         return -ENOMEM;
1840                 our_env[n_env++] = x;
1841
1842                 joined = strv_join(fdnames, ":");
1843                 if (!joined)
1844                         return -ENOMEM;
1845
1846                 x = strjoin("LISTEN_FDNAMES=", joined);
1847                 if (!x)
1848                         return -ENOMEM;
1849                 our_env[n_env++] = x;
1850         }
1851
1852         if ((p->flags & EXEC_SET_WATCHDOG) && p->watchdog_usec > 0) {
1853                 if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid_cached()) < 0)
1854                         return -ENOMEM;
1855                 our_env[n_env++] = x;
1856
1857                 if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, p->watchdog_usec) < 0)
1858                         return -ENOMEM;
1859                 our_env[n_env++] = x;
1860         }
1861
1862         /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use blocking
1863          * Varlink calls back to us for look up dynamic users in PID 1. Break the deadlock between D-Bus and
1864          * PID 1 by disabling use of PID1' NSS interface for looking up dynamic users. */
1865         if (p->flags & EXEC_NSS_DYNAMIC_BYPASS) {
1866                 x = strdup("SYSTEMD_NSS_DYNAMIC_BYPASS=1");
1867                 if (!x)
1868                         return -ENOMEM;
1869                 our_env[n_env++] = x;
1870         }
1871
1872         if (home) {
1873                 x = strjoin("HOME=", home);
1874                 if (!x)
1875                         return -ENOMEM;
1876
1877                 path_simplify(x + 5);
1878                 our_env[n_env++] = x;
1879         }
1880
1881         if (username) {
1882                 x = strjoin("LOGNAME=", username);
1883                 if (!x)
1884                         return -ENOMEM;
1885                 our_env[n_env++] = x;
1886
1887                 x = strjoin("USER=", username);
1888                 if (!x)
1889                         return -ENOMEM;
1890                 our_env[n_env++] = x;
1891         }
1892
1893         if (shell) {
1894                 x = strjoin("SHELL=", shell);
1895                 if (!x)
1896                         return -ENOMEM;
1897
1898                 path_simplify(x + 6);
1899                 our_env[n_env++] = x;
1900         }
1901
1902         if (!sd_id128_is_null(u->invocation_id)) {
1903                 if (asprintf(&x, "INVOCATION_ID=" SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(u->invocation_id)) < 0)
1904                         return -ENOMEM;
1905
1906                 our_env[n_env++] = x;
1907         }
1908
1909         if (exec_context_needs_term(c)) {
1910                 const char *tty_path, *term = NULL;
1911
1912                 tty_path = exec_context_tty_path(c);
1913
1914                 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1915                  * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1916                  * container manager passes to PID 1 ends up all the way in the console login shown. */
1917
1918                 if (path_equal_ptr(tty_path, "/dev/console") && getppid() == 1)
1919                         term = getenv("TERM");
1920
1921                 if (!term)
1922                         term = default_term_for_tty(tty_path);
1923
1924                 x = strjoin("TERM=", term);
1925                 if (!x)
1926                         return -ENOMEM;
1927                 our_env[n_env++] = x;
1928         }
1929
1930         if (journal_stream_dev != 0 && journal_stream_ino != 0) {
1931                 if (asprintf(&x, "JOURNAL_STREAM=" DEV_FMT ":" INO_FMT, journal_stream_dev, journal_stream_ino) < 0)
1932                         return -ENOMEM;
1933
1934                 our_env[n_env++] = x;
1935         }
1936
1937         if (c->log_namespace) {
1938                 x = strjoin("LOG_NAMESPACE=", c->log_namespace);
1939                 if (!x)
1940                         return -ENOMEM;
1941
1942                 our_env[n_env++] = x;
1943         }
1944
1945         for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
1946                 _cleanup_free_ char *joined = NULL;
1947                 const char *n;
1948
1949                 if (!p->prefix[t])
1950                         continue;
1951
1952                 if (c->directories[t].n_items == 0)
1953                         continue;
1954
1955                 n = exec_directory_env_name_to_string(t);
1956                 if (!n)
1957                         continue;
1958
1959                 for (size_t i = 0; i < c->directories[t].n_items; i++) {
1960                         _cleanup_free_ char *prefixed = NULL;
1961
1962                         prefixed = path_join(p->prefix[t], c->directories[t].items[i].path);
1963                         if (!prefixed)
1964                                 return -ENOMEM;
1965
1966                         if (!strextend_with_separator(&joined, ":", prefixed))
1967                                 return -ENOMEM;
1968                 }
1969
1970                 x = strjoin(n, "=", joined);
1971                 if (!x)
1972                         return -ENOMEM;
1973
1974                 our_env[n_env++] = x;
1975         }
1976
1977         if (exec_context_has_credentials(c) && p->prefix[EXEC_DIRECTORY_RUNTIME]) {
1978                 x = strjoin("CREDENTIALS_DIRECTORY=", p->prefix[EXEC_DIRECTORY_RUNTIME], "/credentials/", u->id);
1979                 if (!x)
1980                         return -ENOMEM;
1981
1982                 our_env[n_env++] = x;
1983         }
1984
1985         if (asprintf(&x, "SYSTEMD_EXEC_PID=" PID_FMT, getpid_cached()) < 0)
1986                 return -ENOMEM;
1987
1988         our_env[n_env++] = x;
1989
1990         our_env[n_env++] = NULL;
1991         assert(n_env <= N_ENV_VARS + _EXEC_DIRECTORY_TYPE_MAX);
1992 #undef N_ENV_VARS
1993
1994         *ret = TAKE_PTR(our_env);
1995
1996         return 0;
1997 }
1998
1999 static int build_pass_environment(const ExecContext *c, char ***ret) {
2000         _cleanup_strv_free_ char **pass_env = NULL;
2001         size_t n_env = 0;
2002
2003         STRV_FOREACH(i, c->pass_environment) {
2004                 _cleanup_free_ char *x = NULL;
2005                 char *v;
2006
2007                 v = getenv(*i);
2008                 if (!v)
2009                         continue;
2010                 x = strjoin(*i, "=", v);
2011                 if (!x)
2012                         return -ENOMEM;
2013
2014                 if (!GREEDY_REALLOC(pass_env, n_env + 2))
2015                         return -ENOMEM;
2016
2017                 pass_env[n_env++] = TAKE_PTR(x);
2018                 pass_env[n_env] = NULL;
2019         }
2020
2021         *ret = TAKE_PTR(pass_env);
2022
2023         return 0;
2024 }
2025
2026 bool exec_needs_network_namespace(const ExecContext *context) {
2027         assert(context);
2028
2029         return context->private_network || context->network_namespace_path;
2030 }
2031
2032 static bool exec_needs_ipc_namespace(const ExecContext *context) {
2033         assert(context);
2034
2035         return context->private_ipc || context->ipc_namespace_path;
2036 }
2037
2038 bool exec_needs_mount_namespace(
2039                 const ExecContext *context,
2040                 const ExecParameters *params,
2041                 const ExecRuntime *runtime) {
2042
2043         assert(context);
2044
2045         if (context->root_image)
2046                 return true;
2047
2048         if (!strv_isempty(context->read_write_paths) ||
2049             !strv_isempty(context->read_only_paths) ||
2050             !strv_isempty(context->inaccessible_paths) ||
2051             !strv_isempty(context->exec_paths) ||
2052             !strv_isempty(context->no_exec_paths))
2053                 return true;
2054
2055         if (context->n_bind_mounts > 0)
2056                 return true;
2057
2058         if (context->n_temporary_filesystems > 0)
2059                 return true;
2060
2061         if (context->n_mount_images > 0)
2062                 return true;
2063
2064         if (context->n_extension_images > 0)
2065                 return true;
2066
2067         if (!strv_isempty(context->extension_directories))
2068                 return true;
2069
2070         if (!IN_SET(context->mount_flags, 0, MS_SHARED))
2071                 return true;
2072
2073         if (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir))
2074                 return true;
2075
2076         if (context->private_devices ||
2077             context->private_mounts ||
2078             context->protect_system != PROTECT_SYSTEM_NO ||
2079             context->protect_home != PROTECT_HOME_NO ||
2080             context->protect_kernel_tunables ||
2081             context->protect_kernel_modules ||
2082             context->protect_kernel_logs ||
2083             context->protect_control_groups ||
2084             context->protect_proc != PROTECT_PROC_DEFAULT ||
2085             context->proc_subset != PROC_SUBSET_ALL ||
2086             exec_needs_ipc_namespace(context))
2087                 return true;
2088
2089         if (context->root_directory) {
2090                 if (exec_context_get_effective_mount_apivfs(context))
2091                         return true;
2092
2093                 for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
2094                         if (params && !params->prefix[t])
2095                                 continue;
2096
2097                         if (context->directories[t].n_items > 0)
2098                                 return true;
2099                 }
2100         }
2101
2102         if (context->dynamic_user &&
2103             (context->directories[EXEC_DIRECTORY_STATE].n_items > 0 ||
2104              context->directories[EXEC_DIRECTORY_CACHE].n_items > 0 ||
2105              context->directories[EXEC_DIRECTORY_LOGS].n_items > 0))
2106                 return true;
2107
2108         if (context->log_namespace)
2109                 return true;
2110
2111         return false;
2112 }
2113
2114 static int setup_private_users(uid_t ouid, gid_t ogid, uid_t uid, gid_t gid) {
2115         _cleanup_free_ char *uid_map = NULL, *gid_map = NULL;
2116         _cleanup_close_pair_ int errno_pipe[2] = PIPE_EBADF;
2117         _cleanup_close_ int unshare_ready_fd = -EBADF;
2118         _cleanup_(sigkill_waitp) pid_t pid = 0;
2119         uint64_t c = 1;
2120         ssize_t n;
2121         int r;
2122
2123         /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2124          * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2125          * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2126          * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2127          * which waits for the parent to create the new user namespace while staying in the original namespace. The
2128          * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2129          * continues execution normally.
2130          * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2131          * does not need CAP_SETUID to write the single line mapping to itself. */
2132
2133         /* Can only set up multiple mappings with CAP_SETUID. */
2134         if (have_effective_cap(CAP_SETUID) > 0 && uid != ouid && uid_is_valid(uid))
2135                 r = asprintf(&uid_map,
2136                              UID_FMT " " UID_FMT " 1\n"     /* Map $OUID → $OUID */
2137                              UID_FMT " " UID_FMT " 1\n",    /* Map $UID → $UID */
2138                              ouid, ouid, uid, uid);
2139         else
2140                 r = asprintf(&uid_map,
2141                              UID_FMT " " UID_FMT " 1\n",    /* Map $OUID → $OUID */
2142                              ouid, ouid);
2143
2144         if (r < 0)
2145                 return -ENOMEM;
2146
2147         /* Can only set up multiple mappings with CAP_SETGID. */
2148         if (have_effective_cap(CAP_SETGID) > 0 && gid != ogid && gid_is_valid(gid))
2149                 r = asprintf(&gid_map,
2150                              GID_FMT " " GID_FMT " 1\n"     /* Map $OGID → $OGID */
2151                              GID_FMT " " GID_FMT " 1\n",    /* Map $GID → $GID */
2152                              ogid, ogid, gid, gid);
2153         else
2154                 r = asprintf(&gid_map,
2155                              GID_FMT " " GID_FMT " 1\n",    /* Map $OGID -> $OGID */
2156                              ogid, ogid);
2157
2158         if (r < 0)
2159                 return -ENOMEM;
2160
2161         /* Create a communication channel so that the parent can tell the child when it finished creating the user
2162          * namespace. */
2163         unshare_ready_fd = eventfd(0, EFD_CLOEXEC);
2164         if (unshare_ready_fd < 0)
2165                 return -errno;
2166
2167         /* Create a communication channel so that the child can tell the parent a proper error code in case it
2168          * failed. */
2169         if (pipe2(errno_pipe, O_CLOEXEC) < 0)
2170                 return -errno;
2171
2172         r = safe_fork("(sd-userns)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &pid);
2173         if (r < 0)
2174                 return r;
2175         if (r == 0) {
2176                 _cleanup_close_ int fd = -EBADF;
2177                 const char *a;
2178                 pid_t ppid;
2179
2180                 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2181                  * here, after the parent opened its own user namespace. */
2182
2183                 ppid = getppid();
2184                 errno_pipe[0] = safe_close(errno_pipe[0]);
2185
2186                 /* Wait until the parent unshared the user namespace */
2187                 if (read(unshare_ready_fd, &c, sizeof(c)) < 0) {
2188                         r = -errno;
2189                         goto child_fail;
2190                 }
2191
2192                 /* Disable the setgroups() system call in the child user namespace, for good. */
2193                 a = procfs_file_alloca(ppid, "setgroups");
2194                 fd = open(a, O_WRONLY|O_CLOEXEC);
2195                 if (fd < 0) {
2196                         if (errno != ENOENT) {
2197                                 r = -errno;
2198                                 goto child_fail;
2199                         }
2200
2201                         /* If the file is missing the kernel is too old, let's continue anyway. */
2202                 } else {
2203                         if (write(fd, "deny\n", 5) < 0) {
2204                                 r = -errno;
2205                                 goto child_fail;
2206                         }
2207
2208                         fd = safe_close(fd);
2209                 }
2210
2211                 /* First write the GID map */
2212                 a = procfs_file_alloca(ppid, "gid_map");
2213                 fd = open(a, O_WRONLY|O_CLOEXEC);
2214                 if (fd < 0) {
2215                         r = -errno;
2216                         goto child_fail;
2217                 }
2218                 if (write(fd, gid_map, strlen(gid_map)) < 0) {
2219                         r = -errno;
2220                         goto child_fail;
2221                 }
2222                 fd = safe_close(fd);
2223
2224                 /* The write the UID map */
2225                 a = procfs_file_alloca(ppid, "uid_map");
2226                 fd = open(a, O_WRONLY|O_CLOEXEC);
2227                 if (fd < 0) {
2228                         r = -errno;
2229                         goto child_fail;
2230                 }
2231                 if (write(fd, uid_map, strlen(uid_map)) < 0) {
2232                         r = -errno;
2233                         goto child_fail;
2234                 }
2235
2236                 _exit(EXIT_SUCCESS);
2237
2238         child_fail:
2239                 (void) write(errno_pipe[1], &r, sizeof(r));
2240                 _exit(EXIT_FAILURE);
2241         }
2242
2243         errno_pipe[1] = safe_close(errno_pipe[1]);
2244
2245         if (unshare(CLONE_NEWUSER) < 0)
2246                 return -errno;
2247
2248         /* Let the child know that the namespace is ready now */
2249         if (write(unshare_ready_fd, &c, sizeof(c)) < 0)
2250                 return -errno;
2251
2252         /* Try to read an error code from the child */
2253         n = read(errno_pipe[0], &r, sizeof(r));
2254         if (n < 0)
2255                 return -errno;
2256         if (n == sizeof(r)) { /* an error code was sent to us */
2257                 if (r < 0)
2258                         return r;
2259                 return -EIO;
2260         }
2261         if (n != 0) /* on success we should have read 0 bytes */
2262                 return -EIO;
2263
2264         r = wait_for_terminate_and_check("(sd-userns)", TAKE_PID(pid), 0);
2265         if (r < 0)
2266                 return r;
2267         if (r != EXIT_SUCCESS) /* If something strange happened with the child, let's consider this fatal, too */
2268                 return -EIO;
2269
2270         return 0;
2271 }
2272
2273 static bool exec_directory_is_private(const ExecContext *context, ExecDirectoryType type) {
2274         if (!context->dynamic_user)
2275                 return false;
2276
2277         if (type == EXEC_DIRECTORY_CONFIGURATION)
2278                 return false;
2279
2280         if (type == EXEC_DIRECTORY_RUNTIME && context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO)
2281                 return false;
2282
2283         return true;
2284 }
2285
2286 static int create_many_symlinks(const char *root, const char *source, char **symlinks) {
2287         _cleanup_free_ char *src_abs = NULL;
2288         int r;
2289
2290         assert(source);
2291
2292         src_abs = path_join(root, source);
2293         if (!src_abs)
2294                 return -ENOMEM;
2295
2296         STRV_FOREACH(dst, symlinks) {
2297                 _cleanup_free_ char *dst_abs = NULL;
2298
2299                 dst_abs = path_join(root, *dst);
2300                 if (!dst_abs)
2301                         return -ENOMEM;
2302
2303                 r = mkdir_parents_label(dst_abs, 0755);
2304                 if (r < 0)
2305                         return r;
2306
2307                 r = symlink_idempotent(src_abs, dst_abs, true);
2308                 if (r < 0)
2309                         return r;
2310         }
2311
2312         return 0;
2313 }
2314
2315 static int setup_exec_directory(
2316                 const ExecContext *context,
2317                 const ExecParameters *params,
2318                 uid_t uid,
2319                 gid_t gid,
2320                 ExecDirectoryType type,
2321                 bool needs_mount_namespace,
2322                 int *exit_status) {
2323
2324         static const int exit_status_table[_EXEC_DIRECTORY_TYPE_MAX] = {
2325                 [EXEC_DIRECTORY_RUNTIME] = EXIT_RUNTIME_DIRECTORY,
2326                 [EXEC_DIRECTORY_STATE] = EXIT_STATE_DIRECTORY,
2327                 [EXEC_DIRECTORY_CACHE] = EXIT_CACHE_DIRECTORY,
2328                 [EXEC_DIRECTORY_LOGS] = EXIT_LOGS_DIRECTORY,
2329                 [EXEC_DIRECTORY_CONFIGURATION] = EXIT_CONFIGURATION_DIRECTORY,
2330         };
2331         int r;
2332
2333         assert(context);
2334         assert(params);
2335         assert(type >= 0 && type < _EXEC_DIRECTORY_TYPE_MAX);
2336         assert(exit_status);
2337
2338         if (!params->prefix[type])
2339                 return 0;
2340
2341         if (params->flags & EXEC_CHOWN_DIRECTORIES) {
2342                 if (!uid_is_valid(uid))
2343                         uid = 0;
2344                 if (!gid_is_valid(gid))
2345                         gid = 0;
2346         }
2347
2348         for (size_t i = 0; i < context->directories[type].n_items; i++) {
2349                 _cleanup_free_ char *p = NULL, *pp = NULL;
2350
2351                 p = path_join(params->prefix[type], context->directories[type].items[i].path);
2352                 if (!p) {
2353                         r = -ENOMEM;
2354                         goto fail;
2355                 }
2356
2357                 r = mkdir_parents_label(p, 0755);
2358                 if (r < 0)
2359                         goto fail;
2360
2361                 if (exec_directory_is_private(context, type)) {
2362                         /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2363                          * case we want to avoid leaving a directory around fully accessible that is owned by
2364                          * a dynamic user whose UID is later on reused. To lock this down we use the same
2365                          * trick used by container managers to prohibit host users to get access to files of
2366                          * the same UID in containers: we place everything inside a directory that has an
2367                          * access mode of 0700 and is owned root:root, so that it acts as security boundary
2368                          * for unprivileged host code. We then use fs namespacing to make this directory
2369                          * permeable for the service itself.
2370                          *
2371                          * Specifically: for a service which wants a special directory "foo/" we first create
2372                          * a directory "private/" with access mode 0700 owned by root:root. Then we place
2373                          * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2374                          * "private/foo". This way, privileged host users can access "foo/" as usual, but
2375                          * unprivileged host users can't look into it. Inside of the namespace of the unit
2376                          * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2377                          * "private/foo/" is mounted under the same name, thus disabling the access boundary
2378                          * for the service and making sure it only gets access to the dirs it needs but no
2379                          * others. Tricky? Yes, absolutely, but it works!
2380                          *
2381                          * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2382                          * to be owned by the service itself.
2383                          *
2384                          * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2385                          * for sharing files or sockets with other services. */
2386
2387                         pp = path_join(params->prefix[type], "private");
2388                         if (!pp) {
2389                                 r = -ENOMEM;
2390                                 goto fail;
2391                         }
2392
2393                         /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2394                         r = mkdir_safe_label(pp, 0700, 0, 0, MKDIR_WARN_MODE);
2395                         if (r < 0)
2396                                 goto fail;
2397
2398                         if (!path_extend(&pp, context->directories[type].items[i].path)) {
2399                                 r = -ENOMEM;
2400                                 goto fail;
2401                         }
2402
2403                         /* Create all directories between the configured directory and this private root, and mark them 0755 */
2404                         r = mkdir_parents_label(pp, 0755);
2405                         if (r < 0)
2406                                 goto fail;
2407
2408                         if (is_dir(p, false) > 0 &&
2409                             (laccess(pp, F_OK) < 0 && errno == ENOENT)) {
2410
2411                                 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2412                                  * it over. Most likely the service has been upgraded from one that didn't use
2413                                  * DynamicUser=1, to one that does. */
2414
2415                                 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2416                                          "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2417                                          exec_directory_type_to_string(type), p, pp);
2418
2419                                 if (rename(p, pp) < 0) {
2420                                         r = -errno;
2421                                         goto fail;
2422                                 }
2423                         } else {
2424                                 /* Otherwise, create the actual directory for the service */
2425
2426                                 r = mkdir_label(pp, context->directories[type].mode);
2427                                 if (r < 0 && r != -EEXIST)
2428                                         goto fail;
2429                         }
2430
2431                         if (!context->directories[type].items[i].only_create) {
2432                                 /* And link it up from the original place.
2433                                  * Notes
2434                                  * 1) If a mount namespace is going to be used, then this symlink remains on
2435                                  *    the host, and a new one for the child namespace will be created later.
2436                                  * 2) It is not necessary to create this symlink when one of its parent
2437                                  *    directories is specified and already created. E.g.
2438                                  *        StateDirectory=foo foo/bar
2439                                  *    In that case, the inode points to pp and p for "foo/bar" are the same:
2440                                  *        pp = "/var/lib/private/foo/bar"
2441                                  *        p = "/var/lib/foo/bar"
2442                                  *    and, /var/lib/foo is a symlink to /var/lib/private/foo. So, not only
2443                                  *    we do not need to create the symlink, but we cannot create the symlink.
2444                                  *    See issue #24783. */
2445                                 r = symlink_idempotent(pp, p, true);
2446                                 if (r < 0)
2447                                         goto fail;
2448                         }
2449
2450                 } else {
2451                         _cleanup_free_ char *target = NULL;
2452
2453                         if (type != EXEC_DIRECTORY_CONFIGURATION &&
2454                             readlink_and_make_absolute(p, &target) >= 0) {
2455                                 _cleanup_free_ char *q = NULL, *q_resolved = NULL, *target_resolved = NULL;
2456
2457                                 /* This already exists and is a symlink? Interesting. Maybe it's one created
2458                                  * by DynamicUser=1 (see above)?
