src/core/execute.c

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