src/core/execute.c

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