2459                                  *
2460                                  * We do this for all directory types except for ConfigurationDirectory=,
2461                                  * since they all support the private/ symlink logic at least in some
2462                                  * configurations, see above. */
2463
2464                                 r = chase_symlinks(target, NULL, 0, &target_resolved, NULL);
2465                                 if (r < 0)
2466                                         goto fail;
2467
2468                                 q = path_join(params->prefix[type], "private", context->directories[type].items[i].path);
2469                                 if (!q) {
2470                                         r = -ENOMEM;
2471                                         goto fail;
2472                                 }
2473
2474                                 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2475                                 r = chase_symlinks(q, NULL, CHASE_NONEXISTENT, &q_resolved, NULL);
2476                                 if (r < 0)
2477                                         goto fail;
2478
2479                                 if (path_equal(q_resolved, target_resolved)) {
2480
2481                                         /* Hmm, apparently DynamicUser= was once turned on for this service,
2482                                          * but is no longer. Let's move the directory back up. */
2483
2484                                         log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2485                                                  "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2486                                                  exec_directory_type_to_string(type), q, p);
2487
2488                                         if (unlink(p) < 0) {
2489                                                 r = -errno;
2490                                                 goto fail;
2491                                         }
2492
2493                                         if (rename(q, p) < 0) {
2494                                                 r = -errno;
2495                                                 goto fail;
2496                                         }
2497                                 }
2498                         }
2499
2500                         r = mkdir_label(p, context->directories[type].mode);
2501                         if (r < 0) {
2502                                 if (r != -EEXIST)
2503                                         goto fail;
2504
2505                                 if (type == EXEC_DIRECTORY_CONFIGURATION) {
2506                                         struct stat st;
2507
2508                                         /* Don't change the owner/access mode of the configuration directory,
2509                                          * as in the common case it is not written to by a service, and shall
2510                                          * not be writable. */
2511
2512                                         if (stat(p, &st) < 0) {
2513                                                 r = -errno;
2514                                                 goto fail;
2515                                         }
2516
2517                                         /* Still complain if the access mode doesn't match */
2518                                         if (((st.st_mode ^ context->directories[type].mode) & 07777) != 0)
2519                                                 log_warning("%s \'%s\' already exists but the mode is different. "
2520                                                             "(File system: %o %sMode: %o)",
2521                                                             exec_directory_type_to_string(type), context->directories[type].items[i].path,
2522                                                             st.st_mode & 07777, exec_directory_type_to_string(type), context->directories[type].mode & 07777);
2523
2524                                         continue;
2525                                 }
2526                         }
2527                 }
2528
2529                 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2530                  * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2531                  * current UID/GID ownership.) */
2532                 r = chmod_and_chown(pp ?: p, context->directories[type].mode, UID_INVALID, GID_INVALID);
2533                 if (r < 0)
2534                         goto fail;
2535
2536                 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2537                  * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2538                  * assignments to exist. */
2539                 r = path_chown_recursive(pp ?: p, uid, gid, context->dynamic_user ? 01777 : 07777);
2540                 if (r < 0)
2541                         goto fail;
2542         }
2543
2544         /* If we are not going to run in a namespace, set up the symlinks - otherwise
2545          * they are set up later, to allow configuring empty var/run/etc. */
2546         if (!needs_mount_namespace)
2547                 for (size_t i = 0; i < context->directories[type].n_items; i++) {
2548                         r = create_many_symlinks(params->prefix[type],
2549                                                  context->directories[type].items[i].path,
2550                                                  context->directories[type].items[i].symlinks);
2551                         if (r < 0)
2552                                 goto fail;
2553                 }
2554
2555         return 0;
2556
2557 fail:
2558         *exit_status = exit_status_table[type];
2559         return r;
2560 }
2561
2562 static int write_credential(
2563                 int dfd,
2564                 const char *id,
2565                 const void *data,
2566                 size_t size,
2567                 uid_t uid,
2568                 bool ownership_ok) {
2569
2570         _cleanup_(unlink_and_freep) char *tmp = NULL;
2571         _cleanup_close_ int fd = -EBADF;
2572         int r;
2573
2574         r = tempfn_random_child("", "cred", &tmp);
2575         if (r < 0)
2576                 return r;
2577
2578         fd = openat(dfd, tmp, O_CREAT|O_RDWR|O_CLOEXEC|O_EXCL|O_NOFOLLOW|O_NOCTTY, 0600);
2579         if (fd < 0) {
2580                 tmp = mfree(tmp);
2581                 return -errno;
2582         }
2583
2584         r = loop_write(fd, data, size, /* do_poll = */ false);
2585         if (r < 0)
2586                 return r;
2587
2588         if (fchmod(fd, 0400) < 0) /* Take away "w" bit */
2589                 return -errno;
2590
2591         if (uid_is_valid(uid) && uid != getuid()) {
2592                 r = fd_add_uid_acl_permission(fd, uid, ACL_READ);
2593                 if (r < 0) {
2594                         if (!ERRNO_IS_NOT_SUPPORTED(r) && !ERRNO_IS_PRIVILEGE(r))
2595                                 return r;
2596
2597                         if (!ownership_ok) /* Ideally we use ACLs, since we can neatly express what we want
2598                                             * to express: that the user gets read access and nothing
2599                                             * else. But if the backing fs can't support that (e.g. ramfs)
2600                                             * then we can use file ownership instead. But that's only safe if
2601                                             * we can then re-mount the whole thing read-only, so that the
2602                                             * user can no longer chmod() the file to gain write access. */
2603                                 return r;
2604
2605                         if (fchown(fd, uid, GID_INVALID) < 0)
2606                                 return -errno;
2607                 }
2608         }
2609
2610         if (renameat(dfd, tmp, dfd, id) < 0)
2611                 return -errno;
2612
2613         tmp = mfree(tmp);
2614         return 0;
2615 }
2616
2617 static char **credential_search_path(
2618                 const ExecParameters *params,
2619                 bool encrypted) {
2620
2621         _cleanup_strv_free_ char **l = NULL;
2622
2623         assert(params);
2624
2625         /* Assemble a search path to find credentials in. We'll look in /etc/credstore/ (and similar
2626          * directories in /usr/lib/ + /run/) for all types of credentials. If we are looking for encrypted
2627          * credentials, also look in /etc/credstore.encrypted/ (and similar dirs). */
2628
2629         if (encrypted) {
2630                 if (strv_extend(&l, params->received_encrypted_credentials_directory) < 0)
2631                         return NULL;
2632
2633                 if (strv_extend_strv(&l, CONF_PATHS_STRV("credstore.encrypted"), /* filter_duplicates= */ true) < 0)
2634                         return NULL;
2635         }
2636
2637         if (params->received_credentials_directory)
2638                 if (strv_extend(&l, params->received_credentials_directory) < 0)
2639                         return NULL;
2640
2641         if (strv_extend_strv(&l, CONF_PATHS_STRV("credstore"), /* filter_duplicates= */ true) < 0)
2642                 return NULL;
2643
2644         if (DEBUG_LOGGING) {
2645                 _cleanup_free_ char *t = strv_join(l, ":");
2646
2647                 log_debug("Credential search path is: %s", t);
2648         }
2649
2650         return TAKE_PTR(l);
2651 }
2652
2653 static int load_credential(
2654                 const ExecContext *context,
2655                 const ExecParameters *params,
2656                 const char *id,
2657                 const char *path,
2658                 bool encrypted,
2659                 const char *unit,
2660                 int read_dfd,
2661                 int write_dfd,
2662                 uid_t uid,
2663                 bool ownership_ok,
2664                 uint64_t *left) {
2665
2666         ReadFullFileFlags flags = READ_FULL_FILE_SECURE|READ_FULL_FILE_FAIL_WHEN_LARGER;
2667         _cleanup_strv_free_ char **search_path = NULL;
2668         _cleanup_(erase_and_freep) char *data = NULL;
2669         _cleanup_free_ char *bindname = NULL;
2670         const char *source = NULL;
2671         bool missing_ok = true;
2672         size_t size, add, maxsz;
2673         int r;
2674
2675         assert(context);
2676         assert(params);
2677         assert(id);
2678         assert(path);
2679         assert(unit);
2680         assert(read_dfd >= 0 || read_dfd == AT_FDCWD);
2681         assert(write_dfd >= 0);
2682         assert(left);
2683
2684         if (read_dfd >= 0) {
2685                 /* If a directory fd is specified, then read the file directly from that dir. In this case we
2686                  * won't do AF_UNIX stuff (we simply don't want to recursively iterate down a tree of AF_UNIX
2687                  * IPC sockets). It's OK if a file vanishes here in the time we enumerate it and intend to
2688                  * open it. */
2689
2690                 if (!filename_is_valid(path)) /* safety check */
2691                         return -EINVAL;
2692
2693                 missing_ok = true;
2694                 source = path;
2695
2696         } else if (path_is_absolute(path)) {
2697                 /* If this is an absolute path, read the data directly from it, and support AF_UNIX
2698                  * sockets */
2699
2700                 if (!path_is_valid(path)) /* safety check */
2701                         return -EINVAL;
2702
2703                 flags |= READ_FULL_FILE_CONNECT_SOCKET;
2704
2705                 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2706                  * via the source socket address in case we read off an AF_UNIX socket. */
2707                 if (asprintf(&bindname, "@%" PRIx64"/unit/%s/%s", random_u64(), unit, id) < 0)
2708                         return -ENOMEM;
2709
2710                 missing_ok = false;
2711                 source = path;
2712
2713         } else if (credential_name_valid(path)) {
2714                 /* If this is a relative path, take it as credential name relative to the credentials
2715                  * directory we received ourselves. We don't support the AF_UNIX stuff in this mode, since we
2716                  * are operating on a credential store, i.e. this is guaranteed to be regular files. */
2717
2718                 search_path = credential_search_path(params, encrypted);
2719                 if (!search_path)
2720                         return -ENOMEM;
2721
2722                 missing_ok = true;
2723         } else
2724                 source = NULL;
2725
2726         if (encrypted)
2727                 flags |= READ_FULL_FILE_UNBASE64;
2728
2729         maxsz = encrypted ? CREDENTIAL_ENCRYPTED_SIZE_MAX : CREDENTIAL_SIZE_MAX;
2730
2731         if (search_path) {
2732                 STRV_FOREACH(d, search_path) {
2733                         _cleanup_free_ char *j = NULL;
2734
2735                         j = path_join(*d, path);
2736                         if (!j)
2737                                 return -ENOMEM;
2738
2739                         r = read_full_file_full(
2740                                         AT_FDCWD, j, /* path is absolute, hence pass AT_FDCWD as nop dir fd here */
2741                                         UINT64_MAX,
2742                                         maxsz,
2743                                         flags,
2744                                         NULL,
2745                                         &data, &size);
2746                         if (r != -ENOENT)
2747                                 break;
2748                 }
2749         } else if (source)
2750                 r = read_full_file_full(
2751                                 read_dfd, source,
2752                                 UINT64_MAX,
2753                                 maxsz,
2754                                 flags,
2755                                 bindname,
2756                                 &data, &size);
2757         else
2758                 r = -ENOENT;
2759
2760         if (r == -ENOENT && (missing_ok || hashmap_contains(context->set_credentials, id))) {
2761                 /* Make a missing inherited credential non-fatal, let's just continue. After all apps
2762                  * will get clear errors if we don't pass such a missing credential on as they
2763                  * themselves will get ENOENT when trying to read them, which should not be much
2764                  * worse than when we handle the error here and make it fatal.
2765                  *
2766                  * Also, if the source file doesn't exist, but a fallback is set via SetCredentials=
2767                  * we are fine, too. */
2768                 log_debug_errno(r, "Couldn't read inherited credential '%s', skipping: %m", path);
2769                 return 0;
2770         }
2771         if (r < 0)
2772                 return log_debug_errno(r, "Failed to read credential '%s': %m", path);
2773
2774         if (encrypted) {
2775                 _cleanup_free_ void *plaintext = NULL;
2776                 size_t plaintext_size = 0;
2777
2778                 r = decrypt_credential_and_warn(id, now(CLOCK_REALTIME), NULL, NULL, data, size, &plaintext, &plaintext_size);
2779                 if (r < 0)
2780                         return r;
2781
2782                 free_and_replace(data, plaintext);
2783                 size = plaintext_size;
2784         }
2785
2786         add = strlen(id) + size;
2787         if (add > *left)
2788                 return -E2BIG;
2789
2790         r = write_credential(write_dfd, id, data, size, uid, ownership_ok);
2791         if (r < 0)
2792                 return log_debug_errno(r, "Failed to write credential '%s': %m", id);
2793
2794         *left -= add;
2795         return 0;
2796 }
2797
2798 struct load_cred_args {
2799         const ExecContext *context;
2800         const ExecParameters *params;
2801         bool encrypted;
2802         const char *unit;
2803         int dfd;
2804         uid_t uid;
2805         bool ownership_ok;
2806         uint64_t *left;
2807 };
2808
2809 static int load_cred_recurse_dir_cb(
2810                 RecurseDirEvent event,
2811                 const char *path,
2812                 int dir_fd,
2813                 int inode_fd,
2814                 const struct dirent *de,
2815                 const struct statx *sx,
2816                 void *userdata) {
2817
2818         struct load_cred_args *args = ASSERT_PTR(userdata);
2819         _cleanup_free_ char *sub_id = NULL;
2820         int r;
2821
2822         if (event != RECURSE_DIR_ENTRY)
2823                 return RECURSE_DIR_CONTINUE;
2824
2825         if (!IN_SET(de->d_type, DT_REG, DT_SOCK))
2826                 return RECURSE_DIR_CONTINUE;
2827
2828         sub_id = strreplace(path, "/", "_");
2829         if (!sub_id)
2830                 return -ENOMEM;
2831
2832         if (!credential_name_valid(sub_id))
2833                 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Credential would get ID %s, which is not valid, refusing", sub_id);
2834
2835         if (faccessat(args->dfd, sub_id, F_OK, AT_SYMLINK_NOFOLLOW) >= 0) {
2836                 log_debug("Skipping credential with duplicated ID %s at %s", sub_id, path);
2837                 return RECURSE_DIR_CONTINUE;
2838         }
2839         if (errno != ENOENT)
2840                 return log_debug_errno(errno, "Failed to test if credential %s exists: %m", sub_id);
2841
2842         r = load_credential(
2843                         args->context,
2844                         args->params,
2845                         sub_id,
2846                         de->d_name,
2847                         args->encrypted,
2848                         args->unit,
2849                         dir_fd,
2850                         args->dfd,
2851                         args->uid,
2852                         args->ownership_ok,
2853                         args->left);
2854         if (r < 0)
2855                 return r;
2856
2857         return RECURSE_DIR_CONTINUE;
2858 }
2859
2860 static int acquire_credentials(
2861                 const ExecContext *context,
2862                 const ExecParameters *params,
2863                 const char *unit,
2864                 const char *p,
2865                 uid_t uid,
2866                 bool ownership_ok) {
2867
2868         uint64_t left = CREDENTIALS_TOTAL_SIZE_MAX;
2869         _cleanup_close_ int dfd = -EBADF;
2870         ExecLoadCredential *lc;
2871         ExecSetCredential *sc;
2872         int r;
2873
2874         assert(context);
2875         assert(p);
2876
2877         dfd = open(p, O_DIRECTORY|O_CLOEXEC);
2878         if (dfd < 0)
2879                 return -errno;
2880
2881         /* First, load credentials off disk (or acquire via AF_UNIX socket) */
2882         HASHMAP_FOREACH(lc, context->load_credentials) {
2883                 _cleanup_close_ int sub_fd = -EBADF;
2884
2885                 /* If this is an absolute path, then try to open it as a directory. If that works, then we'll
2886                  * recurse into it. If it is an absolute path but it isn't a directory, then we'll open it as
2887                  * a regular file. Finally, if it's a relative path we will use it as a credential name to
2888                  * propagate a credential passed to us from further up. */
2889
2890                 if (path_is_absolute(lc->path)) {
2891                         sub_fd = open(lc->path, O_DIRECTORY|O_CLOEXEC|O_RDONLY);
2892                         if (sub_fd < 0 && !IN_SET(errno,
2893                                                   ENOTDIR,  /* Not a directory */
2894                                                   ENOENT))  /* Doesn't exist? */
2895                                 return log_debug_errno(errno, "Failed to open '%s': %m", lc->path);
2896                 }
2897
2898                 if (sub_fd < 0)
2899                         /* Regular file (incl. a credential passed in from higher up) */
2900                         r = load_credential(
2901                                         context,
2902                                         params,
2903                                         lc->id,
2904                                         lc->path,
2905                                         lc->encrypted,
2906                                         unit,
2907                                         AT_FDCWD,
2908                                         dfd,
2909                                         uid,
2910                                         ownership_ok,
2911                                         &left);
2912                 else
2913                         /* Directory */
2914                         r = recurse_dir(
2915                                         sub_fd,
2916                                         /* path= */ lc->id, /* recurse_dir() will suffix the subdir paths from here to the top-level id */
2917                                         /* statx_mask= */ 0,
2918                                         /* n_depth_max= */ UINT_MAX,
2919                                         RECURSE_DIR_SORT|RECURSE_DIR_IGNORE_DOT|RECURSE_DIR_ENSURE_TYPE,
2920                                         load_cred_recurse_dir_cb,
2921                                         &(struct load_cred_args) {
2922                                                 .context = context,
2923                                                 .params = params,
2924                                                 .encrypted = lc->encrypted,
2925                                                 .unit = unit,
2926                                                 .dfd = dfd,
2927                                                 .uid = uid,
2928                                                 .ownership_ok = ownership_ok,
2929                                                 .left = &left,
2930                                         });
2931                 if (r < 0)
2932                         return r;
2933         }
2934
2935         /* Second, we add in literally specified credentials. If the credentials already exist, we'll not add
2936          * them, so that they can act as a "default" if the same credential is specified multiple times. */
2937         HASHMAP_FOREACH(sc, context->set_credentials) {
2938                 _cleanup_(erase_and_freep) void *plaintext = NULL;
2939                 const char *data;
2940                 size_t size, add;
2941
2942                 /* Note that we check ahead of time here instead of relying on O_EXCL|O_CREAT later to return
2943                  * EEXIST if the credential already exists. That's because the TPM2-based decryption is kinda
2944                  * slow and involved, hence it's nice to be able to skip that if the credential already
2945                  * exists anyway. */
2946                 if (faccessat(dfd, sc->id, F_OK, AT_SYMLINK_NOFOLLOW) >= 0)
2947                         continue;
2948                 if (errno != ENOENT)
2949                         return log_debug_errno(errno, "Failed to test if credential %s exists: %m", sc->id);
2950
2951                 if (sc->encrypted) {
2952                         r = decrypt_credential_and_warn(sc->id, now(CLOCK_REALTIME), NULL, NULL, sc->data, sc->size, &plaintext, &size);
2953                         if (r < 0)
2954                                 return r;
2955
2956                         data = plaintext;
2957                 } else {
2958                         data = sc->data;
2959                         size = sc->size;
2960                 }
2961
2962                 add = strlen(sc->id) + size;
2963                 if (add > left)
2964                         return -E2BIG;
2965
2966                 r = write_credential(dfd, sc->id, data, size, uid, ownership_ok);
2967                 if (r < 0)
2968                         return r;
2969
2970                 left -= add;
2971         }
2972
2973         if (fchmod(dfd, 0500) < 0) /* Now take away the "w" bit */
2974                 return -errno;
2975
2976         /* After we created all keys with the right perms, also make sure the credential store as a whole is
2977          * accessible */
2978
2979         if (uid_is_valid(uid) && uid != getuid()) {
2980                 r = fd_add_uid_acl_permission(dfd, uid, ACL_READ | ACL_EXECUTE);
2981                 if (r < 0) {
2982                         if (!ERRNO_IS_NOT_SUPPORTED(r) && !ERRNO_IS_PRIVILEGE(r))
2983                                 return r;
2984
2985                         if (!ownership_ok)
2986                                 return r;
2987
2988                         if (fchown(dfd, uid, GID_INVALID) < 0)
2989                                 return -errno;
2990                 }
2991         }
2992
2993         return 0;
2994 }
2995
2996 static int setup_credentials_internal(
2997                 const ExecContext *context,
2998                 const ExecParameters *params,
2999                 const char *unit,
3000                 const char *final,        /* This is where the credential store shall eventually end up at */
3001                 const char *workspace,    /* This is where we can prepare it before moving it to the final place */
3002                 bool reuse_workspace,     /* Whether to reuse any existing workspace mount if it already is a mount */
3003                 bool must_mount,          /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
3004                 uid_t uid) {
3005
3006         int r, workspace_mounted; /* negative if we don't know yet whether we have/can mount something; true
3007                                    * if we mounted something; false if we definitely can't mount anything */
3008         bool final_mounted;
3009         const char *where;
3010
3011         assert(context);
3012         assert(final);
3013         assert(workspace);
3014
3015         if (reuse_workspace) {
3016                 r = path_is_mount_point(workspace, NULL, 0);
3017                 if (r < 0)
3018                         return r;
3019                 if (r > 0)
3020                         workspace_mounted = true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
3021                 else
3022                         workspace_mounted = -1; /* We need to figure out if we can mount something to the workspace */
3023         } else
3024                 workspace_mounted = -1; /* ditto */
3025
3026         r = path_is_mount_point(final, NULL, 0);
3027         if (r < 0)
3028                 return r;
3029         if (r > 0) {
3030                 /* If the final place already has something mounted, we use that. If the workspace also has
3031                  * something mounted we assume it's actually the same mount (but with MS_RDONLY
3032                  * different). */
3033                 final_mounted = true;
3034
3035                 if (workspace_mounted < 0) {
3036                         /* If the final place is mounted, but the workspace isn't, then let's bind mount
3037                          * the final version to the workspace, and make it writable, so that we can make
3038                          * changes */
3039
3040                         r = mount_nofollow_verbose(LOG_DEBUG, final, workspace, NULL, MS_BIND|MS_REC, NULL);
3041                         if (r < 0)
3042                                 return r;
3043
3044                         r = mount_nofollow_verbose(LOG_DEBUG, NULL, workspace, NULL, MS_BIND|MS_REMOUNT|MS_NODEV|MS_NOEXEC|MS_NOSUID, NULL);
3045                         if (r < 0)
3046                                 return r;
3047
3048                         workspace_mounted = true;
3049                 }
3050         } else
3051                 final_mounted = false;
3052
3053         if (workspace_mounted < 0) {
3054                 /* Nothing is mounted on the workspace yet, let's try to mount something now */
3055                 for (int try = 0;; try++) {
3056
3057                         if (try == 0) {
3058                                 /* Try "ramfs" first, since it's not swap backed */
3059                                 r = mount_nofollow_verbose(LOG_DEBUG, "ramfs", workspace, "ramfs", MS_NODEV|MS_NOEXEC|MS_NOSUID, "mode=0700");
3060                                 if (r >= 0) {
3061                                         workspace_mounted = true;
3062                                         break;
3063                                 }
3064
3065                         } else if (try == 1) {
3066                                 _cleanup_free_ char *opts = NULL;
3067
3068                                 if (asprintf(&opts, "mode=0700,nr_inodes=1024,size=%zu", (size_t) CREDENTIALS_TOTAL_SIZE_MAX) < 0)
3069                                         return -ENOMEM;
3070
3071                                 /* Fall back to "tmpfs" otherwise */
3072                                 r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", workspace, "tmpfs", MS_NODEV|MS_NOEXEC|MS_NOSUID, opts);
3073                                 if (r >= 0) {
3074                                         workspace_mounted = true;
3075                                         break;
3076                                 }
3077
3078                         } else {
3079                                 /* If that didn't work, try to make a bind mount from the final to the workspace, so that we can make it writable there. */
3080                                 r = mount_nofollow_verbose(LOG_DEBUG, final, workspace, NULL, MS_BIND|MS_REC, NULL);
3081                                 if (r < 0) {
3082                                         if (!ERRNO_IS_PRIVILEGE(r)) /* Propagate anything that isn't a permission problem */
3083                                                 return r;
3084
3085                                         if (must_mount) /* If we it's not OK to use the plain directory
3086                                                          * fallback, propagate all errors too */
3087                                                 return r;
3088
3089                                         /* If we lack privileges to bind mount stuff, then let's gracefully
3090                                          * proceed for compat with container envs, and just use the final dir
3091                                          * as is. */
3092
3093                                         workspace_mounted = false;
3094                                         break;
3095                                 }
3096
3097                                 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
3098                                 r = mount_nofollow_verbose(LOG_DEBUG, NULL, workspace, NULL, MS_BIND|MS_REMOUNT|MS_NODEV|MS_NOEXEC|MS_NOSUID, NULL);
3099                                 if (r < 0)
3100                                         return r;
3101
3102                                 workspace_mounted = true;
3103                                 break;
3104                         }
3105                 }
3106         }
3107
3108         assert(!must_mount || workspace_mounted > 0);
3109         where = workspace_mounted ? workspace : final;
3110
3111         (void) label_fix_full(AT_FDCWD, where, final, 0);
3112
3113         r = acquire_credentials(context, params, unit, where, uid, workspace_mounted);
3114         if (r < 0)
3115                 return r;
3116
3117         if (workspace_mounted) {
3118                 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
3119                 r = mount_nofollow_verbose(LOG_DEBUG, NULL, workspace, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NODEV|MS_NOEXEC|MS_NOSUID, NULL);
3120                 if (r < 0)
3121                         return r;
3122
3123                 /* And mount it to the final place, read-only */
3124                 if (final_mounted)
3125                         r = umount_verbose(LOG_DEBUG, workspace, MNT_DETACH|UMOUNT_NOFOLLOW);
3126                 else
3127                         r = mount_nofollow_verbose(LOG_DEBUG, workspace, final, NULL, MS_MOVE, NULL);
3128                 if (r < 0)
3129                         return r;
3130         } else {
3131                 _cleanup_free_ char *parent = NULL;
3132
3133                 /* If we do not have our own mount put used the plain directory fallback, then we need to
3134                  * open access to the top-level credential directory and the per-service directory now */
3135
3136                 r = path_extract_directory(final, &parent);
3137                 if (r < 0)
3138                         return r;
3139                 if (chmod(parent, 0755) < 0)
3140                         return -errno;
3141         }
3142
3143         return 0;
3144 }
3145
3146 static int setup_credentials(
3147                 const ExecContext *context,
3148                 const ExecParameters *params,
3149                 const char *unit,
3150                 uid_t uid) {
3151
3152         _cleanup_free_ char *p = NULL, *q = NULL;
3153         int r;
3154
3155         assert(context);
3156         assert(params);
3157
3158         if (!exec_context_has_credentials(context))
3159                 return 0;
3160
3161         if (!params->prefix[EXEC_DIRECTORY_RUNTIME])
3162                 return -EINVAL;
3163
3164         /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
3165          * and the subdir we mount over with a read-only file system readable by the service's user */
3166         q = path_join(params->prefix[EXEC_DIRECTORY_RUNTIME], "credentials");
3167         if (!q)
3168                 return -ENOMEM;
3169
3170         r = mkdir_label(q, 0755); /* top-level dir: world readable/searchable */
3171         if (r < 0 && r != -EEXIST)
3172                 return r;
3173
3174         p = path_join(q, unit);
3175         if (!p)
3176                 return -ENOMEM;
3177
3178         r = mkdir_label(p, 0700); /* per-unit dir: private to user */
3179         if (r < 0 && r != -EEXIST)
3180                 return r;
3181
3182         r = safe_fork("(sd-mkdcreds)", FORK_DEATHSIG|FORK_WAIT|FORK_NEW_MOUNTNS, NULL);
3183         if (r < 0) {
3184                 _cleanup_free_ char *t = NULL, *u = NULL;
3185
3186                 /* If this is not a privilege or support issue then propagate the error */
3187                 if (!ERRNO_IS_NOT_SUPPORTED(r) && !ERRNO_IS_PRIVILEGE(r))
3188                         return r;
3189
3190                 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
3191                  * it into place, so that users can't access half-initialized credential stores. */
3192                 t = path_join(params->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/temporary-credentials");
3193                 if (!t)
3194                         return -ENOMEM;
3195
3196                 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
3197                  * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
3198                  * after it is fully set up */
3199                 u = path_join(t, unit);
3200                 if (!u)
3201                         return -ENOMEM;
3202
3203                 FOREACH_STRING(i, t, u) {
3204                         r = mkdir_label(i, 0700);
3205                         if (r < 0 && r != -EEXIST)
3206                                 return r;
3207                 }
3208
3209                 r = setup_credentials_internal(
3210                                 context,
3211                                 params,
3212                                 unit,
3213                                 p,       /* final mount point */
3214                                 u,       /* temporary workspace to overmount */
3215                                 true,    /* reuse the workspace if it is already a mount */
3216                                 false,   /* it's OK to fall back to a plain directory if we can't mount anything */
3217                                 uid);
3218
3219                 (void) rmdir(u); /* remove the workspace again if we can. */
3220
3221                 if (r < 0)
3222                         return r;
3223
3224         } else if (r == 0) {
3225
3226                 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
3227                  * we can use the same directory for all cases, after turning off propagation. Question
3228                  * though is: where do we turn off propagation exactly, and where do we place the workspace
3229                  * directory? We need some place that is guaranteed to be a mount point in the host, and
3230                  * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
3231                  * since we ultimately want to move the resulting file system there, i.e. we need propagation
3232                  * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
3233                  * would be visible in the host mount table all the time, which we want to avoid. Hence, what
3234                  * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
3235                  * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
3236                  * propagation on the former, and then overmount the latter.
3237                  *
3238                  * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
3239                  * for this purpose, but there are few other candidates that work equally well for us, and
3240                  * given that the we do this in a privately namespaced short-lived single-threaded process
3241                  * that no one else sees this should be OK to do. */
3242
3243                 r = mount_nofollow_verbose(LOG_DEBUG, NULL, "/dev", NULL, MS_SLAVE|MS_REC, NULL); /* Turn off propagation from our namespace to host */
3244                 if (r < 0)
3245                         goto child_fail;
3246
3247                 r = setup_credentials_internal(
3248                                 context,
3249                                 params,
3250                                 unit,
3251                                 p,           /* final mount point */
3252                                 "/dev/shm",  /* temporary workspace to overmount */
3253                                 false,       /* do not reuse /dev/shm if it is already a mount, under no circumstances */
3254                                 true,        /* insist that something is mounted, do not allow fallback to plain directory */
3255                                 uid);
3256                 if (r < 0)
3257                         goto child_fail;
3258
3259                 _exit(EXIT_SUCCESS);
3260
3261         child_fail:
3262                 _exit(EXIT_FAILURE);
3263         }
3264
3265         return 0;
3266 }
3267
3268 #if ENABLE_SMACK
3269 static int setup_smack(
3270                 const Manager *manager,
3271                 const ExecContext *context,
3272                 int executable_fd) {
3273         int r;
3274
3275         assert(context);
3276         assert(executable_fd >= 0);
3277
3278         if (context->smack_process_label) {
3279                 r = mac_smack_apply_pid(0, context->smack_process_label);
3280                 if (r < 0)
3281                         return r;
3282         } else if (manager->default_smack_process_label) {
3283                 _cleanup_free_ char *exec_label = NULL;
3284
3285                 r = mac_smack_read_fd(executable_fd, SMACK_ATTR_EXEC, &exec_label);
3286                 if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r))
3287                         return r;
3288
3289                 r = mac_smack_apply_pid(0, exec_label ? : manager->default_smack_process_label);
3290                 if (r < 0)
3291                         return r;
3292         }
3293
3294         return 0;
3295 }
3296 #endif
3297
3298 static int compile_bind_mounts(
3299                 const ExecContext *context,
3300                 const ExecParameters *params,
3301                 BindMount **ret_bind_mounts,
3302                 size_t *ret_n_bind_mounts,
3303                 char ***ret_empty_directories) {
3304
3305         _cleanup_strv_free_ char **empty_directories = NULL;
3306         BindMount *bind_mounts;
3307         size_t n, h = 0;
3308         int r;
3309
3310         assert(context);
3311         assert(params);
3312         assert(ret_bind_mounts);
3313         assert(ret_n_bind_mounts);
3314         assert(ret_empty_directories);
3315
3316         n = context->n_bind_mounts;
3317         for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
3318                 if (!params->prefix[t])
3319                         continue;
3320
3321                 for (size_t i = 0; i < context->directories[t].n_items; i++)
3322                         n += !context->directories[t].items[i].only_create;
3323         }
3324
3325         if (n <= 0) {
3326                 *ret_bind_mounts = NULL;
3327                 *ret_n_bind_mounts = 0;
3328                 *ret_empty_directories = NULL;
3329                 return 0;
3330         }
3331
3332         bind_mounts = new(BindMount, n);
3333         if (!bind_mounts)
3334                 return -ENOMEM;
3335
3336         for (size_t i = 0; i < context->n_bind_mounts; i++) {
3337                 BindMount *item = context->bind_mounts + i;
3338                 char *s, *d;
3339
3340                 s = strdup(item->source);
3341                 if (!s) {
3342                         r = -ENOMEM;
3343                         goto finish;
3344                 }
3345
3346                 d = strdup(item->destination);
3347                 if (!d) {
3348                         free(s);
3349                         r = -ENOMEM;
3350                         goto finish;
3351                 }
3352
3353                 bind_mounts[h++] = (BindMount) {
3354                         .source = s,
3355                         .destination = d,
3356                         .read_only = item->read_only,
3357                         .recursive = item->recursive,
3358                         .ignore_enoent = item->ignore_enoent,
3359                 };
3360         }
3361
3362         for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
3363                 if (!params->prefix[t])
3364                         continue;
3365
3366                 if (context->directories[t].n_items == 0)
3367                         continue;
3368
3369                 if (exec_directory_is_private(context, t) &&
3370                     !exec_context_with_rootfs(context)) {
3371                         char *private_root;
3372
3373                         /* So this is for a dynamic user, and we need to make sure the process can access its own
3374                          * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3375                          * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3376
3377                         private_root = path_join(params->prefix[t], "private");
3378                         if (!private_root) {
3379                                 r = -ENOMEM;
3380                                 goto finish;
3381                         }
3382
3383                         r = strv_consume(&empty_directories, private_root);
3384                         if (r < 0)
3385                                 goto finish;
3386                 }
3387
3388                 for (size_t i = 0; i < context->directories[t].n_items; i++) {
3389                         char *s, *d;
3390
3391                         /* When one of the parent directories is in the list, we cannot create the symlink
3392                          * for the child directory. See also the comments in setup_exec_directory(). */
3393                         if (context->directories[t].items[i].only_create)
3394                                 continue;
3395
3396                         if (exec_directory_is_private(context, t))
3397                                 s = path_join(params->prefix[t], "private", context->directories[t].items[i].path);
3398                         else
3399                                 s = path_join(params->prefix[t], context->directories[t].items[i].path);
3400                         if (!s) {
3401                                 r = -ENOMEM;
3402                                 goto finish;
3403                         }
3404
3405                         if (exec_directory_is_private(context, t) &&
3406                             exec_context_with_rootfs(context))
3407                                 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3408                                  * directory is not created on the root directory. So, let's bind-mount the directory
3409                                  * on the 'non-private' place. */
3410                                 d = path_join(params->prefix[t], context->directories[t].items[i].path);
3411                         else
3412                                 d = strdup(s);
3413                         if (!d) {
3414                                 free(s);
3415                                 r = -ENOMEM;
3416                                 goto finish;
3417                         }
3418
3419                         bind_mounts[h++] = (BindMount) {
3420                                 .source = s,
3421                                 .destination = d,
3422                                 .read_only = false,
3423                                 .nosuid = context->dynamic_user, /* don't allow suid/sgid when DynamicUser= is on */
3424                                 .recursive = true,
3425                                 .ignore_enoent = false,
3426                         };
3427                 }
3428         }
3429
3430         assert(h == n);
3431
3432         *ret_bind_mounts = bind_mounts;
3433         *ret_n_bind_mounts = n;
3434         *ret_empty_directories = TAKE_PTR(empty_directories);
3435
3436         return (int) n;
3437
3438 finish:
3439         bind_mount_free_many(bind_mounts, h);
3440         return r;
3441 }
3442
3443 /* ret_symlinks will contain a list of pairs src:dest that describes
3444  * the symlinks to create later on. For example, the symlinks needed
3445  * to safely give private directories to DynamicUser=1 users. */
3446 static int compile_symlinks(
3447                 const ExecContext *context,
3448                 const ExecParameters *params,
3449                 char ***ret_symlinks) {
3450
3451         _cleanup_strv_free_ char **symlinks = NULL;
3452         int r;
3453
3454         assert(context);
3455         assert(params);
3456         assert(ret_symlinks);
3457
3458         for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
3459                 for (size_t i = 0; i < context->directories[dt].n_items; i++) {
3460                         _cleanup_free_ char *private_path = NULL, *path = NULL;
3461
3462                         STRV_FOREACH(symlink, context->directories[dt].items[i].symlinks) {
3463                                 _cleanup_free_ char *src_abs = NULL, *dst_abs = NULL;
3464
3465                                 src_abs = path_join(params->prefix[dt], context->directories[dt].items[i].path);
3466                                 dst_abs = path_join(params->prefix[dt], *symlink);
3467                                 if (!src_abs || !dst_abs)
3468                                         return -ENOMEM;
3469
3470                                 r = strv_consume_pair(&symlinks, TAKE_PTR(src_abs), TAKE_PTR(dst_abs));
3471                                 if (r < 0)
3472                                         return r;
3473                         }
3474
3475                         if (!exec_directory_is_private(context, dt) ||
3476                             exec_context_with_rootfs(context) ||
3477                             context->directories[dt].items[i].only_create)
3478                                 continue;
3479
3480                         private_path = path_join(params->prefix[dt], "private", context->directories[dt].items[i].path);
3481                         if (!private_path)
3482                                 return -ENOMEM;
3483
3484                         path = path_join(params->prefix[dt], context->directories[dt].items[i].path);
3485                         if (!path)
3486                                 return -ENOMEM;
3487
3488                         r = strv_consume_pair(&symlinks, TAKE_PTR(private_path), TAKE_PTR(path));
3489                         if (r < 0)
3490                                 return r;
3491                 }
3492         }
3493
3494         *ret_symlinks = TAKE_PTR(symlinks);
3495
3496         return 0;
3497 }
3498
3499 static bool insist_on_sandboxing(
3500                 const ExecContext *context,
3501                 const char *root_dir,
3502                 const char *root_image,
3503                 const BindMount *bind_mounts,
3504                 size_t n_bind_mounts) {
3505
3506         assert(context);
3507         assert(n_bind_mounts == 0 || bind_mounts);
3508
3509         /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3510          * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3511          * rearrange stuff in a way we cannot ignore gracefully. */
3512
3513         if (context->n_temporary_filesystems > 0)
3514                 return true;
3515
3516         if (root_dir || root_image)
3517                 return true;
3518
3519         if (context->n_mount_images > 0)
3520                 return true;
3521
3522         if (context->dynamic_user)
3523                 return true;
3524
3525         if (context->n_extension_images > 0 || !strv_isempty(context->extension_directories))
3526                 return true;
3527
3528         /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3529          * essential. */
3530         for (size_t i = 0; i < n_bind_mounts; i++)
3531                 if (!path_equal(bind_mounts[i].source, bind_mounts[i].destination))
3532                         return true;
3533
3534         if (context->log_namespace)
3535                 return true;
3536
3537         return false;
3538 }
3539
3540 static int apply_mount_namespace(
3541                 const Unit *u,
3542                 ExecCommandFlags command_flags,
3543                 const ExecContext *context,
3544                 const ExecParameters *params,
3545                 const ExecRuntime *runtime,
3546                 char **error_path) {
3547
3548         _cleanup_strv_free_ char **empty_directories = NULL, **symlinks = NULL;
3549         const char *tmp_dir = NULL, *var_tmp_dir = NULL;
3550         const char *root_dir = NULL, *root_image = NULL;
3551         _cleanup_free_ char *creds_path = NULL, *incoming_dir = NULL, *propagate_dir = NULL,
3552                         *extension_dir = NULL;
3553         NamespaceInfo ns_info;
3554         bool needs_sandboxing;
3555         BindMount *bind_mounts = NULL;
3556         size_t n_bind_mounts = 0;
3557         int r;
3558
3559         assert(context);
3560
3561         if (params->flags & EXEC_APPLY_CHROOT) {
3562                 root_image = context->root_image;
3563
3564                 if (!root_image)
3565                         root_dir = context->root_directory;
3566         }
3567
3568         r = compile_bind_mounts(context, params, &bind_mounts, &n_bind_mounts, &empty_directories);
3569         if (r < 0)
3570                 return r;
3571
3572         /* Symlinks for exec dirs are set up after other mounts, before they are made read-only. */
3573         r = compile_symlinks(context, params, &symlinks);
3574         if (r < 0)
3575                 goto finalize;
3576
3577         needs_sandboxing = (params->flags & EXEC_APPLY_SANDBOXING) && !(command_flags & EXEC_COMMAND_FULLY_PRIVILEGED);
3578         if (needs_sandboxing) {
3579                 /* The runtime struct only contains the parent of the private /tmp,
3580                  * which is non-accessible to world users. Inside of it there's a /tmp
3581                  * that is sticky, and that's the one we want to use here.
3582                  * This does not apply when we are using /run/systemd/empty as fallback. */
3583
3584                 if (context->private_tmp && runtime) {
3585                         if (streq_ptr(runtime->tmp_dir, RUN_SYSTEMD_EMPTY))
3586                                 tmp_dir = runtime->tmp_dir;
3587                         else if (runtime->tmp_dir)
3588                                 tmp_dir = strjoina(runtime->tmp_dir, "/tmp");
3589
3590                         if (streq_ptr(runtime->var_tmp_dir, RUN_SYSTEMD_EMPTY))
3591                                 var_tmp_dir = runtime->var_tmp_dir;
3592                         else if (runtime->var_tmp_dir)
3593                                 var_tmp_dir = strjoina(runtime->var_tmp_dir, "/tmp");
3594                 }
3595
3596                 ns_info = (NamespaceInfo) {
3597                         .ignore_protect_paths = false,
3598                         .private_dev = context->private_devices,
3599                         .protect_control_groups = context->protect_control_groups,
3600                         .protect_kernel_tunables = context->protect_kernel_tunables,
3601                         .protect_kernel_modules = context->protect_kernel_modules,
3602                         .protect_kernel_logs = context->protect_kernel_logs,
3603                         .protect_hostname = context->protect_hostname,
3604                         .mount_apivfs = exec_context_get_effective_mount_apivfs(context),
3605                         .protect_home = context->protect_home,
3606                         .protect_system = context->protect_system,
3607                         .protect_proc = context->protect_proc,
3608                         .proc_subset = context->proc_subset,
3609                         .private_ipc = exec_needs_ipc_namespace(context),
3610                         /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
3611                         .mount_nosuid = context->no_new_privileges && !mac_selinux_use(),
3612                 };
3613         } else if (!context->dynamic_user && root_dir)
3614                 /*
3615                  * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3616                  * sandbox info, otherwise enforce it, don't ignore protected paths and
3617                  * fail if we are enable to apply the sandbox inside the mount namespace.
3618                  */
3619                 ns_info = (NamespaceInfo) {
3620                         .ignore_protect_paths = true,
3621                 };
3622         else
3623                 ns_info = (NamespaceInfo) {};
3624
3625         if (context->mount_flags == MS_SHARED)
3626                 log_unit_debug(u, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3627
3628         if (exec_context_has_credentials(context) &&
3629             params->prefix[EXEC_DIRECTORY_RUNTIME] &&
3630             FLAGS_SET(params->flags, EXEC_WRITE_CREDENTIALS)) {
3631                 creds_path = path_join(params->prefix[EXEC_DIRECTORY_RUNTIME], "credentials", u->id);
3632                 if (!creds_path) {
3633                         r = -ENOMEM;
3634                         goto finalize;
3635                 }
3636         }
3637
3638         if (MANAGER_IS_SYSTEM(u->manager)) {
3639                 propagate_dir = path_join("/run/systemd/propagate/", u->id);
3640                 if (!propagate_dir) {
3641                         r = -ENOMEM;
3642                         goto finalize;
3643                 }
3644
3645                 incoming_dir = strdup("/run/systemd/incoming");
3646                 if (!incoming_dir) {
3647                         r = -ENOMEM;
3648                         goto finalize;
3649                 }
3650
3651                 extension_dir = strdup("/run/systemd/unit-extensions");
3652                 if (!extension_dir) {
3653                         r = -ENOMEM;
3654                         goto finalize;
3655                 }
3656         } else
3657                 if (asprintf(&extension_dir, "/run/user/" UID_FMT "/systemd/unit-extensions", geteuid()) < 0) {
3658                         r = -ENOMEM;
3659                         goto finalize;
3660                 }
3661
3662         r = setup_namespace(root_dir, root_image, context->root_image_options,
3663                             &ns_info, context->read_write_paths,
3664                             needs_sandboxing ? context->read_only_paths : NULL,
3665                             needs_sandboxing ? context->inaccessible_paths : NULL,
3666                             needs_sandboxing ? context->exec_paths : NULL,
3667                             needs_sandboxing ? context->no_exec_paths : NULL,
3668                             empty_directories,
3669                             symlinks,
3670                             bind_mounts,
3671                             n_bind_mounts,
3672                             context->temporary_filesystems,
3673                             context->n_temporary_filesystems,
3674                             context->mount_images,
3675                             context->n_mount_images,
3676                             tmp_dir,
3677                             var_tmp_dir,
3678                             creds_path,
3679                             context->log_namespace,
3680                             context->mount_flags,
3681                             context->root_hash, context->root_hash_size, context->root_hash_path,
3682                             context->root_hash_sig, context->root_hash_sig_size, context->root_hash_sig_path,
3683                             context->root_verity,
3684                             context->extension_images,
3685                             context->n_extension_images,
3686                             context->extension_directories,
3687                             propagate_dir,
3688                             incoming_dir,
3689                             extension_dir,
3690                             root_dir || root_image ? params->notify_socket : NULL,
3691                             error_path);
3692
3693         /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3694          * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3695          * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3696          * completely different execution environment. */
3697         if (r == -ENOANO) {
3698                 if (insist_on_sandboxing(
3699                                     context,
3700                                     root_dir, root_image,
3701                                     bind_mounts,
3702                                     n_bind_mounts)) {
3703                         log_unit_debug(u, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3704                                        "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3705                                        n_bind_mounts, context->n_temporary_filesystems, yes_no(root_dir), yes_no(root_image), yes_no(context->dynamic_user));
3706
3707                         r = -EOPNOTSUPP;
3708                 } else {
3709                         log_unit_debug(u, "Failed to set up namespace, assuming containerized execution and ignoring.");
3710                         r = 0;
3711                 }
3712         }
3713
3714 finalize:
3715         bind_mount_free_many(bind_mounts, n_bind_mounts);
3716         return r;
3717 }
3718
3719 static int apply_working_directory(
3720                 const ExecContext *context,
3721                 const ExecParameters *params,
3722                 const char *home,
3723                 int *exit_status) {
3724
3725         const char *d, *wd;
3726
3727         assert(context);
3728         assert(exit_status);
3729
3730         if (context->working_directory_home) {
3731
3732                 if (!home) {
3733                         *exit_status = EXIT_CHDIR;
3734                         return -ENXIO;
3735                 }
3736
3737                 wd = home;
3738
3739         } else
3740                 wd = empty_to_root(context->working_directory);
3741
3742         if (params->flags & EXEC_APPLY_CHROOT)
3743                 d = wd;
3744         else
3745                 d = prefix_roota(context->root_directory, wd);
3746
3747         if (chdir(d) < 0 && !context->working_directory_missing_ok) {
3748                 *exit_status = EXIT_CHDIR;
3749                 return -errno;
3750         }
3751
3752         return 0;
3753 }
3754
3755 static int apply_root_directory(
3756                 const ExecContext *context,
3757                 const ExecParameters *params,
3758                 const bool needs_mount_ns,
3759                 int *exit_status) {
3760
3761         assert(context);
3762         assert(exit_status);
3763
3764         if (params->flags & EXEC_APPLY_CHROOT)
3765                 if (!needs_mount_ns && context->root_directory)
3766                         if (chroot(context->root_directory) < 0) {
3767                                 *exit_status = EXIT_CHROOT;
3768                                 return -errno;
3769                         }
3770
3771         return 0;
3772 }
3773
3774 static int setup_keyring(
3775                 const Unit *u,
3776                 const ExecContext *context,
3777                 const ExecParameters *p,
3778                 uid_t uid, gid_t gid) {
3779
3780         key_serial_t keyring;
3781         int r = 0;
3782         uid_t saved_uid;
3783         gid_t saved_gid;
3784
3785         assert(u);
3786         assert(context);
3787         assert(p);
3788
3789         /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3790          * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3791          * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3792          * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3793          * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3794          * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3795
3796         if (context->keyring_mode == EXEC_KEYRING_INHERIT)
3797                 return 0;
3798
3799         /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3800          * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3801          * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3802          * & group is just as nasty as acquiring a reference to the user keyring. */
3803
3804         saved_uid = getuid();
3805         saved_gid = getgid();
3806
3807         if (gid_is_valid(gid) && gid != saved_gid) {
3808                 if (setregid(gid, -1) < 0)
3809                         return log_unit_error_errno(u, errno, "Failed to change GID for user keyring: %m");
3810         }
3811
3812         if (uid_is_valid(uid) && uid != saved_uid) {
3813                 if (setreuid(uid, -1) < 0) {
3814                         r = log_unit_error_errno(u, errno, "Failed to change UID for user keyring: %m");
3815                         goto out;
3816                 }
3817         }
3818
3819         keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
3820         if (keyring == -1) {
3821                 if (errno == ENOSYS)
3822                         log_unit_debug_errno(u, errno, "Kernel keyring not supported, ignoring.");
3823                 else if (ERRNO_IS_PRIVILEGE(errno))
3824                         log_unit_debug_errno(u, errno, "Kernel keyring access prohibited, ignoring.");
3825                 else if (errno == EDQUOT)
3826                         log_unit_debug_errno(u, errno, "Out of kernel keyrings to allocate, ignoring.");
3827                 else
3828                         r = log_unit_error_errno(u, errno, "Setting up kernel keyring failed: %m");
3829
3830                 goto out;
3831         }
3832
3833         /* When requested link the user keyring into the session keyring. */
3834         if (context->keyring_mode == EXEC_KEYRING_SHARED) {
3835
3836                 if (keyctl(KEYCTL_LINK,
3837                            KEY_SPEC_USER_KEYRING,
3838                            KEY_SPEC_SESSION_KEYRING, 0, 0) < 0) {
3839                         r = log_unit_error_errno(u, errno, "Failed to link user keyring into session keyring: %m");
3840                         goto out;
3841                 }
3842         }
3843
3844         /* Restore uid/gid back */
3845         if (uid_is_valid(uid) && uid != saved_uid) {
3846                 if (setreuid(saved_uid, -1) < 0) {
3847                         r = log_unit_error_errno(u, errno, "Failed to change UID back for user keyring: %m");
3848                         goto out;
3849                 }
3850         }
3851
3852         if (gid_is_valid(gid) && gid != saved_gid) {
3853                 if (setregid(saved_gid, -1) < 0)
3854                         return log_unit_error_errno(u, errno, "Failed to change GID back for user keyring: %m");
3855         }
3856
3857         /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3858         if (!sd_id128_is_null(u->invocation_id)) {
3859                 key_serial_t key;
3860
3861                 key = add_key("user", "invocation_id", &u->invocation_id, sizeof(u->invocation_id), KEY_SPEC_SESSION_KEYRING);
3862                 if (key == -1)
3863                         log_unit_debug_errno(u, errno, "Failed to add invocation ID to keyring, ignoring: %m");
3864                 else {
3865                         if (keyctl(KEYCTL_SETPERM, key,
3866                                    KEY_POS_VIEW|KEY_POS_READ|KEY_POS_SEARCH|
3867                                    KEY_USR_VIEW|KEY_USR_READ|KEY_USR_SEARCH, 0, 0) < 0)
3868                                 r = log_unit_error_errno(u, errno, "Failed to restrict invocation ID permission: %m");
3869                 }
3870         }
3871
3872 out:
3873         /* Revert back uid & gid for the last time, and exit */
3874         /* no extra logging, as only the first already reported error matters */
3875         if (getuid() != saved_uid)
3876                 (void) setreuid(saved_uid, -1);
3877
3878         if (getgid() != saved_gid)
3879                 (void) setregid(saved_gid, -1);
3880
3881         return r;
3882 }
3883
3884 static void append_socket_pair(int *array, size_t *n, const int pair[static 2]) {
3885         assert(array);
3886         assert(n);
3887         assert(pair);
3888
3889         if (pair[0] >= 0)
3890                 array[(*n)++] = pair[0];
3891         if (pair[1] >= 0)
3892                 array[(*n)++] = pair[1];
3893 }
3894
3895 static int close_remaining_fds(
3896                 const ExecParameters *params,
3897                 const ExecRuntime *runtime,
3898                 const DynamicCreds *dcreds,
3899                 int user_lookup_fd,
3900                 int socket_fd,
3901                 const int *fds, size_t n_fds) {
3902
3903         size_t n_dont_close = 0;
3904         int dont_close[n_fds + 12];
3905
3906         assert(params);
3907
3908         if (params->stdin_fd >= 0)
3909                 dont_close[n_dont_close++] = params->stdin_fd;
3910         if (params->stdout_fd >= 0)
3911                 dont_close[n_dont_close++] = params->stdout_fd;
3912         if (params->stderr_fd >= 0)
3913                 dont_close[n_dont_close++] = params->stderr_fd;
3914
3915         if (socket_fd >= 0)
3916                 dont_close[n_dont_close++] = socket_fd;
3917         if (n_fds > 0) {
3918                 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
3919                 n_dont_close += n_fds;
3920         }
3921
3922         if (runtime) {
3923                 append_socket_pair(dont_close, &n_dont_close, runtime->netns_storage_socket);
3924                 append_socket_pair(dont_close, &n_dont_close, runtime->ipcns_storage_socket);
3925         }
3926
3927         if (dcreds) {
3928                 if (dcreds->user)
3929                         append_socket_pair(dont_close, &n_dont_close, dcreds->user->storage_socket);
3930                 if (dcreds->group)
3931                         append_socket_pair(dont_close, &n_dont_close, dcreds->group->storage_socket);
3932         }
3933
3934         if (user_lookup_fd >= 0)
3935                 dont_close[n_dont_close++] = user_lookup_fd;
3936
3937         return close_all_fds(dont_close, n_dont_close);
3938 }
3939
3940 static int send_user_lookup(
3941                 Unit *unit,
3942                 int user_lookup_fd,
3943                 uid_t uid,
3944                 gid_t gid) {
3945
3946         assert(unit);
3947
3948         /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3949          * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3950          * specified. */
3951
3952         if (user_lookup_fd < 0)
3953                 return 0;
3954
3955         if (!uid_is_valid(uid) && !gid_is_valid(gid))
3956                 return 0;
3957
3958         if (writev(user_lookup_fd,
3959                (struct iovec[]) {
3960                            IOVEC_INIT(&uid, sizeof(uid)),
3961                            IOVEC_INIT(&gid, sizeof(gid)),
3962                            IOVEC_INIT_STRING(unit->id) }, 3) < 0)
3963                 return -errno;
3964
3965         return 0;
3966 }
3967
3968 static int acquire_home(const ExecContext *c, uid_t uid, const char** home, char **buf) {
3969         int r;
3970
3971         assert(c);
3972         assert(home);
3973         assert(buf);
3974
3975         /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3976
3977         if (*home)
3978                 return 0;
3979
3980         if (!c->working_directory_home)
3981                 return 0;
3982
3983         r = get_home_dir(buf);
3984         if (r < 0)
3985                 return r;
3986
3987         *home = *buf;
3988         return 1;
3989 }
3990
3991 static int compile_suggested_paths(const ExecContext *c, const ExecParameters *p, char ***ret) {
3992         _cleanup_strv_free_ char ** list = NULL;
3993         int r;
3994
3995         assert(c);
3996         assert(p);
3997         assert(ret);
3998
3999         assert(c->dynamic_user);
4000
4001         /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
4002          * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
4003          * directories. */
4004
4005         for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
4006                 if (t == EXEC_DIRECTORY_CONFIGURATION)
4007                         continue;
4008
4009                 if (!p->prefix[t])
4010                         continue;
4011
4012                 for (size_t i = 0; i < c->directories[t].n_items; i++) {
4013                         char *e;
4014
4015                         if (exec_directory_is_private(c, t))
4016                                 e = path_join(p->prefix[t], "private", c->directories[t].items[i].path);
4017                         else
4018                                 e = path_join(p->prefix[t], c->directories[t].items[i].path);
4019                         if (!e)
4020                                 return -ENOMEM;
4021
4022                         r = strv_consume(&list, e);
4023                         if (r < 0)
4024                                 return r;
4025                 }
4026         }
4027
4028         *ret = TAKE_PTR(list);
4029
4030         return 0;
4031 }
4032
4033 static int exec_parameters_get_cgroup_path(const ExecParameters *params, char **ret) {
4034         bool using_subcgroup;
4035         char *p;
4036
4037         assert(params);
4038         assert(ret);
4039
4040         if (!params->cgroup_path)
4041                 return -EINVAL;
4042
4043         /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
4044          * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
4045          * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
4046          * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
4047          * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
4048          * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
4049          * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
4050          * flag, which is only passed for the former statements, not for the latter. */
4051
4052         using_subcgroup = FLAGS_SET(params->flags, EXEC_CONTROL_CGROUP|EXEC_CGROUP_DELEGATE|EXEC_IS_CONTROL);
4053         if (using_subcgroup)
4054                 p = path_join(params->cgroup_path, ".control");
4055         else
4056                 p = strdup(params->cgroup_path);
4057         if (!p)
4058                 return -ENOMEM;
4059
4060         *ret = p;
4061         return using_subcgroup;
4062 }
4063
4064 static int exec_context_cpu_affinity_from_numa(const ExecContext *c, CPUSet *ret) {
4065         _cleanup_(cpu_set_reset) CPUSet s = {};
4066         int r;
4067
4068         assert(c);
4069         assert(ret);
4070
4071         if (!c->numa_policy.nodes.set) {
4072                 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
4073                 return 0;
4074         }
4075
4076         r = numa_to_cpu_set(&c->numa_policy, &s);
4077         if (r < 0)
4078                 return r;
4079
4080         cpu_set_reset(ret);
4081
4082         return cpu_set_add_all(ret, &s);
4083 }
4084
4085 bool exec_context_get_cpu_affinity_from_numa(const ExecContext *c) {
4086         assert(c);
4087
4088         return c->cpu_affinity_from_numa;
4089 }
4090
4091 static int add_shifted_fd(int *fds, size_t fds_size, size_t *n_fds, int fd, int *ret_fd) {
4092         int r;
4093
4094         assert(fds);
4095         assert(n_fds);
4096         assert(*n_fds < fds_size);
4097         assert(ret_fd);
4098
4099         if (fd < 0) {
4100                 *ret_fd = -EBADF;
4101                 return 0;
4102         }
4103
4104         if (fd < 3 + (int) *n_fds) {
4105                 /* Let's move the fd up, so that it's outside of the fd range we will use to store
4106                  * the fds we pass to the process (or which are closed only during execve). */
4107
4108                 r = fcntl(fd, F_DUPFD_CLOEXEC, 3 + (int) *n_fds);
4109                 if (r < 0)
4110                         return -errno;
4111
4112                 close_and_replace(fd, r);
4113         }
4114
4115         *ret_fd = fds[*n_fds] = fd;
4116         (*n_fds) ++;
4117         return 1;
4118 }
4119
4120 static int connect_unix_harder(Unit *u, const OpenFile *of, int ofd) {
4121         union sockaddr_union addr = {
4122                 .un.sun_family = AF_UNIX,
4123         };
4124         socklen_t sa_len;
4125         static const int socket_types[] = { SOCK_DGRAM, SOCK_STREAM, SOCK_SEQPACKET };
4126         int r;
4127
4128         assert(u);
4129         assert(of);
4130         assert(ofd >= 0);
4131
4132         r = sockaddr_un_set_path(&addr.un, FORMAT_PROC_FD_PATH(ofd));
4133         if (r < 0)
4134                 return log_unit_error_errno(u, r, "Failed to set sockaddr for %s: %m", of->path);
4135
4136         sa_len = r;
4137
4138         for (size_t i = 0; i < ELEMENTSOF(socket_types); i++) {
4139                 _cleanup_close_ int fd = -EBADF;
4140
4141                 fd = socket(AF_UNIX, socket_types[i] | SOCK_CLOEXEC, 0);
4142                 if (fd < 0)
4143                         return log_unit_error_errno(u, errno, "Failed to create socket for %s: %m", of->path);
4144
4145                 r = RET_NERRNO(connect(fd, &addr.sa, sa_len));
4146                 if (r == -EPROTOTYPE)
4147                         continue;
4148                 if (r < 0)
4149                         return log_unit_error_errno(u, r, "Failed to connect socket for %s: %m", of->path);
4150
4151                 return TAKE_FD(fd);
4152         }
4153
4154         return log_unit_error_errno(u, SYNTHETIC_ERRNO(EPROTOTYPE), "Failed to connect socket for \"%s\".", of->path);
4155 }
4156
4157 static int get_open_file_fd(Unit *u, const OpenFile *of) {
4158         struct stat st;
4159         _cleanup_close_ int fd = -EBADF, ofd = -EBADF;
4160
4161         assert(u);
4162         assert(of);
4163
4164         ofd = open(of->path, O_PATH | O_CLOEXEC);
4165         if (ofd < 0)
4166                 return log_error_errno(errno, "Could not open \"%s\": %m", of->path);
4167         if (fstat(ofd, &st) < 0)
4168                 return log_error_errno(errno, "Failed to stat %s: %m", of->path);
4169
4170         if (S_ISSOCK(st.st_mode)) {
4171                 fd = connect_unix_harder(u, of, ofd);
4172                 if (fd < 0)
4173                         return fd;
4174
4175                 if (FLAGS_SET(of->flags, OPENFILE_READ_ONLY) && shutdown(fd, SHUT_WR) < 0)
4176                         return log_error_errno(errno, "Failed to shutdown send for socket %s: %m", of->path);
4177
4178                 log_unit_debug(u, "socket %s opened (fd=%d)", of->path, fd);
4179         } else {
4180                 int flags = FLAGS_SET(of->flags, OPENFILE_READ_ONLY) ? O_RDONLY : O_RDWR;
4181                 if (FLAGS_SET(of->flags, OPENFILE_APPEND))
4182                         flags |= O_APPEND;
4183                 else if (FLAGS_SET(of->flags, OPENFILE_TRUNCATE))
4184                         flags |= O_TRUNC;
4185
4186                 fd = fd_reopen(ofd, flags | O_CLOEXEC);
4187                 if (fd < 0)
4188                         return log_unit_error_errno(u, fd, "Failed to open file %s: %m", of->path);
4189
4190                 log_unit_debug(u, "file %s opened (fd=%d)", of->path, fd);
4191         }
4192
4193         return TAKE_FD(fd);
4194 }
4195
4196 static int collect_open_file_fds(
4197                 Unit *u,
4198                 OpenFile* open_files,
4199                 int **fds,
4200                 char ***fdnames,
4201                 size_t *n_fds) {
4202         int r;
4203
4204         assert(u);
4205         assert(fds);
4206         assert(fdnames);
4207         assert(n_fds);
4208
4209         LIST_FOREACH(open_files, of, open_files) {
4210                 _cleanup_close_ int fd = -EBADF;
4211
4212                 fd = get_open_file_fd(u, of);
4213                 if (fd < 0) {
4214                         if (FLAGS_SET(of->flags, OPENFILE_GRACEFUL)) {
4215                                 log_unit_debug_errno(u, fd, "Failed to get OpenFile= file descriptor for %s, ignoring: %m", of->path);
4216                                 continue;
4217                         }
4218
4219                         return fd;
4220                 }
4221
4222                 if (!GREEDY_REALLOC(*fds, *n_fds + 1))
4223                         return -ENOMEM;
4224
4225                 r = strv_extend(fdnames, of->fdname);
4226                 if (r < 0)
4227                         return r;
4228
4229                 (*fds)[*n_fds] = TAKE_FD(fd);
4230
4231                 (*n_fds)++;
4232         }
4233
4234         return 0;
4235 }
4236
4237 static int exec_child(
4238                 Unit *unit,
4239                 const ExecCommand *command,
4240                 const ExecContext *context,
4241                 const ExecParameters *params,
4242                 ExecRuntime *runtime,
4243                 DynamicCreds *dcreds,
4244                 int socket_fd,
4245                 const int named_iofds[static 3],
4246                 int *params_fds,
4247                 size_t n_socket_fds,
4248                 size_t n_storage_fds,
4249                 char **files_env,
4250                 int user_lookup_fd,
4251                 int *exit_status) {
4252
4253         _cleanup_strv_free_ char **our_env = NULL, **pass_env = NULL, **joined_exec_search_path = NULL, **accum_env = NULL, **replaced_argv = NULL;
4254         int r, ngids = 0, exec_fd;
4255         _cleanup_free_ gid_t *supplementary_gids = NULL;
4256         const char *username = NULL, *groupname = NULL;
4257         _cleanup_free_ char *home_buffer = NULL;
4258         const char *home = NULL, *shell = NULL;
4259         char **final_argv = NULL;
4260         dev_t journal_stream_dev = 0;
4261         ino_t journal_stream_ino = 0;
4262         bool userns_set_up = false;
4263         bool needs_sandboxing,          /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
4264                 needs_setuid,           /* Do we need to do the actual setresuid()/setresgid() calls? */
4265                 needs_mount_namespace,  /* Do we need to set up a mount namespace for this kernel? */
4266                 needs_ambient_hack;     /* Do we need to apply the ambient capabilities hack? */
4267 #if HAVE_SELINUX
4268         _cleanup_free_ char *mac_selinux_context_net = NULL;
4269         bool use_selinux = false;
4270 #endif
4271 #if ENABLE_SMACK
4272         bool use_smack = false;
4273 #endif
4274 #if HAVE_APPARMOR
4275         bool use_apparmor = false;
4276 #endif
4277         uid_t saved_uid = getuid();
4278         gid_t saved_gid = getgid();
4279         uid_t uid = UID_INVALID;
4280         gid_t gid = GID_INVALID;
4281         size_t n_fds = n_socket_fds + n_storage_fds, /* fds to pass to the child */
4282                n_keep_fds; /* total number of fds not to close */
4283         int secure_bits;
4284         _cleanup_free_ gid_t *gids_after_pam = NULL;
4285         int ngids_after_pam = 0;
4286         _cleanup_free_ int *fds = NULL;
4287         _cleanup_strv_free_ char **fdnames = NULL;
4288
4289         assert(unit);
4290         assert(command);
4291         assert(context);
4292         assert(params);
4293         assert(exit_status);
4294
4295         /* Explicitly test for CVE-2021-4034 inspired invocations */
4296         assert(command->path);
4297         assert(!strv_isempty(command->argv));
4298
4299         rename_process_from_path(command->path);
4300
4301         /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main
4302          * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially,
4303          * both of which will be demoted to SIG_DFL. */
4304         (void) default_signals(SIGNALS_CRASH_HANDLER,
4305                                SIGNALS_IGNORE);
4306
4307         if (context->ignore_sigpipe)
4308                 (void) ignore_signals(SIGPIPE);
4309
4310         r = reset_signal_mask();
4311         if (r < 0) {
4312                 *exit_status = EXIT_SIGNAL_MASK;
4313                 return log_unit_error_errno(unit, r, "Failed to set process signal mask: %m");
4314         }
4315
4316         if (params->idle_pipe)
4317                 do_idle_pipe_dance(params->idle_pipe);
4318
4319         /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
4320          * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
4321          * any fds open we don't really want open during the transition. In order to make logging work, we switch the
4322          * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
4323
4324         log_forget_fds();
4325         log_set_open_when_needed(true);
4326
4327         /* In case anything used libc syslog(), close this here, too */
4328         closelog();
4329
4330         fds = newdup(int, params_fds, n_fds);
4331         if (!fds) {
4332                 *exit_status = EXIT_MEMORY;
4333                 return log_oom();
4334         }
4335
4336         fdnames = strv_copy((char**) params->fd_names);
4337         if (!fdnames) {
4338                 *exit_status = EXIT_MEMORY;
4339                 return log_oom();
4340         }
4341
4342         r = collect_open_file_fds(unit, params->open_files, &fds, &fdnames, &n_fds);
4343         if (r < 0) {
4344                 *exit_status = EXIT_FDS;
4345                 return log_unit_error_errno(unit, r, "Failed to get OpenFile= file descriptors: %m");
4346         }
4347
4348         int keep_fds[n_fds + 3];
4349         memcpy_safe(keep_fds, fds, n_fds * sizeof(int));
4350         n_keep_fds = n_fds;
4351
4352         r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, params->exec_fd, &exec_fd);
4353         if (r < 0) {
4354                 *exit_status = EXIT_FDS;
4355                 return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m");
4356         }
4357
4358 #if HAVE_LIBBPF
4359         if (unit->manager->restrict_fs) {
4360                 int bpf_map_fd = lsm_bpf_map_restrict_fs_fd(unit);
4361                 if (bpf_map_fd < 0) {
4362                         *exit_status = EXIT_FDS;
4363                         return log_unit_error_errno(unit, bpf_map_fd, "Failed to get restrict filesystems BPF map fd: %m");
4364                 }
4365
4366                 r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, bpf_map_fd, &bpf_map_fd);
4367                 if (r < 0) {
4368                         *exit_status = EXIT_FDS;
4369                         return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m");
4370                 }
4371         }
4372 #endif
4373
4374         r = close_remaining_fds(params, runtime, dcreds, user_lookup_fd, socket_fd, keep_fds, n_keep_fds);
4375         if (r < 0) {
4376                 *exit_status = EXIT_FDS;
4377                 return log_unit_error_errno(unit, r, "Failed to close unwanted file descriptors: %m");
4378         }
4379
4380         if (!context->same_pgrp &&
4381             setsid() < 0) {
4382                 *exit_status = EXIT_SETSID;
4383                 return log_unit_error_errno(unit, errno, "Failed to create new process session: %m");
4384         }
4385
4386         exec_context_tty_reset(context, params);
4387
4388         if (unit_shall_confirm_spawn(unit)) {
4389                 _cleanup_free_ char *cmdline = NULL;
4390
4391                 cmdline = quote_command_line(command->argv, SHELL_ESCAPE_EMPTY);
4392                 if (!cmdline) {
4393                         *exit_status = EXIT_MEMORY;
4394                         return log_oom();
4395                 }
4396
4397                 r = ask_for_confirmation(context, params->confirm_spawn, unit, cmdline);
4398                 if (r != CONFIRM_EXECUTE) {
4399                         if (r == CONFIRM_PRETEND_SUCCESS) {
4400                                 *exit_status = EXIT_SUCCESS;
4401                                 return 0;
4402                         }
4403                         *exit_status = EXIT_CONFIRM;
4404                         return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ECANCELED),
4405                                                     "Execution cancelled by the user");
4406                 }
4407         }
4408
4409         /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
4410          * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
4411          * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
4412          * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
4413          * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
4414         if (setenv("SYSTEMD_ACTIVATION_UNIT", unit->id, true) != 0 ||
4415             setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit->manager) ? "system" : "user", true) != 0) {
4416                 *exit_status = EXIT_MEMORY;
4417                 return log_unit_error_errno(unit, errno, "Failed to update environment: %m");
4418         }
4419
4420         if (context->dynamic_user && dcreds) {
4421                 _cleanup_strv_free_ char **suggested_paths = NULL;
4422
4423                 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
4424                  * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
4425                 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
4426                         *exit_status = EXIT_USER;
4427                         return log_unit_error_errno(unit, errno, "Failed to update environment: %m");
4428                 }
4429
4430                 r = compile_suggested_paths(context, params, &suggested_paths);
4431                 if (r < 0) {
4432                         *exit_status = EXIT_MEMORY;
4433                         return log_oom();
4434                 }
4435
4436                 r = dynamic_creds_realize(dcreds, suggested_paths, &uid, &gid);
4437                 if (r < 0) {
4438                         *exit_status = EXIT_USER;
4439                         if (r == -EILSEQ)
4440                                 return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP),
4441                                                             "Failed to update dynamic user credentials: User or group with specified name already exists.");
4442                         return log_unit_error_errno(unit, r, "Failed to update dynamic user credentials: %m");
4443                 }
4444
4445                 if (!uid_is_valid(uid)) {
4446                         *exit_status = EXIT_USER;
4447                         return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ESRCH), "UID validation failed for \""UID_FMT"\"", uid);
4448                 }
4449
4450                 if (!gid_is_valid(gid)) {
4451                         *exit_status = EXIT_USER;
4452                         return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ESRCH), "GID validation failed for \""GID_FMT"\"", gid);
4453                 }
4454
4455                 if (dcreds->user)
4456                         username = dcreds->user->name;
4457
4458         } else {
4459                 r = get_fixed_user(context, &username, &uid, &gid, &home, &shell);
4460                 if (r < 0) {
4461                         *exit_status = EXIT_USER;
4462                         return log_unit_error_errno(unit, r, "Failed to determine user credentials: %m");
4463                 }
4464
4465                 r = get_fixed_group(context, &groupname, &gid);
4466                 if (r < 0) {
4467                         *exit_status = EXIT_GROUP;
4468                         return log_unit_error_errno(unit, r, "Failed to determine group credentials: %m");
4469                 }
4470         }
4471
4472         /* Initialize user supplementary groups and get SupplementaryGroups= ones */
4473         r = get_supplementary_groups(context, username, groupname, gid,
4474                                      &supplementary_gids, &ngids);
4475         if (r < 0) {
4476                 *exit_status = EXIT_GROUP;
4477                 return log_unit_error_errno(unit, r, "Failed to determine supplementary groups: %m");
4478         }
4479
4480         r = send_user_lookup(unit, user_lookup_fd, uid, gid);
4481         if (r < 0) {
4482                 *exit_status = EXIT_USER;
4483                 return log_unit_error_errno(unit, r, "Failed to send user credentials to PID1: %m");
4484         }
4485
4486         user_lookup_fd = safe_close(user_lookup_fd);
4487
4488         r = acquire_home(context, uid, &home, &home_buffer);
4489         if (r < 0) {
4490                 *exit_status = EXIT_CHDIR;
4491                 return log_unit_error_errno(unit, r, "Failed to determine $HOME for user: %m");
4492         }
4493
4494         /* If a socket is connected to STDIN/STDOUT/STDERR, we
4495          * must sure to drop O_NONBLOCK */
4496         if (socket_fd >= 0)
4497                 (void) fd_nonblock(socket_fd, false);
4498
4499         /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
4500          * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
4501         if (params->cgroup_path) {
4502                 _cleanup_free_ char *p = NULL;
4503
4504                 r = exec_parameters_get_cgroup_path(params, &p);
4505                 if (r < 0) {
4506                         *exit_status = EXIT_CGROUP;
4507                         return log_unit_error_errno(unit, r, "Failed to acquire cgroup path: %m");
4508                 }
4509
4510                 r = cg_attach_everywhere(params->cgroup_supported, p, 0, NULL, NULL);
4511                 if (r == -EUCLEAN) {
4512                         *exit_status = EXIT_CGROUP;
4513                         return log_unit_error_errno(unit, r, "Failed to attach process to cgroup %s "
4514                                                     "because the cgroup or one of its parents or "
4515                                                     "siblings is in the threaded mode: %m", p);
4516                 }
4517                 if (r < 0) {
4518                         *exit_status = EXIT_CGROUP;
4519                         return log_unit_error_errno(unit, r, "Failed to attach to cgroup %s: %m", p);
4520                 }
4521         }
4522
4523         if (context->network_namespace_path && runtime && runtime->netns_storage_socket[0] >= 0) {
4524                 r = open_shareable_ns_path(runtime->netns_storage_socket, context->network_namespace_path, CLONE_NEWNET);
4525                 if (r < 0) {
4526                         *exit_status = EXIT_NETWORK;
4527                         return log_unit_error_errno(unit, r, "Failed to open network namespace path %s: %m", context->network_namespace_path);
4528                 }
4529         }
4530
4531         if (context->ipc_namespace_path && runtime && runtime->ipcns_storage_socket[0] >= 0) {
4532                 r = open_shareable_ns_path(runtime->ipcns_storage_socket, context->ipc_namespace_path, CLONE_NEWIPC);
4533                 if (r < 0) {
4534                         *exit_status = EXIT_NAMESPACE;
4535                         return log_unit_error_errno(unit, r, "Failed to open IPC namespace path %s: %m", context->ipc_namespace_path);
4536                 }
4537         }
4538
4539         r = setup_input(context, params, socket_fd, named_iofds);
4540         if (r < 0) {
4541                 *exit_status = EXIT_STDIN;
4542                 return log_unit_error_errno(unit, r, "Failed to set up standard input: %m");
4543         }
4544
4545         r = setup_output(unit, context, params, STDOUT_FILENO, socket_fd, named_iofds, basename(command->path), uid, gid, &journal_stream_dev, &journal_stream_ino);
4546         if (r < 0) {
4547                 *exit_status = EXIT_STDOUT;
4548                 return log_unit_error_errno(unit, r, "Failed to set up standard output: %m");
4549         }
4550
4551         r = setup_output(unit, context, params, STDERR_FILENO, socket_fd, named_iofds, basename(command->path), uid, gid, &journal_stream_dev, &journal_stream_ino);
4552         if (r < 0) {
4553                 *exit_status = EXIT_STDERR;
4554                 return log_unit_error_errno(unit, r, "Failed to set up standard error output: %m");
4555         }
4556
4557         if (context->oom_score_adjust_set) {
4558                 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
4559                  * prohibit write access to this file, and we shouldn't trip up over that. */
4560                 r = set_oom_score_adjust(context->oom_score_adjust);
4561                 if (ERRNO_IS_PRIVILEGE(r))
4562                         log_unit_debug_errno(unit, r, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
4563                 else if (r < 0) {
4564                         *exit_status = EXIT_OOM_ADJUST;
4565                         return log_unit_error_errno(unit, r, "Failed to adjust OOM setting: %m");
4566                 }
4567         }
4568
4569         if (context->coredump_filter_set) {
4570                 r = set_coredump_filter(context->coredump_filter);
4571                 if (ERRNO_IS_PRIVILEGE(r))
4572                         log_unit_debug_errno(unit, r, "Failed to adjust coredump_filter, ignoring: %m");
4573                 else if (r < 0)
4574                         return log_unit_error_errno(unit, r, "Failed to adjust coredump_filter: %m");
4575         }
4576
4577         if (context->nice_set) {
4578                 r = setpriority_closest(context->nice);
4579                 if (r < 0)
4580                         return log_unit_error_errno(unit, r, "Failed to set up process scheduling priority (nice level): %m");
4581         }
4582
4583         if (context->cpu_sched_set) {
4584                 struct sched_param param = {
4585                         .sched_priority = context->cpu_sched_priority,
4586                 };
4587
4588                 r = sched_setscheduler(0,
4589                                        context->cpu_sched_policy |
4590                                        (context->cpu_sched_reset_on_fork ?
4591                                         SCHED_RESET_ON_FORK : 0),
4592                                        &param);
4593                 if (r < 0) {
4594                         *exit_status = EXIT_SETSCHEDULER;
4595                         return log_unit_error_errno(unit, errno, "Failed to set up CPU scheduling: %m");
4596                 }
4597         }
4598
4599         if (context->cpu_affinity_from_numa || context->cpu_set.set) {
4600                 _cleanup_(cpu_set_reset) CPUSet converted_cpu_set = {};
4601                 const CPUSet *cpu_set;
4602
4603                 if (context->cpu_affinity_from_numa) {
4604                         r = exec_context_cpu_affinity_from_numa(context, &converted_cpu_set);
4605                         if (r < 0) {
4606                                 *exit_status = EXIT_CPUAFFINITY;
4607                                 return log_unit_error_errno(unit, r, "Failed to derive CPU affinity mask from NUMA mask: %m");
4608                         }
4609
4610                         cpu_set = &converted_cpu_set;
4611                 } else
4612                         cpu_set = &context->cpu_set;
4613
4614                 if (sched_setaffinity(0, cpu_set->allocated, cpu_set->set) < 0) {
4615                         *exit_status = EXIT_CPUAFFINITY;
4616                         return log_unit_error_errno(unit, errno, "Failed to set up CPU affinity: %m");
4617                 }
4618         }
4619
4620         if (mpol_is_valid(numa_policy_get_type(&context->numa_policy))) {
4621                 r = apply_numa_policy(&context->numa_policy);
4622                 if (r == -EOPNOTSUPP)
4623                         log_unit_debug_errno(unit, r, "NUMA support not available, ignoring.");
4624                 else if (r < 0) {
4625                         *exit_status = EXIT_NUMA_POLICY;
4626                         return log_unit_error_errno(unit, r, "Failed to set NUMA memory policy: %m");
4627                 }
4628         }
4629
4630         if (context->ioprio_set)
4631                 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
4632                         *exit_status = EXIT_IOPRIO;
4633                         return log_unit_error_errno(unit, errno, "Failed to set up IO scheduling priority: %m");
4634                 }
4635
4636         if (context->timer_slack_nsec != NSEC_INFINITY)
4637                 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
4638                         *exit_status = EXIT_TIMERSLACK;
4639                         return log_unit_error_errno(unit, errno, "Failed to set up timer slack: %m");
4640                 }
4641
4642         if (context->personality != PERSONALITY_INVALID) {
4643                 r = safe_personality(context->personality);
4644                 if (r < 0) {
4645                         *exit_status = EXIT_PERSONALITY;
4646                         return log_unit_error_errno(unit, r, "Failed to set up execution domain (personality): %m");
4647                 }
4648         }
4649
4650         if (context->utmp_id) {
4651                 const char *line = context->tty_path ?
4652                         (path_startswith(context->tty_path, "/dev/") ?: context->tty_path) :
4653                         NULL;
4654                 utmp_put_init_process(context->utmp_id, getpid_cached(), getsid(0),
4655                                       line,
4656                                       context->utmp_mode == EXEC_UTMP_INIT  ? INIT_PROCESS :
4657                                       context->utmp_mode == EXEC_UTMP_LOGIN ? LOGIN_PROCESS :
4658                                       USER_PROCESS,
4659                                       username);
4660         }
4661
4662         if (uid_is_valid(uid)) {
4663                 r = chown_terminal(STDIN_FILENO, uid);
4664                 if (r < 0) {
4665                         *exit_status = EXIT_STDIN;
4666                         return log_unit_error_errno(unit, r, "Failed to change ownership of terminal: %m");
4667                 }
4668         }
4669
4670         /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4671          * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4672          * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4673          * touch a single hierarchy too. */
4674         if (params->cgroup_path && context->user && (params->flags & EXEC_CGROUP_DELEGATE)) {
4675                 r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, uid, gid);
4676                 if (r < 0) {
4677                         *exit_status = EXIT_CGROUP;
4678                         return log_unit_error_errno(unit, r, "Failed to adjust control group access: %m");
4679                 }
4680         }
4681
4682         needs_mount_namespace = exec_needs_mount_namespace(context, params, runtime);
4683
4684         for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
4685                 r = setup_exec_directory(context, params, uid, gid, dt, needs_mount_namespace, exit_status);
4686                 if (r < 0)
4687                         return log_unit_error_errno(unit, r, "Failed to set up special execution directory in %s: %m", params->prefix[dt]);
4688         }
4689
4690         if (FLAGS_SET(params->flags, EXEC_WRITE_CREDENTIALS)) {
4691                 r = setup_credentials(context, params, unit->id, uid);
4692                 if (r < 0) {
4693                         *exit_status = EXIT_CREDENTIALS;
4694                         return log_unit_error_errno(unit, r, "Failed to set up credentials: %m");
4695                 }
4696         }
4697
4698         r = build_environment(
4699                         unit,
4700                         context,
4701                         params,
4702                         n_fds,
4703                         fdnames,
4704                         home,
4705                         username,
4706                         shell,
4707                         journal_stream_dev,
4708                         journal_stream_ino,
4709                         &our_env);
4710         if (r < 0) {
4711                 *exit_status = EXIT_MEMORY;
4712                 return log_oom();
4713         }
4714
4715         r = build_pass_environment(context, &pass_env);
4716         if (r < 0) {
4717                 *exit_status = EXIT_MEMORY;
4718                 return log_oom();
4719         }
4720
4721         /* The $PATH variable is set to the default path in params->environment. However, this is overridden
4722          * if user-specified fields have $PATH set. The intention is to also override $PATH if the unit does
4723          * not specify PATH but the unit has ExecSearchPath. */
4724         if (!strv_isempty(context->exec_search_path)) {
4725                 _cleanup_free_ char *joined = NULL;
4726
4727                 joined = strv_join(context->exec_search_path, ":");
4728                 if (!joined) {
4729                         *exit_status = EXIT_MEMORY;
4730                         return log_oom();
4731                 }
4732
4733                 r = strv_env_assign(&joined_exec_search_path, "PATH", joined);
4734                 if (r < 0) {
4735                         *exit_status = EXIT_MEMORY;
4736                         return log_oom();
4737                 }
4738         }
4739
4740         accum_env = strv_env_merge(params->environment,
4741                                    our_env,
4742                                    joined_exec_search_path,
4743                                    pass_env,
4744                                    context->environment,
4745                                    files_env);
4746         if (!accum_env) {
4747                 *exit_status = EXIT_MEMORY;
4748                 return log_oom();
4749         }
4750         accum_env = strv_env_clean(accum_env);
4751
4752         (void) umask(context->umask);
4753
4754         r = setup_keyring(unit, context, params, uid, gid);
4755         if (r < 0) {
4756                 *exit_status = EXIT_KEYRING;
4757                 return log_unit_error_errno(unit, r, "Failed to set up kernel keyring: %m");
4758         }
4759
4760         /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted
4761          * from it. */
4762         needs_sandboxing = (params->flags & EXEC_APPLY_SANDBOXING) && !(command->flags & EXEC_COMMAND_FULLY_PRIVILEGED);
4763
4764         /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked
4765          * for it, and the kernel doesn't actually support ambient caps. */
4766         needs_ambient_hack = (params->flags & EXEC_APPLY_SANDBOXING) && (command->flags & EXEC_COMMAND_AMBIENT_MAGIC) && !ambient_capabilities_supported();
4767
4768         /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly
4769          * excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not
4770          * desired. */
4771         if (needs_ambient_hack)
4772                 needs_setuid = false;
4773         else
4774                 needs_setuid = (params->flags & EXEC_APPLY_SANDBOXING) && !(command->flags & (EXEC_COMMAND_FULLY_PRIVILEGED|EXEC_COMMAND_NO_SETUID));
4775
4776         if (needs_sandboxing) {
4777                 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on
4778                  * /sys being present. The actual MAC context application will happen later, as late as
4779                  * possible, to avoid impacting our own code paths. */
4780
4781 #if HAVE_SELINUX
4782                 use_selinux = mac_selinux_use();
4783 #endif
4784 #if ENABLE_SMACK
4785                 use_smack = mac_smack_use();
4786 #endif
4787 #if HAVE_APPARMOR
4788                 use_apparmor = mac_apparmor_use();
4789 #endif
4790         }
4791
4792         if (needs_sandboxing) {
4793                 int which_failed;
4794
4795                 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4796                  * is set here. (See below.) */
4797
4798                 r = setrlimit_closest_all((const struct rlimit* const *) context->rlimit, &which_failed);
4799                 if (r < 0) {
4800                         *exit_status = EXIT_LIMITS;
4801                         return log_unit_error_errno(unit, r, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
4802                 }
4803         }
4804
4805         if (needs_setuid && context->pam_name && username) {
4806                 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4807                  * wins here. (See above.) */
4808
4809                 /* All fds passed in the fds array will be closed in the pam child process. */
4810                 r = setup_pam(context->pam_name, username, uid, gid, context->tty_path, &accum_env, fds, n_fds);
4811                 if (r < 0) {
4812                         *exit_status = EXIT_PAM;
4813                         return log_unit_error_errno(unit, r, "Failed to set up PAM session: %m");
4814                 }
4815
4816                 ngids_after_pam = getgroups_alloc(&gids_after_pam);
4817                 if (ngids_after_pam < 0) {
4818                         *exit_status = EXIT_MEMORY;
4819                         return log_unit_error_errno(unit, ngids_after_pam, "Failed to obtain groups after setting up PAM: %m");
4820                 }
4821         }
4822
4823         if (needs_sandboxing && context->private_users && have_effective_cap(CAP_SYS_ADMIN) <= 0) {
4824                 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4825                  * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4826                  * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4827
4828                 userns_set_up = true;
4829                 r = setup_private_users(saved_uid, saved_gid, uid, gid);
4830                 if (r < 0) {
4831                         *exit_status = EXIT_USER;
4832                         return log_unit_error_errno(unit, r, "Failed to set up user namespacing for unprivileged user: %m");
4833                 }
4834         }
4835
4836         if (exec_needs_network_namespace(context) && runtime && runtime->netns_storage_socket[0] >= 0) {
4837
4838                 if (ns_type_supported(NAMESPACE_NET)) {
4839                         r = setup_shareable_ns(runtime->netns_storage_socket, CLONE_NEWNET);
4840                         if (r == -EPERM)
4841                                 log_unit_warning_errno(unit, r,
4842                                                        "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4843                         else if (r < 0) {
4844                                 *exit_status = EXIT_NETWORK;
4845                                 return log_unit_error_errno(unit, r, "Failed to set up network namespacing: %m");
4846                         }
4847                 } else if (context->network_namespace_path) {
4848                         *exit_status = EXIT_NETWORK;
4849                         return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP),
4850                                                     "NetworkNamespacePath= is not supported, refusing.");
4851                 } else
4852                         log_unit_warning(unit, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4853         }
4854
4855         if (exec_needs_ipc_namespace(context) && runtime && runtime->ipcns_storage_socket[0] >= 0) {
4856
4857                 if (ns_type_supported(NAMESPACE_IPC)) {
4858                         r = setup_shareable_ns(runtime->ipcns_storage_socket, CLONE_NEWIPC);
4859                         if (r == -EPERM)
4860                                 log_unit_warning_errno(unit, r,
4861                                                        "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
4862                         else if (r < 0) {
4863                                 *exit_status = EXIT_NAMESPACE;
4864                                 return log_unit_error_errno(unit, r, "Failed to set up IPC namespacing: %m");
4865                         }
4866                 } else if (context->ipc_namespace_path) {
4867                         *exit_status = EXIT_NAMESPACE;
4868                         return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP),
4869                                                     "IPCNamespacePath= is not supported, refusing.");
4870                 } else
4871                         log_unit_warning(unit, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
4872         }
4873
4874         if (needs_mount_namespace) {
4875                 _cleanup_free_ char *error_path = NULL;
4876
4877                 r = apply_mount_namespace(unit, command->flags, context, params, runtime, &error_path);
4878                 if (r < 0) {
4879                         *exit_status = EXIT_NAMESPACE;
4880                         return log_unit_error_errno(unit, r, "Failed to set up mount namespacing%s%s: %m",
4881                                                     error_path ? ": " : "", strempty(error_path));
4882                 }
4883         }
4884
4885         if (needs_sandboxing) {
4886                 r = apply_protect_hostname(unit, context, exit_status);
4887                 if (r < 0)
4888                         return r;
4889         }
4890
4891         /* Drop groups as early as possible.
4892          * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4893          * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4894         if (needs_setuid) {
4895                 _cleanup_free_ gid_t *gids_to_enforce = NULL;
4896                 int ngids_to_enforce = 0;
4897
4898                 ngids_to_enforce = merge_gid_lists(supplementary_gids,
4899                                                    ngids,
4900                                                    gids_after_pam,
4901                                                    ngids_after_pam,
4902                                                    &gids_to_enforce);
4903                 if (ngids_to_enforce < 0) {
4904                         *exit_status = EXIT_MEMORY;
4905                         return log_unit_error_errno(unit,
4906                                                     ngids_to_enforce,
4907                                                     "Failed to merge group lists. Group membership might be incorrect: %m");
4908                 }
4909
4910                 r = enforce_groups(gid, gids_to_enforce, ngids_to_enforce);
4911                 if (r < 0) {
4912                         *exit_status = EXIT_GROUP;
4913                         return log_unit_error_errno(unit, r, "Changing group credentials failed: %m");
4914                 }
4915         }
4916
4917         /* If the user namespace was not set up above, try to do it now.
4918          * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4919          * restricted by rules pertaining to combining user namespaces with other namespaces (e.g. in the
4920          * case of mount namespaces being less privileged when the mount point list is copied from a
4921          * different user namespace). */
4922
4923         if (needs_sandboxing && context->private_users && !userns_set_up) {
4924                 r = setup_private_users(saved_uid, saved_gid, uid, gid);
4925                 if (r < 0) {
4926                         *exit_status = EXIT_USER;
4927                         return log_unit_error_errno(unit, r, "Failed to set up user namespacing: %m");
4928                 }
4929         }
4930
4931         /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4932          * shall execute. */
4933
4934         _cleanup_free_ char *executable = NULL;
4935         _cleanup_close_ int executable_fd = -EBADF;
4936         r = find_executable_full(command->path, /* root= */ NULL, context->exec_search_path, false, &executable, &executable_fd);
4937         if (r < 0) {
4938                 if (r != -ENOMEM && (command->flags & EXEC_COMMAND_IGNORE_FAILURE)) {
4939                         log_unit_struct_errno(unit, LOG_INFO, r,
4940                                               "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR,
4941                                               LOG_UNIT_INVOCATION_ID(unit),
4942                                               LOG_UNIT_MESSAGE(unit, "Executable %s missing, skipping: %m",
4943                                                                command->path),
4944                                               "EXECUTABLE=%s", command->path);
4945                         return 0;
4946                 }
4947
4948                 *exit_status = EXIT_EXEC;
4949
4950                 return log_unit_struct_errno(unit, LOG_INFO, r,
4951                                              "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR,
4952                                              LOG_UNIT_INVOCATION_ID(unit),
4953                                              LOG_UNIT_MESSAGE(unit, "Failed to locate executable %s: %m",
4954                                                               command->path),
4955                                              "EXECUTABLE=%s", command->path);
4956         }
4957
4958         r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, executable_fd, &executable_fd);
4959         if (r < 0) {
4960                 *exit_status = EXIT_FDS;
4961                 return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m");
4962         }
4963
4964 #if HAVE_SELINUX
4965         if (needs_sandboxing && use_selinux && params->selinux_context_net) {
4966                 int fd = -EBADF;
4967
4968                 if (socket_fd >= 0)
4969                         fd = socket_fd;
4970                 else if (params->n_socket_fds == 1)
4971                         /* If stdin is not connected to a socket but we are triggered by exactly one socket unit then we
4972                          * use context from that fd to compute the label. */
4973                         fd = params->fds[0];
4974
4975                 if (fd >= 0) {
4976                         r = mac_selinux_get_child_mls_label(fd, executable, context->selinux_context, &mac_selinux_context_net);
4977                         if (r < 0) {
4978                                 if (!context->selinux_context_ignore) {
4979                                         *exit_status = EXIT_SELINUX_CONTEXT;
4980                                         return log_unit_error_errno(unit, r, "Failed to determine SELinux context: %m");
4981                                 }
4982                                 log_unit_debug_errno(unit, r, "Failed to determine SELinux context, ignoring: %m");
4983                         }
4984                 }
4985         }
4986 #endif
4987
4988         /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4989          * more aggressive this time since socket_fd and the netns and ipcns fds we don't need anymore. We do keep the exec_fd
4990          * however if we have it as we want to keep it open until the final execve(). */
4991
4992         r = close_all_fds(keep_fds, n_keep_fds);
4993         if (r >= 0)
4994                 r = shift_fds(fds, n_fds);
4995         if (r >= 0)
4996                 r = flags_fds(fds, n_socket_fds, n_fds, context->non_blocking);
4997         if (r < 0) {
4998                 *exit_status = EXIT_FDS;
4999                 return log_unit_error_errno(unit, r, "Failed to adjust passed file descriptors: %m");
5000         }
5001
5002         /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
5003          * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
5004          * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
5005          * came this far. */
5006
5007         secure_bits = context->secure_bits;
5008
5009         if (needs_sandboxing) {
5010                 uint64_t bset;
5011
5012                 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
5013                  * requested. (Note this is placed after the general resource limit initialization, see
5014                  * above, in order to take precedence.) */
5015                 if (context->restrict_realtime && !context->rlimit[RLIMIT_RTPRIO]) {
5016                         if (setrlimit(RLIMIT_RTPRIO, &RLIMIT_MAKE_CONST(0)) < 0) {
5017                                 *exit_status = EXIT_LIMITS;
5018                                 return log_unit_error_errno(unit, errno, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
5019                         }
5020                 }
5021
5022 #if ENABLE_SMACK
5023                 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
5024                  * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
5025                 if (use_smack) {
5026                         r = setup_smack(unit->manager, context, executable_fd);
5027                         if (r < 0 && !context->smack_process_label_ignore) {
5028                                 *exit_status = EXIT_SMACK_PROCESS_LABEL;
5029                                 return log_unit_error_errno(unit, r, "Failed to set SMACK process label: %m");
5030                         }
5031                 }
5032 #endif
5033
5034                 bset = context->capability_bounding_set;
5035                 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
5036                  * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
5037                  * instead of us doing that */
5038                 if (needs_ambient_hack)
5039                         bset |= (UINT64_C(1) << CAP_SETPCAP) |
5040                                 (UINT64_C(1) << CAP_SETUID) |
5041                                 (UINT64_C(1) << CAP_SETGID);
5042
5043                 if (!cap_test_all(bset)) {
5044                         r = capability_bounding_set_drop(bset, false);
5045                         if (r < 0) {
5046                                 *exit_status = EXIT_CAPABILITIES;
5047                                 return log_unit_error_errno(unit, r, "Failed to drop capabilities: %m");
5048                         }
5049                 }
5050
5051                 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
5052                  * keep-caps set.
5053                  *
5054                  * To be able to raise the ambient capabilities after setresuid() they have to be added to
5055                  * the inherited set and keep caps has to be set (done in enforce_user()).  After setresuid()
5056                  * the ambient capabilities can be raised as they are present in the permitted and
5057                  * inhertiable set. However it is possible that someone wants to set ambient capabilities
5058                  * without changing the user, so we also set the ambient capabilities here.
5059                  *
5060                  * The requested ambient capabilities are raised in the inheritable set if the second
5061                  * argument is true. */
5062                 if (!needs_ambient_hack) {
5063                         r = capability_ambient_set_apply(context->capability_ambient_set, true);
5064                         if (r < 0) {
5065                                 *exit_status = EXIT_CAPABILITIES;
5066                                 return log_unit_error_errno(unit, r, "Failed to apply ambient capabilities (before UID change): %m");
5067                         }
5068                 }
5069         }
5070
5071         /* chroot to root directory first, before we lose the ability to chroot */
5072         r = apply_root_directory(context, params, needs_mount_namespace, exit_status);
5073         if (r < 0)
5074                 return log_unit_error_errno(unit, r, "Chrooting to the requested root directory failed: %m");
5075
5076         if (needs_setuid) {
5077                 if (uid_is_valid(uid)) {
5078                         r = enforce_user(context, uid);
5079                         if (r < 0) {
5080                                 *exit_status = EXIT_USER;
5081                                 return log_unit_error_errno(unit, r, "Failed to change UID to " UID_FMT ": %m", uid);
5082                         }
5083
5084                         if (!needs_ambient_hack &&
5085                             context->capability_ambient_set != 0) {
5086
5087                                 /* Raise the ambient capabilities after user change. */
5088                                 r = capability_ambient_set_apply(context->capability_ambient_set, false);
5089                                 if (r < 0) {
5090                                         *exit_status = EXIT_CAPABILITIES;
5091                                         return log_unit_error_errno(unit, r, "Failed to apply ambient capabilities (after UID change): %m");
5092                                 }
5093                         }
5094                 }
5095         }
5096
5097         /* Apply working directory here, because the working directory might be on NFS and only the user running
5098          * this service might have the correct privilege to change to the working directory */
5099         r = apply_working_directory(context, params, home, exit_status);
5100         if (r < 0)
5101                 return log_unit_error_errno(unit, r, "Changing to the requested working directory failed: %m");
5102
5103         if (needs_sandboxing) {
5104                 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
5105                  * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
5106                  * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
5107                  * are restricted. */
5108
5109 #if HAVE_SELINUX
5110                 if (use_selinux) {
5111                         char *exec_context = mac_selinux_context_net ?: context->selinux_context;
5112
5113                         if (exec_context) {
5114                                 r = setexeccon(exec_context);
5115                                 if (r < 0) {
5116                                         if (!context->selinux_context_ignore) {
5117                                                 *exit_status = EXIT_SELINUX_CONTEXT;
5118                                                 return log_unit_error_errno(unit, r, "Failed to change SELinux context to %s: %m", exec_context);
5119                                         }
5120                                         log_unit_debug_errno(unit, r, "Failed to change SELinux context to %s, ignoring: %m", exec_context);
5121                                 }
5122                         }
5123                 }
5124 #endif
5125
5126 #if HAVE_APPARMOR
5127                 if (use_apparmor && context->apparmor_profile) {
5128                         r = aa_change_onexec(context->apparmor_profile);
5129                         if (r < 0 && !context->apparmor_profile_ignore) {
5130                                 *exit_status = EXIT_APPARMOR_PROFILE;
5131                                 return log_unit_error_errno(unit, errno, "Failed to prepare AppArmor profile change to %s: %m", context->apparmor_profile);
5132                         }
5133                 }
5134 #endif
5135
5136                 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential
5137                  * EPERMs we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits
5138                  * requires CAP_SETPCAP. */
5139                 if (prctl(PR_GET_SECUREBITS) != secure_bits) {
5140                         /* CAP_SETPCAP is required to set securebits. This capability is raised into the
5141                          * effective set here.
5142                          *
5143                          * The effective set is overwritten during execve() with the following values:
5144                          *
5145                          * - ambient set (for non-root processes)
5146                          *
5147                          * - (inheritable | bounding) set for root processes)
5148                          *
5149                          * Hence there is no security impact to raise it in the effective set before execve
5150                          */
5151                         r = capability_gain_cap_setpcap(/* return_caps= */ NULL);
5152                         if (r < 0) {
5153                                 *exit_status = EXIT_CAPABILITIES;
5154                                 return log_unit_error_errno(unit, r, "Failed to gain CAP_SETPCAP for setting secure bits");
5155                         }
5156                         if (prctl(PR_SET_SECUREBITS, secure_bits) < 0) {
5157                                 *exit_status = EXIT_SECUREBITS;
5158                                 return log_unit_error_errno(unit, errno, "Failed to set process secure bits: %m");
5159                         }
5160                 }
5161
5162                 if (context_has_no_new_privileges(context))
5163                         if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
5164                                 *exit_status = EXIT_NO_NEW_PRIVILEGES;
5165                                 return log_unit_error_errno(unit, errno, "Failed to disable new privileges: %m");
5166                         }
5167
5168 #if HAVE_SECCOMP
5169                 r = apply_address_families(unit, context);
5170                 if (r < 0) {
5171                         *exit_status = EXIT_ADDRESS_FAMILIES;
5172                         return log_unit_error_errno(unit, r, "Failed to restrict address families: %m");
5173                 }
5174
5175                 r = apply_memory_deny_write_execute(unit, context);
5176                 if (r < 0) {
5177                         *exit_status = EXIT_SECCOMP;
5178                         return log_unit_error_errno(unit, r, "Failed to disable writing to executable memory: %m");
5179                 }
5180
5181                 r = apply_restrict_realtime(unit, context);
5182                 if (r < 0) {
5183                         *exit_status = EXIT_SECCOMP;
5184                         return log_unit_error_errno(unit, r, "Failed to apply realtime restrictions: %m");
5185                 }
5186
5187                 r = apply_restrict_suid_sgid(unit, context);
5188                 if (r < 0) {
5189                         *exit_status = EXIT_SECCOMP;
5190                         return log_unit_error_errno(unit, r, "Failed to apply SUID/SGID restrictions: %m");
5191                 }
5192
5193                 r = apply_restrict_namespaces(unit, context);
5194                 if (r < 0) {
5195                         *exit_status = EXIT_SECCOMP;
5196                         return log_unit_error_errno(unit, r, "Failed to apply namespace restrictions: %m");
5197                 }
5198
5199                 r = apply_protect_sysctl(unit, context);
5200                 if (r < 0) {
5201                         *exit_status = EXIT_SECCOMP;
5202                         return log_unit_error_errno(unit, r, "Failed to apply sysctl restrictions: %m");
5203                 }
5204
5205                 r = apply_protect_kernel_modules(unit, context);
5206                 if (r < 0) {
5207                         *exit_status = EXIT_SECCOMP;
5208                         return log_unit_error_errno(unit, r, "Failed to apply module loading restrictions: %m");
5209                 }
5210
5211                 r = apply_protect_kernel_logs(unit, context);
5212                 if (r < 0) {
5213                         *exit_status = EXIT_SECCOMP;
5214                         return log_unit_error_errno(unit, r, "Failed to apply kernel log restrictions: %m");
5215                 }
5216
5217                 r = apply_protect_clock(unit, context);
5218                 if (r < 0) {
5219                         *exit_status = EXIT_SECCOMP;
5220                         return log_unit_error_errno(unit, r, "Failed to apply clock restrictions: %m");
5221                 }
5222
5223                 r = apply_private_devices(unit, context);
5224                 if (r < 0) {
5225                         *exit_status = EXIT_SECCOMP;
5226                         return log_unit_error_errno(unit, r, "Failed to set up private devices: %m");
5227                 }
5228
5229                 r = apply_syscall_archs(unit, context);
5230                 if (r < 0) {
5231                         *exit_status = EXIT_SECCOMP;
5232                         return log_unit_error_errno(unit, r, "Failed to apply syscall architecture restrictions: %m");
5233                 }
5234
5235                 r = apply_lock_personality(unit, context);
5236                 if (r < 0) {
5237                         *exit_status = EXIT_SECCOMP;
5238                         return log_unit_error_errno(unit, r, "Failed to lock personalities: %m");
5239                 }
5240
5241                 r = apply_syscall_log(unit, context);
5242                 if (r < 0) {
5243                         *exit_status = EXIT_SECCOMP;
5244                         return log_unit_error_errno(unit, r, "Failed to apply system call log filters: %m");
5245                 }
5246
5247                 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
5248                  * by the filter as little as possible. */
5249                 r = apply_syscall_filter(unit, context, needs_ambient_hack);
5250                 if (r < 0) {
5251                         *exit_status = EXIT_SECCOMP;
5252                         return log_unit_error_errno(unit, r, "Failed to apply system call filters: %m");
5253                 }
5254 #endif
5255
5256 #if HAVE_LIBBPF
5257                 r = apply_restrict_filesystems(unit, context);
5258                 if (r < 0) {
5259                         *exit_status = EXIT_BPF;
5260                         return log_unit_error_errno(unit, r, "Failed to restrict filesystems: %m");
5261                 }
5262 #endif
5263
5264         }
5265
5266         if (!strv_isempty(context->unset_environment)) {
5267                 char **ee = NULL;
5268
5269                 ee = strv_env_delete(accum_env, 1, context->unset_environment);
5270                 if (!ee) {
5271                         *exit_status = EXIT_MEMORY;
5272                         return log_oom();
5273                 }
5274
5275                 strv_free_and_replace(accum_env, ee);
5276         }
5277
5278         if (!FLAGS_SET(command->flags, EXEC_COMMAND_NO_ENV_EXPAND)) {
5279                 replaced_argv = replace_env_argv(command->argv, accum_env);
5280                 if (!replaced_argv) {
5281                         *exit_status = EXIT_MEMORY;
5282                         return log_oom();
5283                 }
5284                 final_argv = replaced_argv;
5285         } else
5286                 final_argv = command->argv;
5287
5288         if (DEBUG_LOGGING) {
5289                 _cleanup_free_ char *line = NULL;
5290
5291                 line = quote_command_line(final_argv, SHELL_ESCAPE_EMPTY);
5292                 if (!line) {
5293                         *exit_status = EXIT_MEMORY;
5294                         return log_oom();
5295                 }
5296
5297                 log_unit_struct(unit, LOG_DEBUG,
5298                                 "EXECUTABLE=%s", executable,
5299                                 LOG_UNIT_MESSAGE(unit, "Executing: %s", line));
5300         }
5301
5302         if (exec_fd >= 0) {
5303                 uint8_t hot = 1;
5304
5305                 /* We have finished with all our initializations. Let's now let the manager know that. From this point
5306                  * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
5307
5308                 if (write(exec_fd, &hot, sizeof(hot)) < 0) {
5309                         *exit_status = EXIT_EXEC;
5310                         return log_unit_error_errno(unit, errno, "Failed to enable exec_fd: %m");
5311                 }
5312         }
5313
5314         r = fexecve_or_execve(executable_fd, executable, final_argv, accum_env);
5315
5316         if (exec_fd >= 0) {
5317                 uint8_t hot = 0;
5318
5319                 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
5320                  * that POLLHUP on it no longer means execve() succeeded. */
5321
5322                 if (write(exec_fd, &hot, sizeof(hot)) < 0) {
5323                         *exit_status = EXIT_EXEC;
5324                         return log_unit_error_errno(unit, errno, "Failed to disable exec_fd: %m");
5325                 }
5326         }
5327
5328         *exit_status = EXIT_EXEC;
5329         return log_unit_error_errno(unit, r, "Failed to execute %s: %m", executable);
5330 }
5331
5332 static int exec_context_load_environment(const Unit *unit, const ExecContext *c, char ***l);
5333 static int exec_context_named_iofds(const ExecContext *c, const ExecParameters *p, int named_iofds[static 3]);
5334
5335 int exec_spawn(Unit *unit,
5336                ExecCommand *command,
5337                const ExecContext *context,
5338                const ExecParameters *params,
5339                ExecRuntime *runtime,
5340                DynamicCreds *dcreds,
5341                pid_t *ret) {
5342
5343         int socket_fd, r, named_iofds[3] = { -1, -1, -1 }, *fds = NULL;
5344         _cleanup_free_ char *subcgroup_path = NULL;
5345         _cleanup_strv_free_ char **files_env = NULL;
5346         size_t n_storage_fds = 0, n_socket_fds = 0;
5347         _cleanup_free_ char *line = NULL;
5348         pid_t pid;
5349
5350         assert(unit);
5351         assert(command);
5352         assert(context);
5353         assert(ret);
5354         assert(params);
5355         assert(params->fds || (params->n_socket_fds + params->n_storage_fds <= 0));
5356
5357         if (context->std_input == EXEC_INPUT_SOCKET ||
5358             context->std_output == EXEC_OUTPUT_SOCKET ||
5359             context->std_error == EXEC_OUTPUT_SOCKET) {
5360
5361                 if (params->n_socket_fds > 1)
5362                         return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EINVAL), "Got more than one socket.");
5363
5364                 if (params->n_socket_fds == 0)
5365                         return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EINVAL), "Got no socket.");
5366
5367                 socket_fd = params->fds[0];
5368         } else {
5369                 socket_fd = -EBADF;
5370                 fds = params->fds;
5371                 n_socket_fds = params->n_socket_fds;
5372                 n_storage_fds = params->n_storage_fds;
5373         }
5374
5375         r = exec_context_named_iofds(context, params, named_iofds);
5376         if (r < 0)
5377                 return log_unit_error_errno(unit, r, "Failed to load a named file descriptor: %m");
5378
5379         r = exec_context_load_environment(unit, context, &files_env);
5380         if (r < 0)
5381                 return log_unit_error_errno(unit, r, "Failed to load environment files: %m");
5382
5383         line = quote_command_line(command->argv, SHELL_ESCAPE_EMPTY);
5384         if (!line)
5385                 return log_oom();
5386
5387         /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
5388            and, until the next SELinux policy changes, we save further reloads in future children. */
5389         mac_selinux_maybe_reload();
5390
5391         log_unit_struct(unit, LOG_DEBUG,
5392                         LOG_UNIT_MESSAGE(unit, "About to execute %s", line),
5393                         "EXECUTABLE=%s", command->path, /* We won't know the real executable path until we create
5394                                                            the mount namespace in the child, but we want to log
5395                                                            from the parent, so we need to use the (possibly
5396                                                            inaccurate) path here. */
5397                         LOG_UNIT_INVOCATION_ID(unit));
5398
5399         if (params->cgroup_path) {
5400                 r = exec_parameters_get_cgroup_path(params, &subcgroup_path);
5401                 if (r < 0)
5402                         return log_unit_error_errno(unit, r, "Failed to acquire subcgroup path: %m");
5403                 if (r > 0) { /* We are using a child cgroup */
5404                         r = cg_create(SYSTEMD_CGROUP_CONTROLLER, subcgroup_path);
5405                         if (r < 0)
5406                                 return log_unit_error_errno(unit, r, "Failed to create control group '%s': %m", subcgroup_path);
5407
5408                         /* Normally we would not propagate the xattrs to children but since we created this
5409                          * sub-cgroup internally we should do it. */
5410                         cgroup_oomd_xattr_apply(unit, subcgroup_path);
5411                         cgroup_log_xattr_apply(unit, subcgroup_path);
5412                 }
5413         }
5414
5415         pid = fork();
5416         if (pid < 0)
5417                 return log_unit_error_errno(unit, errno, "Failed to fork: %m");
5418
5419         if (pid == 0) {
5420                 int exit_status = EXIT_SUCCESS;
5421
5422                 r = exec_child(unit,
5423                                command,
5424                                context,
5425                                params,
5426                                runtime,
5427                                dcreds,
5428                                socket_fd,
5429                                named_iofds,
5430                                fds,
5431                                n_socket_fds,
5432                                n_storage_fds,
5433                                files_env,
5434                                unit->manager->user_lookup_fds[1],
5435                                &exit_status);
5436
5437                 if (r < 0) {
5438                         const char *status =
5439                                 exit_status_to_string(exit_status,
5440                                                       EXIT_STATUS_LIBC | EXIT_STATUS_SYSTEMD);
5441
5442                         log_unit_struct_errno(unit, LOG_ERR, r,
5443                                               "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR,
5444                                               LOG_UNIT_INVOCATION_ID(unit),
5445                                               LOG_UNIT_MESSAGE(unit, "Failed at step %s spawning %s: %m",
5446                                                                status, command->path),
5447                                               "EXECUTABLE=%s", command->path);
5448                 }
5449
5450                 _exit(exit_status);
5451         }
5452
5453         log_unit_debug(unit, "Forked %s as "PID_FMT, command->path, pid);
5454
5455         /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
5456          * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
5457          * process will be killed too). */
5458         if (subcgroup_path)
5459                 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER, subcgroup_path, pid);
5460
5461         exec_status_start(&command->exec_status, pid);
5462
5463         *ret = pid;
5464         return 0;
5465 }
5466
5467 void exec_context_init(ExecContext *c) {
5468         assert(c);
5469
5470         c->umask = 0022;
5471         c->ioprio = IOPRIO_DEFAULT_CLASS_AND_PRIO;
5472         c->cpu_sched_policy = SCHED_OTHER;
5473         c->syslog_priority = LOG_DAEMON|LOG_INFO;
5474         c->syslog_level_prefix = true;
5475         c->ignore_sigpipe = true;
5476         c->timer_slack_nsec = NSEC_INFINITY;
5477         c->personality = PERSONALITY_INVALID;
5478         for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++)
5479                 c->directories[t].mode = 0755;
5480         c->timeout_clean_usec = USEC_INFINITY;
5481         c->capability_bounding_set = CAP_MASK_UNSET;
5482         assert_cc(NAMESPACE_FLAGS_INITIAL != NAMESPACE_FLAGS_ALL);
5483         c->restrict_namespaces = NAMESPACE_FLAGS_INITIAL;
5484         c->log_level_max = -1;
5485 #if HAVE_SECCOMP
5486         c->syscall_errno = SECCOMP_ERROR_NUMBER_KILL;
5487 #endif
5488         c->tty_rows = UINT_MAX;
5489         c->tty_cols = UINT_MAX;
5490         numa_policy_reset(&c->numa_policy);
5491 }
5492
5493 void exec_context_done(ExecContext *c) {
5494         assert(c);
5495
5496         c->environment = strv_free(c->environment);
5497         c->environment_files = strv_free(c->environment_files);
5498         c->pass_environment = strv_free(c->pass_environment);
5499         c->unset_environment = strv_free(c->unset_environment);
5500
5501         rlimit_free_all(c->rlimit);
5502
5503         for (size_t l = 0; l < 3; l++) {
5504                 c->stdio_fdname[l] = mfree(c->stdio_fdname[l]);
5505                 c->stdio_file[l] = mfree(c->stdio_file[l]);
5506         }
5507
5508         c->working_directory = mfree(c->working_directory);
5509         c->root_directory = mfree(c->root_directory);
5510         c->root_image = mfree(c->root_image);
5511         c->root_image_options = mount_options_free_all(c->root_image_options);
5512         c->root_hash = mfree(c->root_hash);
5513         c->root_hash_size = 0;
5514         c->root_hash_path = mfree(c->root_hash_path);
5515         c->root_hash_sig = mfree(c->root_hash_sig);
5516         c->root_hash_sig_size = 0;
5517         c->root_hash_sig_path = mfree(c->root_hash_sig_path);
5518         c->root_verity = mfree(c->root_verity);
5519         c->extension_images = mount_image_free_many(c->extension_images, &c->n_extension_images);
5520         c->extension_directories = strv_free(c->extension_directories);
5521         c->tty_path = mfree(c->tty_path);
5522         c->syslog_identifier = mfree(c->syslog_identifier);
5523         c->user = mfree(c->user);
5524         c->group = mfree(c->group);
5525
5526         c->supplementary_groups = strv_free(c->supplementary_groups);
5527
5528         c->pam_name = mfree(c->pam_name);
5529
5530         c->read_only_paths = strv_free(c->read_only_paths);
5531         c->read_write_paths = strv_free(c->read_write_paths);
5532         c->inaccessible_paths = strv_free(c->inaccessible_paths);
5533         c->exec_paths = strv_free(c->exec_paths);
5534         c->no_exec_paths = strv_free(c->no_exec_paths);
5535         c->exec_search_path = strv_free(c->exec_search_path);
5536
5537         bind_mount_free_many(c->bind_mounts, c->n_bind_mounts);
5538         c->bind_mounts = NULL;
5539         c->n_bind_mounts = 0;
5540         temporary_filesystem_free_many(c->temporary_filesystems, c->n_temporary_filesystems);
5541         c->temporary_filesystems = NULL;
5542         c->n_temporary_filesystems = 0;
5543         c->mount_images = mount_image_free_many(c->mount_images, &c->n_mount_images);
5544
5545         cpu_set_reset(&c->cpu_set);
5546         numa_policy_reset(&c->numa_policy);
5547
5548         c->utmp_id = mfree(c->utmp_id);
5549         c->selinux_context = mfree(c->selinux_context);
5550         c->apparmor_profile = mfree(c->apparmor_profile);
5551         c->smack_process_label = mfree(c->smack_process_label);
5552
5553         c->restrict_filesystems = set_free(c->restrict_filesystems);
5554
5555         c->syscall_filter = hashmap_free(c->syscall_filter);
5556         c->syscall_archs = set_free(c->syscall_archs);
5557         c->address_families = set_free(c->address_families);
5558
5559         for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++)
5560                 exec_directory_done(&c->directories[t]);
5561
5562         c->log_level_max = -1;
5563
5564         exec_context_free_log_extra_fields(c);
5565         c->log_filter_allowed_patterns = set_free(c->log_filter_allowed_patterns);
5566         c->log_filter_denied_patterns = set_free(c->log_filter_denied_patterns);
5567
5568         c->log_ratelimit_interval_usec = 0;
5569         c->log_ratelimit_burst = 0;
5570
5571         c->stdin_data = mfree(c->stdin_data);
5572         c->stdin_data_size = 0;
5573
5574         c->network_namespace_path = mfree(c->network_namespace_path);
5575         c->ipc_namespace_path = mfree(c->ipc_namespace_path);
5576
5577         c->log_namespace = mfree(c->log_namespace);
5578
5579         c->load_credentials = hashmap_free(c->load_credentials);
5580         c->set_credentials = hashmap_free(c->set_credentials);
5581 }
5582
5583 int exec_context_destroy_runtime_directory(const ExecContext *c, const char *runtime_prefix) {
5584         assert(c);
5585
5586         if (!runtime_prefix)
5587                 return 0;
5588
5589         for (size_t i = 0; i < c->directories[EXEC_DIRECTORY_RUNTIME].n_items; i++) {
5590                 _cleanup_free_ char *p = NULL;
5591
5592                 if (exec_directory_is_private(c, EXEC_DIRECTORY_RUNTIME))
5593                         p = path_join(runtime_prefix, "private", c->directories[EXEC_DIRECTORY_RUNTIME].items[i].path);
5594                 else
5595                         p = path_join(runtime_prefix, c->directories[EXEC_DIRECTORY_RUNTIME].items[i].path);
5596                 if (!p)
5597                         return -ENOMEM;
5598
5599                 /* We execute this synchronously, since we need to be sure this is gone when we start the
5600                  * service next. */
5601                 (void) rm_rf(p, REMOVE_ROOT);
5602
5603                 STRV_FOREACH(symlink, c->directories[EXEC_DIRECTORY_RUNTIME].items[i].symlinks) {
5604                         _cleanup_free_ char *symlink_abs = NULL;
5605
5606                         if (exec_directory_is_private(c, EXEC_DIRECTORY_RUNTIME))
5607                                 symlink_abs = path_join(runtime_prefix, "private", *symlink);
5608                         else
5609                                 symlink_abs = path_join(runtime_prefix, *symlink);
5610                         if (!symlink_abs)
5611                                 return -ENOMEM;
5612
5613                         (void) unlink(symlink_abs);
5614                 }
5615         }
5616
5617         return 0;
5618 }
5619
5620 int exec_context_destroy_credentials(const ExecContext *c, const char *runtime_prefix, const char *unit) {
5621         _cleanup_free_ char *p = NULL;
5622
5623         assert(c);
5624
5625         if (!runtime_prefix || !unit)
5626                 return 0;
5627
5628         p = path_join(runtime_prefix, "credentials", unit);
5629         if (!p)
5630                 return -ENOMEM;
5631
5632         /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
5633          * unmount it, and afterwards remove the mount point */
5634         (void) umount2(p, MNT_DETACH|UMOUNT_NOFOLLOW);
5635         (void) rm_rf(p, REMOVE_ROOT|REMOVE_CHMOD);
5636
5637         return 0;
5638 }
5639
5640 int exec_context_destroy_mount_ns_dir(Unit *u) {
5641         _cleanup_free_ char *p = NULL;
5642
5643         if (!u || !MANAGER_IS_SYSTEM(u->manager))
5644                 return 0;
5645
5646         p = path_join("/run/systemd/propagate/", u->id);
5647         if (!p)
5648                 return -ENOMEM;
5649
5650         /* This is only filled transiently (see mount_in_namespace()), should be empty or even non-existent*/
5651         if (rmdir(p) < 0 && errno != ENOENT)
5652                 log_unit_debug_errno(u, errno, "Unable to remove propagation dir '%s', ignoring: %m", p);
5653
5654         return 0;
5655 }
5656
5657 static void exec_command_done(ExecCommand *c) {
5658         assert(c);
5659
5660         c->path = mfree(c->path);
5661         c->argv = strv_free(c->argv);
5662 }
5663
5664 void exec_command_done_array(ExecCommand *c, size_t n) {
5665         for (size_t i = 0; i < n; i++)
5666                 exec_command_done(c+i);
5667 }
5668
5669 ExecCommand* exec_command_free_list(ExecCommand *c) {
5670         ExecCommand *i;
5671
5672         while ((i = c)) {
5673                 LIST_REMOVE(command, c, i);
5674                 exec_command_done(i);
5675                 free(i);
5676         }
5677
5678         return NULL;
5679 }
5680
5681 void exec_command_free_array(ExecCommand **c, size_t n) {
5682         for (size_t i = 0; i < n; i++)
5683                 c[i] = exec_command_free_list(c[i]);
5684 }
5685
5686 void exec_command_reset_status_array(ExecCommand *c, size_t n) {
5687         for (size_t i = 0; i < n; i++)
5688                 exec_status_reset(&c[i].exec_status);
5689 }
5690
5691 void exec_command_reset_status_list_array(ExecCommand **c, size_t n) {
5692         for (size_t i = 0; i < n; i++)
5693                 LIST_FOREACH(command, z, c[i])
5694                         exec_status_reset(&z->exec_status);
5695 }
5696
5697 typedef struct InvalidEnvInfo {
5698         const Unit *unit;
5699         const char *path;
5700 } InvalidEnvInfo;
5701
5702 static void invalid_env(const char *p, void *userdata) {
5703         InvalidEnvInfo *info = userdata;
5704
5705         log_unit_error(info->unit, "Ignoring invalid environment assignment '%s': %s", p, info->path);
5706 }
5707
5708 const char* exec_context_fdname(const ExecContext *c, int fd_index) {
5709         assert(c);
5710
5711         switch (fd_index) {
5712
5713         case STDIN_FILENO:
5714                 if (c->std_input != EXEC_INPUT_NAMED_FD)
5715                         return NULL;
5716
5717                 return c->stdio_fdname[STDIN_FILENO] ?: "stdin";
5718
5719         case STDOUT_FILENO:
5720                 if (c->std_output != EXEC_OUTPUT_NAMED_FD)
5721                         return NULL;
5722
5723                 return c->stdio_fdname[STDOUT_FILENO] ?: "stdout";
5724
5725         case STDERR_FILENO:
5726                 if (c->std_error != EXEC_OUTPUT_NAMED_FD)
5727                         return NULL;
5728
5729                 return c->stdio_fdname[STDERR_FILENO] ?: "stderr";
5730
5731         default:
5732                 return NULL;
5733         }
5734 }
5735
5736 static int exec_context_named_iofds(
5737                 const ExecContext *c,
5738                 const ExecParameters *p,
5739                 int named_iofds[static 3]) {
5740
5741         size_t targets;
5742         const char* stdio_fdname[3];
5743         size_t n_fds;
5744
5745         assert(c);
5746         assert(p);
5747         assert(named_iofds);
5748
5749         targets = (c->std_input == EXEC_INPUT_NAMED_FD) +
5750                   (c->std_output == EXEC_OUTPUT_NAMED_FD) +
5751                   (c->std_error == EXEC_OUTPUT_NAMED_FD);
5752
5753         for (size_t i = 0; i < 3; i++)
5754                 stdio_fdname[i] = exec_context_fdname(c, i);
5755
5756         n_fds = p->n_storage_fds + p->n_socket_fds;
5757
5758         for (size_t i = 0; i < n_fds  && targets > 0; i++)
5759                 if (named_iofds[STDIN_FILENO] < 0 &&
5760                     c->std_input == EXEC_INPUT_NAMED_FD &&
5761                     stdio_fdname[STDIN_FILENO] &&
5762                     streq(p->fd_names[i], stdio_fdname[STDIN_FILENO])) {
5763
5764                         named_iofds[STDIN_FILENO] = p->fds[i];
5765                         targets--;
5766
5767                 } else if (named_iofds[STDOUT_FILENO] < 0 &&
5768                            c->std_output == EXEC_OUTPUT_NAMED_FD &&
5769                            stdio_fdname[STDOUT_FILENO] &&
5770                            streq(p->fd_names[i], stdio_fdname[STDOUT_FILENO])) {
5771
5772                         named_iofds[STDOUT_FILENO] = p->fds[i];
5773                         targets--;
5774
5775                 } else if (named_iofds[STDERR_FILENO] < 0 &&
5776                            c->std_error == EXEC_OUTPUT_NAMED_FD &&
5777                            stdio_fdname[STDERR_FILENO] &&
5778                            streq(p->fd_names[i], stdio_fdname[STDERR_FILENO])) {
5779
5780                         named_iofds[STDERR_FILENO] = p->fds[i];
5781                         targets--;
5782                 }
5783
5784         return targets == 0 ? 0 : -ENOENT;
5785 }
5786
5787 static int exec_context_load_environment(const Unit *unit, const ExecContext *c, char ***ret) {
5788         _cleanup_strv_free_ char **v = NULL;
5789         int r;
5790
5791         assert(c);
5792         assert(ret);
5793
5794         STRV_FOREACH(i, c->environment_files) {
5795                 _cleanup_globfree_ glob_t pglob = {};
5796                 bool ignore = false;
5797                 char *fn = *i;
5798
5799                 if (fn[0] == '-') {
5800                         ignore = true;
5801                         fn++;
5802                 }
5803
5804                 if (!path_is_absolute(fn)) {
5805                         if (ignore)
5806                                 continue;
5807                         return -EINVAL;
5808                 }
5809
5810                 /* Filename supports globbing, take all matching files */
5811                 r = safe_glob(fn, 0, &pglob);
5812                 if (r < 0) {
5813                         if (ignore)
5814                                 continue;
5815                         return r;
5816                 }
5817
5818                 /* When we don't match anything, -ENOENT should be returned */
5819                 assert(pglob.gl_pathc > 0);
5820
5821                 for (unsigned n = 0; n < pglob.gl_pathc; n++) {
5822                         _cleanup_strv_free_ char **p = NULL;
5823
5824                         r = load_env_file(NULL, pglob.gl_pathv[n], &p);
5825                         if (r < 0) {
5826                                 if (ignore)
5827                                         continue;
5828                                 return r;
5829                         }
5830
5831                         /* Log invalid environment variables with filename */
5832                         if (p) {
5833                                 InvalidEnvInfo info = {
5834                                         .unit = unit,
5835                                         .path = pglob.gl_pathv[n]
5836                                 };
5837
5838                                 p = strv_env_clean_with_callback(p, invalid_env, &info);
5839                         }
5840
5841                         if (!v)
5842                                 v = TAKE_PTR(p);
5843                         else {
5844                                 char **m = strv_env_merge(v, p);
5845                                 if (!m)
5846                                         return -ENOMEM;
5847
5848                                 strv_free_and_replace(v, m);
5849                         }
5850                 }
5851         }
5852
5853         *ret = TAKE_PTR(v);
5854
5855         return 0;
5856 }
5857
5858 static bool tty_may_match_dev_console(const char *tty) {
5859         _cleanup_free_ char *resolved = NULL;
5860
5861         if (!tty)
5862                 return true;
5863
5864         tty = skip_dev_prefix(tty);
5865
5866         /* trivial identity? */
5867         if (streq(tty, "console"))
5868                 return true;
5869
5870         if (resolve_dev_console(&resolved) < 0)
5871                 return true; /* if we could not resolve, assume it may */
5872
5873         /* "tty0" means the active VC, so it may be the same sometimes */
5874         return path_equal(resolved, tty) || (streq(resolved, "tty0") && tty_is_vc(tty));
5875 }
5876
5877 static bool exec_context_may_touch_tty(const ExecContext *ec) {
5878         assert(ec);
5879
5880         return ec->tty_reset ||
5881                 ec->tty_vhangup ||
5882                 ec->tty_vt_disallocate ||
5883                 is_terminal_input(ec->std_input) ||
5884                 is_terminal_output(ec->std_output) ||
5885                 is_terminal_output(ec->std_error);
5886 }
5887
5888 bool exec_context_may_touch_console(const ExecContext *ec) {
5889
5890         return exec_context_may_touch_tty(ec) &&
5891                tty_may_match_dev_console(exec_context_tty_path(ec));
5892 }
5893
5894 static void strv_fprintf(FILE *f, char **l) {
5895         assert(f);
5896
5897         STRV_FOREACH(g, l)
5898                 fprintf(f, " %s", *g);
5899 }
5900
5901 static void strv_dump(FILE* f, const char *prefix, const char *name, char **strv) {
5902         assert(f);
5903         assert(prefix);
5904         assert(name);
5905
5906         if (!strv_isempty(strv)) {
5907                 fprintf(f, "%s%s:", prefix, name);
5908                 strv_fprintf(f, strv);
5909                 fputs("\n", f);
5910         }
5911 }
5912
5913 void exec_context_dump(const ExecContext *c, FILE* f, const char *prefix) {
5914         int r;
5915
5916         assert(c);
5917         assert(f);
5918
5919         prefix = strempty(prefix);
5920
5921         fprintf(f,
5922                 "%sUMask: %04o\n"
5923                 "%sWorkingDirectory: %s\n"
5924                 "%sRootDirectory: %s\n"
5925                 "%sNonBlocking: %s\n"
5926                 "%sPrivateTmp: %s\n"
5927                 "%sPrivateDevices: %s\n"
5928                 "%sProtectKernelTunables: %s\n"
5929                 "%sProtectKernelModules: %s\n"
5930                 "%sProtectKernelLogs: %s\n"
5931                 "%sProtectClock: %s\n"
5932                 "%sProtectControlGroups: %s\n"
5933                 "%sPrivateNetwork: %s\n"
5934                 "%sPrivateUsers: %s\n"
5935                 "%sProtectHome: %s\n"
5936                 "%sProtectSystem: %s\n"
5937                 "%sMountAPIVFS: %s\n"
5938                 "%sIgnoreSIGPIPE: %s\n"
5939                 "%sMemoryDenyWriteExecute: %s\n"
5940                 "%sRestrictRealtime: %s\n"
5941                 "%sRestrictSUIDSGID: %s\n"
5942                 "%sKeyringMode: %s\n"
5943                 "%sProtectHostname: %s\n"
5944                 "%sProtectProc: %s\n"
5945                 "%sProcSubset: %s\n",
5946                 prefix, c->umask,
5947                 prefix, empty_to_root(c->working_directory),
5948                 prefix, empty_to_root(c->root_directory),
5949                 prefix, yes_no(c->non_blocking),
5950                 prefix, yes_no(c->private_tmp),
5951                 prefix, yes_no(c->private_devices),
5952                 prefix, yes_no(c->protect_kernel_tunables),
5953                 prefix, yes_no(c->protect_kernel_modules),
5954                 prefix, yes_no(c->protect_kernel_logs),
5955                 prefix, yes_no(c->protect_clock),
5956                 prefix, yes_no(c->protect_control_groups),
5957                 prefix, yes_no(c->private_network),
5958                 prefix, yes_no(c->private_users),
5959                 prefix, protect_home_to_string(c->protect_home),
5960                 prefix, protect_system_to_string(c->protect_system),
5961                 prefix, yes_no(exec_context_get_effective_mount_apivfs(c)),
5962                 prefix, yes_no(c->ignore_sigpipe),
5963                 prefix, yes_no(c->memory_deny_write_execute),
5964                 prefix, yes_no(c->restrict_realtime),
5965                 prefix, yes_no(c->restrict_suid_sgid),
5966                 prefix, exec_keyring_mode_to_string(c->keyring_mode),
5967                 prefix, yes_no(c->protect_hostname),
5968                 prefix, protect_proc_to_string(c->protect_proc),
5969                 prefix, proc_subset_to_string(c->proc_subset));
5970
5971         if (c->root_image)
5972                 fprintf(f, "%sRootImage: %s\n", prefix, c->root_image);
5973
5974         if (c->root_image_options) {
5975                 fprintf(f, "%sRootImageOptions:", prefix);
5976                 LIST_FOREACH(mount_options, o, c->root_image_options)
5977                         if (!isempty(o->options))
5978                                 fprintf(f, " %s:%s",
5979                                         partition_designator_to_string(o->partition_designator),
5980                                         o->options);
5981                 fprintf(f, "\n");
5982         }
5983
5984         if (c->root_hash) {
5985                 _cleanup_free_ char *encoded = NULL;
5986                 encoded = hexmem(c->root_hash, c->root_hash_size);
5987                 if (encoded)
5988                         fprintf(f, "%sRootHash: %s\n", prefix, encoded);
5989         }
5990
5991         if (c->root_hash_path)
5992                 fprintf(f, "%sRootHash: %s\n", prefix, c->root_hash_path);
5993
5994         if (c->root_hash_sig) {
5995                 _cleanup_free_ char *encoded = NULL;
5996                 ssize_t len;
5997                 len = base64mem(c->root_hash_sig, c->root_hash_sig_size, &encoded);
5998                 if (len)
5999                         fprintf(f, "%sRootHashSignature: base64:%s\n", prefix, encoded);
6000         }
6001
6002         if (c->root_hash_sig_path)
6003                 fprintf(f, "%sRootHashSignature: %s\n", prefix, c->root_hash_sig_path);
6004
6005         if (c->root_verity)
6006                 fprintf(f, "%sRootVerity: %s\n", prefix, c->root_verity);
6007
6008         STRV_FOREACH(e, c->environment)
6009                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
6010
6011         STRV_FOREACH(e, c->environment_files)
6012                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
6013
6014         STRV_FOREACH(e, c->pass_environment)
6015                 fprintf(f, "%sPassEnvironment: %s\n", prefix, *e);
6016
6017         STRV_FOREACH(e, c->unset_environment)
6018                 fprintf(f, "%sUnsetEnvironment: %s\n", prefix, *e);
6019
6020         fprintf(f, "%sRuntimeDirectoryPreserve: %s\n", prefix, exec_preserve_mode_to_string(c->runtime_directory_preserve_mode));
6021
6022         for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
6023                 fprintf(f, "%s%sMode: %04o\n", prefix, exec_directory_type_to_string(dt), c->directories[dt].mode);
6024
6025                 for (size_t i = 0; i < c->directories[dt].n_items; i++) {
6026                         fprintf(f, "%s%s: %s\n", prefix, exec_directory_type_to_string(dt), c->directories[dt].items[i].path);
6027
6028                         STRV_FOREACH(d, c->directories[dt].items[i].symlinks)
6029                                 fprintf(f, "%s%s: %s:%s\n", prefix, exec_directory_type_symlink_to_string(dt), c->directories[dt].items[i].path, *d);
6030                 }
6031         }
6032
6033         fprintf(f, "%sTimeoutCleanSec: %s\n", prefix, FORMAT_TIMESPAN(c->timeout_clean_usec, USEC_PER_SEC));
6034
6035         if (c->nice_set)
6036                 fprintf(f, "%sNice: %i\n", prefix, c->nice);
6037
6038         if (c->oom_score_adjust_set)
6039                 fprintf(f, "%sOOMScoreAdjust: %i\n", prefix, c->oom_score_adjust);
6040
6041         if (c->coredump_filter_set)
6042                 fprintf(f, "%sCoredumpFilter: 0x%"PRIx64"\n", prefix, c->coredump_filter);
6043
6044         for (unsigned i = 0; i < RLIM_NLIMITS; i++)
6045                 if (c->rlimit[i]) {
6046                         fprintf(f, "%sLimit%s: " RLIM_FMT "\n",
6047                                 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
6048                         fprintf(f, "%sLimit%sSoft: " RLIM_FMT "\n",
6049                                 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_cur);
6050                 }
6051
6052         if (c->ioprio_set) {
6053                 _cleanup_free_ char *class_str = NULL;
6054
6055                 r = ioprio_class_to_string_alloc(ioprio_prio_class(c->ioprio), &class_str);
6056                 if (r >= 0)
6057                         fprintf(f, "%sIOSchedulingClass: %s\n", prefix, class_str);
6058
6059                 fprintf(f, "%sIOPriority: %d\n", prefix, ioprio_prio_data(c->ioprio));
6060         }
6061
6062         if (c->cpu_sched_set) {
6063                 _cleanup_free_ char *policy_str = NULL;
6064
6065                 r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
6066                 if (r >= 0)
6067                         fprintf(f, "%sCPUSchedulingPolicy: %s\n", prefix, policy_str);
6068
6069                 fprintf(f,
6070                         "%sCPUSchedulingPriority: %i\n"
6071                         "%sCPUSchedulingResetOnFork: %s\n",
6072                         prefix, c->cpu_sched_priority,
6073                         prefix, yes_no(c->cpu_sched_reset_on_fork));
6074         }
6075
6076         if (c->cpu_set.set) {
6077                 _cleanup_free_ char *affinity = NULL;
6078
6079                 affinity = cpu_set_to_range_string(&c->cpu_set);
6080                 fprintf(f, "%sCPUAffinity: %s\n", prefix, affinity);
6081         }
6082
6083         if (mpol_is_valid(numa_policy_get_type(&c->numa_policy))) {
6084                 _cleanup_free_ char *nodes = NULL;
6085
6086                 nodes = cpu_set_to_range_string(&c->numa_policy.nodes);
6087                 fprintf(f, "%sNUMAPolicy: %s\n", prefix, mpol_to_string(numa_policy_get_type(&c->numa_policy)));
6088                 fprintf(f, "%sNUMAMask: %s\n", prefix, strnull(nodes));
6089         }
6090
6091         if (c->timer_slack_nsec != NSEC_INFINITY)
6092                 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
6093
6094         fprintf(f,
6095                 "%sStandardInput: %s\n"
6096                 "%sStandardOutput: %s\n"
6097                 "%sStandardError: %s\n",
6098                 prefix, exec_input_to_string(c->std_input),
6099                 prefix, exec_output_to_string(c->std_output),
6100                 prefix, exec_output_to_string(c->std_error));
6101
6102         if (c->std_input == EXEC_INPUT_NAMED_FD)
6103                 fprintf(f, "%sStandardInputFileDescriptorName: %s\n", prefix, c->stdio_fdname[STDIN_FILENO]);
6104         if (c->std_output == EXEC_OUTPUT_NAMED_FD)
6105                 fprintf(f, "%sStandardOutputFileDescriptorName: %s\n", prefix, c->stdio_fdname[STDOUT_FILENO]);
6106         if (c->std_error == EXEC_OUTPUT_NAMED_FD)
6107                 fprintf(f, "%sStandardErrorFileDescriptorName: %s\n", prefix, c->stdio_fdname[STDERR_FILENO]);
6108
6109         if (c->std_input == EXEC_INPUT_FILE)
6110                 fprintf(f, "%sStandardInputFile: %s\n", prefix, c->stdio_file[STDIN_FILENO]);
6111         if (c->std_output == EXEC_OUTPUT_FILE)
6112                 fprintf(f, "%sStandardOutputFile: %s\n", prefix, c->stdio_file[STDOUT_FILENO]);
6113         if (c->std_output == EXEC_OUTPUT_FILE_APPEND)
6114                 fprintf(f, "%sStandardOutputFileToAppend: %s\n", prefix, c->stdio_file[STDOUT_FILENO]);
6115         if (c->std_output == EXEC_OUTPUT_FILE_TRUNCATE)
6116                 fprintf(f, "%sStandardOutputFileToTruncate: %s\n", prefix, c->stdio_file[STDOUT_FILENO]);
6117         if (c->std_error == EXEC_OUTPUT_FILE)
6118                 fprintf(f, "%sStandardErrorFile: %s\n", prefix, c->stdio_file[STDERR_FILENO]);
6119         if (c->std_error == EXEC_OUTPUT_FILE_APPEND)
6120                 fprintf(f, "%sStandardErrorFileToAppend: %s\n", prefix, c->stdio_file[STDERR_FILENO]);
6121         if (c->std_error == EXEC_OUTPUT_FILE_TRUNCATE)
6122                 fprintf(f, "%sStandardErrorFileToTruncate: %s\n", prefix, c->stdio_file[STDERR_FILENO]);
6123
6124         if (c->tty_path)
6125                 fprintf(f,
6126                         "%sTTYPath: %s\n"
6127                         "%sTTYReset: %s\n"
6128                         "%sTTYVHangup: %s\n"
6129                         "%sTTYVTDisallocate: %s\n"
6130                         "%sTTYRows: %u\n"
6131                         "%sTTYColumns: %u\n",
6132                         prefix, c->tty_path,
6133                         prefix, yes_no(c->tty_reset),
6134                         prefix, yes_no(c->tty_vhangup),
6135                         prefix, yes_no(c->tty_vt_disallocate),
6136                         prefix, c->tty_rows,
6137                         prefix, c->tty_cols);
6138
6139         if (IN_SET(c->std_output,
6140                    EXEC_OUTPUT_KMSG,
6141                    EXEC_OUTPUT_JOURNAL,
6142                    EXEC_OUTPUT_KMSG_AND_CONSOLE,
6143                    EXEC_OUTPUT_JOURNAL_AND_CONSOLE) ||
6144             IN_SET(c->std_error,
6145                    EXEC_OUTPUT_KMSG,
6146                    EXEC_OUTPUT_JOURNAL,
6147                    EXEC_OUTPUT_KMSG_AND_CONSOLE,
6148                    EXEC_OUTPUT_JOURNAL_AND_CONSOLE)) {
6149
6150                 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
6151
6152                 r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
6153                 if (r >= 0)
6154                         fprintf(f, "%sSyslogFacility: %s\n", prefix, fac_str);
6155
6156                 r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
6157                 if (r >= 0)
6158                         fprintf(f, "%sSyslogLevel: %s\n", prefix, lvl_str);
6159         }
6160
6161         if (c->log_level_max >= 0) {
6162                 _cleanup_free_ char *t = NULL;
6163
6164                 (void) log_level_to_string_alloc(c->log_level_max, &t);
6165
6166                 fprintf(f, "%sLogLevelMax: %s\n", prefix, strna(t));
6167         }
6168
6169         if (c->log_ratelimit_interval_usec > 0)
6170                 fprintf(f,
6171                         "%sLogRateLimitIntervalSec: %s\n",
6172                         prefix, FORMAT_TIMESPAN(c->log_ratelimit_interval_usec, USEC_PER_SEC));
6173
6174         if (c->log_ratelimit_burst > 0)
6175                 fprintf(f, "%sLogRateLimitBurst: %u\n", prefix, c->log_ratelimit_burst);
6176
6177         if (!set_isempty(c->log_filter_allowed_patterns) || !set_isempty(c->log_filter_denied_patterns)) {
6178                 fprintf(f, "%sLogFilterPatterns:", prefix);
6179
6180                 char *pattern;
6181                 SET_FOREACH(pattern, c->log_filter_allowed_patterns)
6182                         fprintf(f, " %s", pattern);
6183                 SET_FOREACH(pattern, c->log_filter_denied_patterns)
6184                         fprintf(f, " ~%s", pattern);
6185                 fputc('\n', f);
6186         }
6187
6188         for (size_t j = 0; j < c->n_log_extra_fields; j++) {
6189                 fprintf(f, "%sLogExtraFields: ", prefix);
6190                 fwrite(c->log_extra_fields[j].iov_base,
6191                        1, c->log_extra_fields[j].iov_len,
6192                        f);
6193                 fputc('\n', f);
6194         }
6195
6196         if (c->log_namespace)
6197                 fprintf(f, "%sLogNamespace: %s\n", prefix, c->log_namespace);
6198
6199         if (c->secure_bits) {
6200                 _cleanup_free_ char *str = NULL;
6201
6202                 r = secure_bits_to_string_alloc(c->secure_bits, &str);
6203                 if (r >= 0)
6204                         fprintf(f, "%sSecure Bits: %s\n", prefix, str);
6205         }
6206
6207         if (c->capability_bounding_set != CAP_MASK_UNSET) {
6208                 _cleanup_free_ char *str = NULL;
6209
6210                 r = capability_set_to_string(c->capability_bounding_set, &str);
6211                 if (r >= 0)
6212                         fprintf(f, "%sCapabilityBoundingSet: %s\n", prefix, str);
6213         }
6214
6215         if (c->capability_ambient_set != 0) {
6216                 _cleanup_free_ char *str = NULL;
6217
6218                 r = capability_set_to_string(c->capability_ambient_set, &str);
6219                 if (r >= 0)
6220                         fprintf(f, "%sAmbientCapabilities: %s\n", prefix, str);
6221         }
6222
6223         if (c->user)
6224                 fprintf(f, "%sUser: %s\n", prefix, c->user);
6225         if (c->group)
6226                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
6227
6228         fprintf(f, "%sDynamicUser: %s\n", prefix, yes_no(c->dynamic_user));
6229
6230         strv_dump(f, prefix, "SupplementaryGroups", c->supplementary_groups);
6231
6232         if (c->pam_name)
6233                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
6234
6235         strv_dump(f, prefix, "ReadWritePaths", c->read_write_paths);
6236         strv_dump(f, prefix, "ReadOnlyPaths", c->read_only_paths);
6237         strv_dump(f, prefix, "InaccessiblePaths", c->inaccessible_paths);
6238         strv_dump(f, prefix, "ExecPaths", c->exec_paths);
6239         strv_dump(f, prefix, "NoExecPaths", c->no_exec_paths);
6240         strv_dump(f, prefix, "ExecSearchPath", c->exec_search_path);
6241
6242         for (size_t i = 0; i < c->n_bind_mounts; i++)
6243                 fprintf(f, "%s%s: %s%s:%s:%s\n", prefix,
6244                         c->bind_mounts[i].read_only ? "BindReadOnlyPaths" : "BindPaths",
6245                         c->bind_mounts[i].ignore_enoent ? "-": "",
6246                         c->bind_mounts[i].source,
6247                         c->bind_mounts[i].destination,
6248                         c->bind_mounts[i].recursive ? "rbind" : "norbind");
6249
6250         for (size_t i = 0; i < c->n_temporary_filesystems; i++) {
6251                 const TemporaryFileSystem *t = c->temporary_filesystems + i;
6252
6253                 fprintf(f, "%sTemporaryFileSystem: %s%s%s\n", prefix,
6254                         t->path,
6255                         isempty(t->options) ? "" : ":",
6256                         strempty(t->options));
6257         }
6258
6259         if (c->utmp_id)
6260                 fprintf(f,
6261                         "%sUtmpIdentifier: %s\n",
6262                         prefix, c->utmp_id);
6263
6264         if (c->selinux_context)
6265                 fprintf(f,
6266                         "%sSELinuxContext: %s%s\n",
6267                         prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
6268
6269         if (c->apparmor_profile)
6270                 fprintf(f,
6271                         "%sAppArmorProfile: %s%s\n",
6272                         prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
6273
6274         if (c->smack_process_label)
6275                 fprintf(f,
6276                         "%sSmackProcessLabel: %s%s\n",
6277                         prefix, c->smack_process_label_ignore ? "-" : "", c->smack_process_label);
6278
6279         if (c->personality != PERSONALITY_INVALID)
6280                 fprintf(f,
6281                         "%sPersonality: %s\n",
6282                         prefix, strna(personality_to_string(c->personality)));
6283
6284         fprintf(f,
6285                 "%sLockPersonality: %s\n",
6286                 prefix, yes_no(c->lock_personality));
6287
6288         if (c->syscall_filter) {
6289                 fprintf(f,
6290                         "%sSystemCallFilter: ",
6291                         prefix);
6292
6293                 if (!c->syscall_allow_list)
6294                         fputc('~', f);
6295
6296 #if HAVE_SECCOMP
6297                 void *id, *val;
6298                 bool first = true;
6299                 HASHMAP_FOREACH_KEY(val, id, c->syscall_filter) {
6300                         _cleanup_free_ char *name = NULL;
6301                         const char *errno_name = NULL;
6302                         int num = PTR_TO_INT(val);
6303
6304                         if (first)
6305                                 first = false;
6306                         else
6307                                 fputc(' ', f);
6308
6309                         name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
6310                         fputs(strna(name), f);
6311
6312                         if (num >= 0) {
6313                                 errno_name = seccomp_errno_or_action_to_string(num);
6314                                 if (errno_name)
6315                                         fprintf(f, ":%s", errno_name);
6316                                 else
6317                                         fprintf(f, ":%d", num);
6318                         }
6319                 }
6320 #endif
6321
6322                 fputc('\n', f);
6323         }
6324
6325         if (c->syscall_archs) {
6326                 fprintf(f,
6327                         "%sSystemCallArchitectures:",
6328                         prefix);
6329
6330 #if HAVE_SECCOMP
6331                 void *id;
6332                 SET_FOREACH(id, c->syscall_archs)
6333                         fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
6334 #endif
6335                 fputc('\n', f);
6336         }
6337
6338         if (exec_context_restrict_namespaces_set(c)) {
6339                 _cleanup_free_ char *s = NULL;
6340
6341                 r = namespace_flags_to_string(c->restrict_namespaces, &s);
6342                 if (r >= 0)
6343                         fprintf(f, "%sRestrictNamespaces: %s\n",
6344                                 prefix, strna(s));
6345         }
6346
6347 #if HAVE_LIBBPF
6348         if (exec_context_restrict_filesystems_set(c)) {
6349                 char *fs;
6350                 SET_FOREACH(fs, c->restrict_filesystems)
6351                         fprintf(f, "%sRestrictFileSystems: %s\n", prefix, fs);
6352         }
6353 #endif
6354
6355         if (c->network_namespace_path)
6356                 fprintf(f,
6357                         "%sNetworkNamespacePath: %s\n",
6358                         prefix, c->network_namespace_path);
6359
6360         if (c->syscall_errno > 0) {
6361                 fprintf(f, "%sSystemCallErrorNumber: ", prefix);
6362
6363 #if HAVE_SECCOMP
6364                 const char *errno_name = seccomp_errno_or_action_to_string(c->syscall_errno);
6365                 if (errno_name)
6366                         fputs(errno_name, f);
6367                 else
6368                         fprintf(f, "%d", c->syscall_errno);
6369 #endif
6370                 fputc('\n', f);
6371         }
6372
6373         for (size_t i = 0; i < c->n_mount_images; i++) {
6374                 fprintf(f, "%sMountImages: %s%s:%s", prefix,
6375                         c->mount_images[i].ignore_enoent ? "-": "",
6376                         c->mount_images[i].source,
6377                         c->mount_images[i].destination);
6378                 LIST_FOREACH(mount_options, o, c->mount_images[i].mount_options)
6379                         fprintf(f, ":%s:%s",
6380                                 partition_designator_to_string(o->partition_designator),
6381                                 strempty(o->options));
6382                 fprintf(f, "\n");
6383         }
6384
6385         for (size_t i = 0; i < c->n_extension_images; i++) {
6386                 fprintf(f, "%sExtensionImages: %s%s", prefix,
6387                         c->extension_images[i].ignore_enoent ? "-": "",
6388                         c->extension_images[i].source);
6389                 LIST_FOREACH(mount_options, o, c->extension_images[i].mount_options)
6390                         fprintf(f, ":%s:%s",
6391                                 partition_designator_to_string(o->partition_designator),
6392                                 strempty(o->options));
6393                 fprintf(f, "\n");
6394         }
6395
6396         strv_dump(f, prefix, "ExtensionDirectories", c->extension_directories);
6397 }
6398
6399 bool exec_context_maintains_privileges(const ExecContext *c) {
6400         assert(c);
6401
6402         /* Returns true if the process forked off would run under
6403          * an unchanged UID or as root. */
6404
6405         if (!c->user)
6406                 return true;
6407
6408         if (streq(c->user, "root") || streq(c->user, "0"))
6409                 return true;
6410
6411         return false;
6412 }
6413
6414 int exec_context_get_effective_ioprio(const ExecContext *c) {
6415         int p;
6416
6417         assert(c);
6418
6419         if (c->ioprio_set)
6420                 return c->ioprio;
6421
6422         p = ioprio_get(IOPRIO_WHO_PROCESS, 0);
6423         if (p < 0)
6424                 return IOPRIO_DEFAULT_CLASS_AND_PRIO;
6425
6426         return ioprio_normalize(p);
6427 }
6428
6429 bool exec_context_get_effective_mount_apivfs(const ExecContext *c) {
6430         assert(c);
6431
6432         /* Explicit setting wins */
6433         if (c->mount_apivfs_set)
6434                 return c->mount_apivfs;
6435
6436         /* Default to "yes" if root directory or image are specified */
6437         if (exec_context_with_rootfs(c))
6438                 return true;
6439
6440         return false;
6441 }
6442
6443 void exec_context_free_log_extra_fields(ExecContext *c) {
6444         assert(c);
6445
6446         for (size_t l = 0; l < c->n_log_extra_fields; l++)
6447                 free(c->log_extra_fields[l].iov_base);
6448         c->log_extra_fields = mfree(c->log_extra_fields);
6449         c->n_log_extra_fields = 0;
6450 }
6451
6452 void exec_context_revert_tty(ExecContext *c) {
6453         _cleanup_close_ int fd = -EBADF;
6454         const char *path;
6455         struct stat st;
6456         int r;
6457
6458         assert(c);
6459
6460         /* First, reset the TTY (possibly kicking everybody else from the TTY) */
6461         exec_context_tty_reset(c, NULL);
6462
6463         /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
6464          * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
6465          * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
6466         if (!exec_context_may_touch_tty(c))
6467                 return;
6468
6469         path = exec_context_tty_path(c);
6470         if (!path)
6471                 return;
6472
6473         fd = open(path, O_PATH|O_CLOEXEC);
6474         if (fd < 0)
6475                 return (void) log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno,
6476                                              "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m",
6477                                              path);
6478
6479         if (fstat(fd, &st) < 0)
6480                 return (void) log_warning_errno(errno, "Failed to stat TTY '%s', ignoring: %m", path);
6481
6482         /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check
6483          * if things are a character device, since a proper check either means we'd have to open the TTY and
6484          * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects
6485          * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother
6486          * with this at all? → https://github.com/systemd/systemd/issues/19213 */
6487         if (!S_ISCHR(st.st_mode))
6488                 return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path);
6489
6490         r = fchmod_and_chown(fd, TTY_MODE, 0, TTY_GID);
6491         if (r < 0)
6492                 log_warning_errno(r, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path);
6493 }
6494
6495 int exec_context_get_clean_directories(
6496                 ExecContext *c,
6497                 char **prefix,
6498                 ExecCleanMask mask,
6499                 char ***ret) {
6500
6501         _cleanup_strv_free_ char **l = NULL;
6502         int r;
6503
6504         assert(c);
6505         assert(prefix);
6506         assert(ret);
6507
6508         for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
6509                 if (!FLAGS_SET(mask, 1U << t))
6510                         continue;
6511
6512                 if (!prefix[t])
6513                         continue;
6514
6515                 for (size_t i = 0; i < c->directories[t].n_items; i++) {
6516                         char *j;
6517
6518                         j = path_join(prefix[t], c->directories[t].items[i].path);
6519                         if (!j)
6520                                 return -ENOMEM;
6521
6522                         r = strv_consume(&l, j);
6523                         if (r < 0)
6524                                 return r;
6525
6526                         /* Also remove private directories unconditionally. */
6527                         if (t != EXEC_DIRECTORY_CONFIGURATION) {
6528                                 j = path_join(prefix[t], "private", c->directories[t].items[i].path);
6529                                 if (!j)
6530                                         return -ENOMEM;
6531
6532                                 r = strv_consume(&l, j);
6533                                 if (r < 0)
6534                                         return r;
6535                         }
6536
6537                         STRV_FOREACH(symlink, c->directories[t].items[i].symlinks) {
6538                                 j = path_join(prefix[t], *symlink);
6539                                 if (!j)
6540                                         return -ENOMEM;
6541
6542                                 r = strv_consume(&l, j);
6543                                 if (r < 0)
6544                                         return r;
6545                         }
6546                 }
6547         }
6548
6549         *ret = TAKE_PTR(l);
6550         return 0;
6551 }
6552
6553 int exec_context_get_clean_mask(ExecContext *c, ExecCleanMask *ret) {
6554         ExecCleanMask mask = 0;
6555
6556         assert(c);
6557         assert(ret);
6558
6559         for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++)
6560                 if (c->directories[t].n_items > 0)
6561                         mask |= 1U << t;
6562
6563         *ret = mask;
6564         return 0;
6565 }
6566
6567 void exec_status_start(ExecStatus *s, pid_t pid) {
6568         assert(s);
6569
6570         *s = (ExecStatus) {
6571                 .pid = pid,
6572         };
6573
6574         dual_timestamp_get(&s->start_timestamp);
6575 }
6576
6577 void exec_status_exit(ExecStatus *s, const ExecContext *context, pid_t pid, int code, int status) {
6578         assert(s);
6579
6580         if (s->pid != pid)
6581                 *s = (ExecStatus) {
6582                         .pid = pid,
6583                 };
6584
6585         dual_timestamp_get(&s->exit_timestamp);
6586
6587         s->code = code;
6588         s->status = status;
6589
6590         if (context && context->utmp_id)
6591                 (void) utmp_put_dead_process(context->utmp_id, pid, code, status);
6592 }
6593
6594 void exec_status_reset(ExecStatus *s) {
6595         assert(s);
6596
6597         *s = (ExecStatus) {};
6598 }
6599
6600 void exec_status_dump(const ExecStatus *s, FILE *f, const char *prefix) {
6601         assert(s);
6602         assert(f);
6603
6604         if (s->pid <= 0)
6605                 return;
6606
6607         prefix = strempty(prefix);
6608
6609         fprintf(f,
6610                 "%sPID: "PID_FMT"\n",
6611                 prefix, s->pid);
6612
6613         if (dual_timestamp_is_set(&s->start_timestamp))
6614                 fprintf(f,
6615                         "%sStart Timestamp: %s\n",
6616                         prefix, FORMAT_TIMESTAMP(s->start_timestamp.realtime));
6617
6618         if (dual_timestamp_is_set(&s->exit_timestamp))
6619                 fprintf(f,
6620                         "%sExit Timestamp: %s\n"
6621                         "%sExit Code: %s\n"
6622                         "%sExit Status: %i\n",
6623                         prefix, FORMAT_TIMESTAMP(s->exit_timestamp.realtime),
6624                         prefix, sigchld_code_to_string(s->code),
6625                         prefix, s->status);
6626 }
6627
6628 static void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
6629         _cleanup_free_ char *cmd = NULL;
6630         const char *prefix2;
6631
6632         assert(c);
6633         assert(f);
6634
6635         prefix = strempty(prefix);
6636         prefix2 = strjoina(prefix, "\t");
6637
6638         cmd = quote_command_line(c->argv, SHELL_ESCAPE_EMPTY);
6639
6640         fprintf(f,
6641                 "%sCommand Line: %s\n",
6642                 prefix, strnull(cmd));
6643
6644         exec_status_dump(&c->exec_status, f, prefix2);
6645 }
6646
6647 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
6648         assert(f);
6649
6650         prefix = strempty(prefix);
6651
6652         LIST_FOREACH(command, i, c)
6653                 exec_command_dump(i, f, prefix);
6654 }
6655
6656 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
6657         ExecCommand *end;
6658
6659         assert(l);
6660         assert(e);
6661
6662         if (*l) {
6663                 /* It's kind of important, that we keep the order here */
6664                 end = LIST_FIND_TAIL(command, *l);
6665                 LIST_INSERT_AFTER(command, *l, end, e);
6666         } else
6667               *l = e;
6668 }
6669
6670 int exec_command_set(ExecCommand *c, const char *path, ...) {
6671         va_list ap;
6672         char **l, *p;
6673
6674         assert(c);
6675         assert(path);
6676
6677         va_start(ap, path);
6678         l = strv_new_ap(path, ap);
6679         va_end(ap);
6680
6681         if (!l)
6682                 return -ENOMEM;
6683
6684         p = strdup(path);
6685         if (!p) {
6686                 strv_free(l);
6687                 return -ENOMEM;
6688         }
6689
6690         free_and_replace(c->path, p);
6691
6692         return strv_free_and_replace(c->argv, l);
6693 }
6694
6695 int exec_command_append(ExecCommand *c, const char *path, ...) {
6696         _cleanup_strv_free_ char **l = NULL;
6697         va_list ap;
6698         int r;
6699
6700         assert(c);
6701         assert(path);
6702
6703         va_start(ap, path);
6704         l = strv_new_ap(path, ap);
6705         va_end(ap);
6706
6707         if (!l)
6708                 return -ENOMEM;
6709
6710         r = strv_extend_strv(&c->argv, l, false);
6711         if (r < 0)
6712                 return r;
6713
6714         return 0;
6715 }
6716
6717 static void *remove_tmpdir_thread(void *p) {
6718         _cleanup_free_ char *path = p;
6719
6720         (void) rm_rf(path, REMOVE_ROOT|REMOVE_PHYSICAL);
6721         return NULL;
6722 }
6723
6724 static ExecRuntime* exec_runtime_free(ExecRuntime *rt, bool destroy) {
6725         int r;
6726
6727         if (!rt)
6728                 return NULL;
6729
6730         if (rt->manager)
6731                 (void) hashmap_remove(rt->manager->exec_runtime_by_id, rt->id);
6732
6733         /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
6734
6735         if (destroy && rt->tmp_dir && !streq(rt->tmp_dir, RUN_SYSTEMD_EMPTY)) {
6736                 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
6737
6738                 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
6739                 if (r < 0)
6740                         log_warning_errno(r, "Failed to nuke %s: %m", rt->tmp_dir);
6741                 else
6742                         rt->tmp_dir = NULL;
6743         }
6744
6745         if (destroy && rt->var_tmp_dir && !streq(rt->var_tmp_dir, RUN_SYSTEMD_EMPTY)) {
6746                 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
6747
6748                 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
6749                 if (r < 0)
6750                         log_warning_errno(r, "Failed to nuke %s: %m", rt->var_tmp_dir);
6751                 else
6752                         rt->var_tmp_dir = NULL;
6753         }
6754
6755         rt->id = mfree(rt->id);
6756         rt->tmp_dir = mfree(rt->tmp_dir);
6757         rt->var_tmp_dir = mfree(rt->var_tmp_dir);
6758         safe_close_pair(rt->netns_storage_socket);
6759         safe_close_pair(rt->ipcns_storage_socket);
6760         return mfree(rt);
6761 }
6762
6763 static void exec_runtime_freep(ExecRuntime **rt) {
6764         (void) exec_runtime_free(*rt, false);
6765 }
6766
6767 static int exec_runtime_allocate(ExecRuntime **ret, const char *id) {
6768         _cleanup_free_ char *id_copy = NULL;
6769         ExecRuntime *n;
6770
6771         assert(ret);
6772
6773         id_copy = strdup(id);
6774         if (!id_copy)
6775                 return -ENOMEM;
6776
6777         n = new(ExecRuntime, 1);
6778         if (!n)
6779                 return -ENOMEM;
6780
6781         *n = (ExecRuntime) {
6782                 .id = TAKE_PTR(id_copy),
6783                 .netns_storage_socket = PIPE_EBADF,
6784                 .ipcns_storage_socket = PIPE_EBADF,
6785         };
6786
6787         *ret = n;
6788         return 0;
6789 }
6790
6791 static int exec_runtime_add(
6792                 Manager *m,
6793                 const char *id,
6794                 char **tmp_dir,
6795                 char **var_tmp_dir,
6796                 int netns_storage_socket[2],
6797                 int ipcns_storage_socket[2],
6798                 ExecRuntime **ret) {
6799
6800         _cleanup_(exec_runtime_freep) ExecRuntime *rt = NULL;
6801         int r;
6802
6803         assert(m);
6804         assert(id);
6805
6806         /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */
6807
6808         r = exec_runtime_allocate(&rt, id);
6809         if (r < 0)
6810                 return r;
6811
6812         r = hashmap_ensure_put(&m->exec_runtime_by_id, &string_hash_ops, rt->id, rt);
6813         if (r < 0)
6814                 return r;
6815
6816         assert(!!rt->tmp_dir == !!rt->var_tmp_dir); /* We require both to be set together */
6817         rt->tmp_dir = TAKE_PTR(*tmp_dir);
6818         rt->var_tmp_dir = TAKE_PTR(*var_tmp_dir);
6819
6820         if (netns_storage_socket) {
6821                 rt->netns_storage_socket[0] = TAKE_FD(netns_storage_socket[0]);
6822                 rt->netns_storage_socket[1] = TAKE_FD(netns_storage_socket[1]);
6823         }
6824
6825         if (ipcns_storage_socket) {
6826                 rt->ipcns_storage_socket[0] = TAKE_FD(ipcns_storage_socket[0]);
6827                 rt->ipcns_storage_socket[1] = TAKE_FD(ipcns_storage_socket[1]);
6828         }
6829
6830         rt->manager = m;
6831
6832         if (ret)
6833                 *ret = rt;
6834         /* do not remove created ExecRuntime object when the operation succeeds. */
6835         TAKE_PTR(rt);
6836         return 0;
6837 }
6838
6839 static int exec_runtime_make(
6840                 Manager *m,
6841                 const ExecContext *c,
6842                 const char *id,
6843                 ExecRuntime **ret) {
6844
6845         _cleanup_(namespace_cleanup_tmpdirp) char *tmp_dir = NULL, *var_tmp_dir = NULL;
6846         _cleanup_close_pair_ int netns_storage_socket[2] = PIPE_EBADF, ipcns_storage_socket[2] = PIPE_EBADF;
6847         int r;
6848
6849         assert(m);
6850         assert(c);
6851         assert(id);
6852
6853         /* It is not necessary to create ExecRuntime object. */
6854         if (!exec_needs_network_namespace(c) && !exec_needs_ipc_namespace(c) && !c->private_tmp) {
6855                 *ret = NULL;
6856                 return 0;
6857         }
6858
6859         if (c->private_tmp &&
6860             !(prefixed_path_strv_contains(c->inaccessible_paths, "/tmp") &&
6861               (prefixed_path_strv_contains(c->inaccessible_paths, "/var/tmp") ||
6862                prefixed_path_strv_contains(c->inaccessible_paths, "/var")))) {
6863                 r = setup_tmp_dirs(id, &tmp_dir, &var_tmp_dir);
6864                 if (r < 0)
6865                         return r;
6866         }
6867
6868         if (exec_needs_network_namespace(c)) {
6869                 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, netns_storage_socket) < 0)
6870                         return -errno;
6871         }
6872
6873         if (exec_needs_ipc_namespace(c)) {
6874                 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, ipcns_storage_socket) < 0)
6875                         return -errno;
6876         }
6877
6878         r = exec_runtime_add(m, id, &tmp_dir, &var_tmp_dir, netns_storage_socket, ipcns_storage_socket, ret);
6879         if (r < 0)
6880                 return r;
6881
6882         return 1;
6883 }
6884
6885 int exec_runtime_acquire(Manager *m, const ExecContext *c, const char *id, bool create, ExecRuntime **ret) {
6886         ExecRuntime *rt;
6887         int r;
6888
6889         assert(m);
6890         assert(id);
6891         assert(ret);
6892
6893         rt = hashmap_get(m->exec_runtime_by_id, id);
6894         if (rt)
6895                 /* We already have an ExecRuntime object, let's increase the ref count and reuse it */
6896                 goto ref;
6897
6898         if (!create) {
6899                 *ret = NULL;
6900                 return 0;
6901         }
6902
6903         /* If not found, then create a new object. */
6904         r = exec_runtime_make(m, c, id, &rt);
6905         if (r < 0)
6906                 return r;
6907         if (r == 0) {
6908                 /* When r == 0, it is not necessary to create ExecRuntime object. */
6909                 *ret = NULL;
6910                 return 0;
6911         }
6912
6913 ref:
6914         /* increment reference counter. */
6915         rt->n_ref++;
6916         *ret = rt;
6917         return 1;
6918 }
6919
6920 ExecRuntime *exec_runtime_unref(ExecRuntime *rt, bool destroy) {
6921         if (!rt)
6922                 return NULL;
6923
6924         assert(rt->n_ref > 0);
6925
6926         rt->n_ref--;
6927         if (rt->n_ref > 0)
6928                 return NULL;
6929
6930         return exec_runtime_free(rt, destroy);
6931 }
6932
6933 int exec_runtime_serialize(const Manager *m, FILE *f, FDSet *fds) {
6934         ExecRuntime *rt;
6935
6936         assert(m);
6937         assert(f);
6938         assert(fds);
6939
6940         HASHMAP_FOREACH(rt, m->exec_runtime_by_id) {
6941                 fprintf(f, "exec-runtime=%s", rt->id);
6942
6943                 if (rt->tmp_dir)
6944                         fprintf(f, " tmp-dir=%s", rt->tmp_dir);
6945
6946                 if (rt->var_tmp_dir)
6947                         fprintf(f, " var-tmp-dir=%s", rt->var_tmp_dir);
6948
6949                 if (rt->netns_storage_socket[0] >= 0) {
6950                         int copy;
6951
6952                         copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
6953                         if (copy < 0)
6954                                 return copy;
6955
6956                         fprintf(f, " netns-socket-0=%i", copy);
6957                 }
6958
6959                 if (rt->netns_storage_socket[1] >= 0) {
6960                         int copy;
6961
6962                         copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
6963                         if (copy < 0)
6964                                 return copy;
6965
6966                         fprintf(f, " netns-socket-1=%i", copy);
6967                 }
6968
6969                 if (rt->ipcns_storage_socket[0] >= 0) {
6970                         int copy;
6971
6972                         copy = fdset_put_dup(fds, rt->ipcns_storage_socket[0]);
6973                         if (copy < 0)
6974                                 return copy;
6975
6976                         fprintf(f, " ipcns-socket-0=%i", copy);
6977                 }
6978
6979                 if (rt->ipcns_storage_socket[1] >= 0) {
6980                         int copy;
6981
6982                         copy = fdset_put_dup(fds, rt->ipcns_storage_socket[1]);
6983                         if (copy < 0)
6984                                 return copy;
6985
6986                         fprintf(f, " ipcns-socket-1=%i", copy);
6987                 }
6988
6989                 fputc('\n', f);
6990         }
6991
6992         return 0;
6993 }
6994
6995 int exec_runtime_deserialize_compat(Unit *u, const char *key, const char *value, FDSet *fds) {
6996         _cleanup_(exec_runtime_freep) ExecRuntime *rt_create = NULL;
6997         ExecRuntime *rt;
6998         int r;
6999
7000         /* This is for the migration from old (v237 or earlier) deserialization text.
7001          * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
7002          * Even if the ExecRuntime object originally created by the other unit, we cannot judge
7003          * so or not from the serialized text, then we always creates a new object owned by this. */
7004
7005         assert(u);
7006         assert(key);
7007         assert(value);
7008
7009         /* Manager manages ExecRuntime objects by the unit id.
7010          * So, we omit the serialized text when the unit does not have id (yet?)... */
7011         if (isempty(u->id)) {
7012                 log_unit_debug(u, "Invocation ID not found. Dropping runtime parameter.");
7013                 return 0;
7014         }
7015
7016         if (hashmap_ensure_allocated(&u->manager->exec_runtime_by_id, &string_hash_ops) < 0)
7017                 return log_oom();
7018
7019         rt = hashmap_get(u->manager->exec_runtime_by_id, u->id);
7020         if (!rt) {
7021                 if (exec_runtime_allocate(&rt_create, u->id) < 0)
7022                         return log_oom();
7023
7024                 rt = rt_create;
7025         }
7026
7027         if (streq(key, "tmp-dir")) {
7028                 if (free_and_strdup_warn(&rt->tmp_dir, value) < 0)
7029                         return -ENOMEM;
7030
7031         } else if (streq(key, "var-tmp-dir")) {
7032                 if (free_and_strdup_warn(&rt->var_tmp_dir, value) < 0)
7033                         return -ENOMEM;
7034
7035         } else if (streq(key, "netns-socket-0")) {
7036                 int fd;
7037
7038                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd)) {
7039                         log_unit_debug(u, "Failed to parse netns socket value: %s", value);
7040                         return 0;
7041                 }
7042
7043                 safe_close(rt->netns_storage_socket[0]);
7044                 rt->netns_storage_socket[0] = fdset_remove(fds, fd);
7045
7046         } else if (streq(key, "netns-socket-1")) {
7047                 int fd;
7048
7049                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd)) {
7050                         log_unit_debug(u, "Failed to parse netns socket value: %s", value);
7051                         return 0;
7052                 }
7053
7054                 safe_close(rt->netns_storage_socket[1]);
7055                 rt->netns_storage_socket[1] = fdset_remove(fds, fd);
7056
7057         } else
7058                 return 0;
7059
7060         /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
7061         if (rt_create) {
7062                 r = hashmap_put(u->manager->exec_runtime_by_id, rt_create->id, rt_create);
7063                 if (r < 0) {
7064                         log_unit_debug_errno(u, r, "Failed to put runtime parameter to manager's storage: %m");
7065                         return 0;
7066                 }
7067
7068                 rt_create->manager = u->manager;
7069
7070                 /* Avoid cleanup */
7071                 TAKE_PTR(rt_create);
7072         }
7073
7074         return 1;
7075 }
7076
7077 int exec_runtime_deserialize_one(Manager *m, const char *value, FDSet *fds) {
7078         _cleanup_free_ char *tmp_dir = NULL, *var_tmp_dir = NULL;
7079         char *id = NULL;
7080         int r, netns_fdpair[] = {-1, -1}, ipcns_fdpair[] = {-1, -1};
7081         const char *p, *v = ASSERT_PTR(value);
7082         size_t n;
7083
7084         assert(m);
7085         assert(fds);
7086
7087         n = strcspn(v, " ");
7088         id = strndupa_safe(v, n);
7089         if (v[n] != ' ')
7090                 goto finalize;
7091         p = v + n + 1;
7092
7093         v = startswith(p, "tmp-dir=");
7094         if (v) {
7095                 n = strcspn(v, " ");
7096                 tmp_dir = strndup(v, n);
7097                 if (!tmp_dir)
7098                         return log_oom();
7099                 if (v[n] != ' ')
7100                         goto finalize;
7101                 p = v + n + 1;
7102         }
7103
7104         v = startswith(p, "var-tmp-dir=");
7105         if (v) {
7106                 n = strcspn(v, " ");
7107                 var_tmp_dir = strndup(v, n);
7108                 if (!var_tmp_dir)
7109                         return log_oom();
7110                 if (v[n] != ' ')
7111                         goto finalize;
7112                 p = v + n + 1;
7113         }
7114
7115         v = startswith(p, "netns-socket-0=");
7116         if (v) {
7117                 char *buf;
7118
7119                 n = strcspn(v, " ");
7120                 buf = strndupa_safe(v, n);
7121
7122                 r = safe_atoi(buf, &netns_fdpair[0]);
7123                 if (r < 0)
7124                         return log_debug_errno(r, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf);
7125                 if (!fdset_contains(fds, netns_fdpair[0]))
7126                         return log_debug_errno(SYNTHETIC_ERRNO(EBADF),
7127                                                "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair[0]);
7128                 netns_fdpair[0] = fdset_remove(fds, netns_fdpair[0]);
7129                 if (v[n] != ' ')
7130                         goto finalize;
7131                 p = v + n + 1;
7132         }
7133
7134         v = startswith(p, "netns-socket-1=");
7135         if (v) {
7136                 char *buf;
7137
7138                 n = strcspn(v, " ");
7139                 buf = strndupa_safe(v, n);
7140
7141                 r = safe_atoi(buf, &netns_fdpair[1]);
7142                 if (r < 0)
7143                         return log_debug_errno(r, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf);
7144                 if (!fdset_contains(fds, netns_fdpair[1]))
7145                         return log_debug_errno(SYNTHETIC_ERRNO(EBADF),
7146                                                "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair[1]);
7147                 netns_fdpair[1] = fdset_remove(fds, netns_fdpair[1]);
7148                 if (v[n] != ' ')
7149                         goto finalize;
7150                 p = v + n + 1;
7151         }
7152
7153         v = startswith(p, "ipcns-socket-0=");
7154         if (v) {
7155                 char *buf;
7156
7157                 n = strcspn(v, " ");
7158                 buf = strndupa_safe(v, n);
7159
7160                 r = safe_atoi(buf, &ipcns_fdpair[0]);
7161                 if (r < 0)
7162                         return log_debug_errno(r, "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf);
7163                 if (!fdset_contains(fds, ipcns_fdpair[0]))
7164                         return log_debug_errno(SYNTHETIC_ERRNO(EBADF),
7165                                                "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair[0]);
7166                 ipcns_fdpair[0] = fdset_remove(fds, ipcns_fdpair[0]);
7167                 if (v[n] != ' ')
7168                         goto finalize;
7169                 p = v + n + 1;
7170         }
7171
7172         v = startswith(p, "ipcns-socket-1=");
7173         if (v) {
7174                 char *buf;
7175
7176                 n = strcspn(v, " ");
7177                 buf = strndupa_safe(v, n);
7178
7179                 r = safe_atoi(buf, &ipcns_fdpair[1]);
7180                 if (r < 0)
7181                         return log_debug_errno(r, "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf);
7182                 if (!fdset_contains(fds, ipcns_fdpair[1]))
7183                         return log_debug_errno(SYNTHETIC_ERRNO(EBADF),
7184                                                "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair[1]);
7185                 ipcns_fdpair[1] = fdset_remove(fds, ipcns_fdpair[1]);
7186         }
7187
7188 finalize:
7189         r = exec_runtime_add(m, id, &tmp_dir, &var_tmp_dir, netns_fdpair, ipcns_fdpair, NULL);
7190         if (r < 0)
7191                 return log_debug_errno(r, "Failed to add exec-runtime: %m");
7192         return 0;
7193 }
7194
7195 void exec_runtime_vacuum(Manager *m) {
7196         ExecRuntime *rt;
7197
7198         assert(m);
7199
7200         /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
7201
7202         HASHMAP_FOREACH(rt, m->exec_runtime_by_id) {
7203                 if (rt->n_ref > 0)
7204                         continue;
7205
7206                 (void) exec_runtime_free(rt, false);
7207         }
7208 }
7209
7210 void exec_params_clear(ExecParameters *p) {
7211         if (!p)
7212                 return;
7213
7214         p->environment = strv_free(p->environment);
7215         p->fd_names = strv_free(p->fd_names);
7216         p->fds = mfree(p->fds);
7217         p->exec_fd = safe_close(p->exec_fd);
7218 }
7219
7220 ExecSetCredential *exec_set_credential_free(ExecSetCredential *sc) {
7221         if (!sc)
7222                 return NULL;
7223
7224         free(sc->id);
7225         free(sc->data);
7226         return mfree(sc);
7227 }
7228
7229 ExecLoadCredential *exec_load_credential_free(ExecLoadCredential *lc) {
7230         if (!lc)
7231                 return NULL;
7232
7233         free(lc->id);
7234         free(lc->path);
7235         return mfree(lc);
7236 }
7237
7238 void exec_directory_done(ExecDirectory *d) {
7239         if (!d)
7240                 return;
7241
7242         for (size_t i = 0; i < d->n_items; i++) {
7243                 free(d->items[i].path);
7244                 strv_free(d->items[i].symlinks);
7245         }
7246
7247         d->items = mfree(d->items);
7248         d->n_items = 0;
7249         d->mode = 0755;
7250 }
7251
7252 static ExecDirectoryItem *exec_directory_find(ExecDirectory *d, const char *path) {
7253         assert(d);
7254         assert(path);
7255
7256         for (size_t i = 0; i < d->n_items; i++)
7257                 if (path_equal(d->items[i].path, path))
7258                         return &d->items[i];
7259
7260         return NULL;
7261 }
7262
7263 int exec_directory_add(ExecDirectory *d, const char *path, const char *symlink) {
7264         _cleanup_strv_free_ char **s = NULL;
7265         _cleanup_free_ char *p = NULL;
7266         ExecDirectoryItem *existing;
7267         int r;
7268
7269         assert(d);
7270         assert(path);
7271
7272         existing = exec_directory_find(d, path);
7273         if (existing) {
7274                 r = strv_extend(&existing->symlinks, symlink);
7275                 if (r < 0)
7276                         return r;
7277
7278                 return 0; /* existing item is updated */
7279         }
7280
7281         p = strdup(path);
7282         if (!p)
7283                 return -ENOMEM;
7284
7285         if (symlink) {
7286                 s = strv_new(symlink);
7287                 if (!s)
7288                         return -ENOMEM;
7289         }
7290
7291         if (!GREEDY_REALLOC(d->items, d->n_items + 1))
7292                 return -ENOMEM;
7293
7294         d->items[d->n_items++] = (ExecDirectoryItem) {
7295                 .path = TAKE_PTR(p),
7296                 .symlinks = TAKE_PTR(s),
7297         };
7298
7299         return 1; /* new item is added */
7300 }
7301
7302 static int exec_directory_item_compare_func(const ExecDirectoryItem *a, const ExecDirectoryItem *b) {
7303         assert(a);
7304         assert(b);
7305
7306         return path_compare(a->path, b->path);
7307 }
7308
7309 void exec_directory_sort(ExecDirectory *d) {
7310         assert(d);
7311
7312         /* Sort the exec directories to make always parent directories processed at first in
7313          * setup_exec_directory(), e.g., even if StateDirectory=foo/bar foo, we need to create foo at first,
7314          * then foo/bar. Also, set .only_create flag if one of the parent directories is contained in the
7315          * list. See also comments in setup_exec_directory() and issue #24783. */
7316
7317         if (d->n_items <= 1)
7318                 return;
7319
7320         typesafe_qsort(d->items, d->n_items, exec_directory_item_compare_func);
7321
7322         for (size_t i = 1; i < d->n_items; i++)
7323                 for (size_t j = 0; j < i; j++)
7324                         if (path_startswith(d->items[i].path, d->items[j].path)) {
7325                                 d->items[i].only_create = true;
7326                                 break;
7327                         }
7328 }
7329
7330 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops, char, string_hash_func, string_compare_func, ExecSetCredential, exec_set_credential_free);
7331 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_load_credential_hash_ops, char, string_hash_func, string_compare_func, ExecLoadCredential, exec_load_credential_free);
7332
7333 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
7334         [EXEC_INPUT_NULL] = "null",
7335         [EXEC_INPUT_TTY] = "tty",
7336         [EXEC_INPUT_TTY_FORCE] = "tty-force",
7337         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
7338         [EXEC_INPUT_SOCKET] = "socket",
7339         [EXEC_INPUT_NAMED_FD] = "fd",
7340         [EXEC_INPUT_DATA] = "data",
7341         [EXEC_INPUT_FILE] = "file",
7342 };
7343
7344 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
7345
7346 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
7347         [EXEC_OUTPUT_INHERIT] = "inherit",
7348         [EXEC_OUTPUT_NULL] = "null",
7349         [EXEC_OUTPUT_TTY] = "tty",
7350         [EXEC_OUTPUT_KMSG] = "kmsg",
7351         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
7352         [EXEC_OUTPUT_JOURNAL] = "journal",
7353         [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
7354         [EXEC_OUTPUT_SOCKET] = "socket",
7355         [EXEC_OUTPUT_NAMED_FD] = "fd",
7356         [EXEC_OUTPUT_FILE] = "file",
7357         [EXEC_OUTPUT_FILE_APPEND] = "append",
7358         [EXEC_OUTPUT_FILE_TRUNCATE] = "truncate",
7359 };
7360
7361 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
7362
7363 static const char* const exec_utmp_mode_table[_EXEC_UTMP_MODE_MAX] = {
7364         [EXEC_UTMP_INIT] = "init",
7365         [EXEC_UTMP_LOGIN] = "login",
7366         [EXEC_UTMP_USER] = "user",
7367 };
7368
7369 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode, ExecUtmpMode);
7370
7371 static const char* const exec_preserve_mode_table[_EXEC_PRESERVE_MODE_MAX] = {
7372         [EXEC_PRESERVE_NO] = "no",
7373         [EXEC_PRESERVE_YES] = "yes",
7374         [EXEC_PRESERVE_RESTART] = "restart",
7375 };
7376
7377 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode, ExecPreserveMode, EXEC_PRESERVE_YES);
7378
7379 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
7380 static const char* const exec_directory_type_table[_EXEC_DIRECTORY_TYPE_MAX] = {
7381         [EXEC_DIRECTORY_RUNTIME] = "RuntimeDirectory",
7382         [EXEC_DIRECTORY_STATE] = "StateDirectory",
7383         [EXEC_DIRECTORY_CACHE] = "CacheDirectory",
7384         [EXEC_DIRECTORY_LOGS] = "LogsDirectory",
7385         [EXEC_DIRECTORY_CONFIGURATION] = "ConfigurationDirectory",
7386 };
7387
7388 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type, ExecDirectoryType);
7389
7390 /* This table maps ExecDirectoryType to the symlink setting it is configured with in the unit */
7391 static const char* const exec_directory_type_symlink_table[_EXEC_DIRECTORY_TYPE_MAX] = {
7392         [EXEC_DIRECTORY_RUNTIME]       = "RuntimeDirectorySymlink",
7393         [EXEC_DIRECTORY_STATE]         = "StateDirectorySymlink",
7394         [EXEC_DIRECTORY_CACHE]         = "CacheDirectorySymlink",
7395         [EXEC_DIRECTORY_LOGS]          = "LogsDirectorySymlink",
7396         [EXEC_DIRECTORY_CONFIGURATION] = "ConfigurationDirectorySymlink",
7397 };
7398
7399 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type_symlink, ExecDirectoryType);
7400
7401 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
7402  * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
7403  * directories, specifically .timer units with their timestamp touch file. */
7404 static const char* const exec_resource_type_table[_EXEC_DIRECTORY_TYPE_MAX] = {
7405         [EXEC_DIRECTORY_RUNTIME] = "runtime",
7406         [EXEC_DIRECTORY_STATE] = "state",
7407         [EXEC_DIRECTORY_CACHE] = "cache",
7408         [EXEC_DIRECTORY_LOGS] = "logs",
7409         [EXEC_DIRECTORY_CONFIGURATION] = "configuration",
7410 };
7411
7412 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type, ExecDirectoryType);
7413
7414 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
7415  * the service payload in. */
7416 static const char* const exec_directory_env_name_table[_EXEC_DIRECTORY_TYPE_MAX] = {
7417         [EXEC_DIRECTORY_RUNTIME] = "RUNTIME_DIRECTORY",
7418         [EXEC_DIRECTORY_STATE] = "STATE_DIRECTORY",
7419         [EXEC_DIRECTORY_CACHE] = "CACHE_DIRECTORY",
7420         [EXEC_DIRECTORY_LOGS] = "LOGS_DIRECTORY",
7421         [EXEC_DIRECTORY_CONFIGURATION] = "CONFIGURATION_DIRECTORY",
7422 };
7423
7424 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name, ExecDirectoryType);
7425
7426 static const char* const exec_keyring_mode_table[_EXEC_KEYRING_MODE_MAX] = {
7427         [EXEC_KEYRING_INHERIT] = "inherit",
7428         [EXEC_KEYRING_PRIVATE] = "private",
7429         [EXEC_KEYRING_SHARED] = "shared",
7430 };
7431
7432 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode, ExecKeyringMode);