src/core/execute.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2010 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <errno.h>
  21 #include <fcntl.h>
  22 #include <glob.h>
  23 #include <grp.h>
  24 #include <poll.h>
  25 #include <signal.h>
  26 #include <string.h>
  27 #include <sys/capability.h>
  28 #include <sys/eventfd.h>
  29 #include <sys/mman.h>
  30 #include <sys/personality.h>
  31 #include <sys/prctl.h>
  32 #include <sys/shm.h>
  33 #include <sys/socket.h>
  34 #include <sys/stat.h>
  35 #include <sys/types.h>
  36 #include <sys/un.h>
  37 #include <unistd.h>
  38 #include <utmpx.h>
  39
  40 #ifdef HAVE_PAM
  41 #include <security/pam_appl.h>
  42 #endif
  43
  44 #ifdef HAVE_SELINUX
  45 #include <selinux/selinux.h>
  46 #endif
  47
  48 #ifdef HAVE_SECCOMP
  49 #include <seccomp.h>
  50 #endif
  51
  52 #ifdef HAVE_APPARMOR
  53 #include <sys/apparmor.h>
  54 #endif
  55
  56 #include "sd-messages.h"
  57
  58 #include "af-list.h"
  59 #include "alloc-util.h"
  60 #ifdef HAVE_APPARMOR
  61 #include "apparmor-util.h"
  62 #endif
  63 #include "async.h"
  64 #include "barrier.h"
  65 #include "cap-list.h"
  66 #include "capability-util.h"
  67 #include "def.h"
  68 #include "env-util.h"
  69 #include "errno-list.h"
  70 #include "execute.h"
  71 #include "exit-status.h"
  72 #include "fd-util.h"
  73 #include "fileio.h"
  74 #include "format-util.h"
  75 #include "fs-util.h"
  76 #include "glob-util.h"
  77 #include "io-util.h"
  78 #include "ioprio.h"
  79 #include "log.h"
  80 #include "macro.h"
  81 #include "missing.h"
  82 #include "mkdir.h"
  83 #include "namespace.h"
  84 #include "parse-util.h"
  85 #include "path-util.h"
  86 #include "process-util.h"
  87 #include "rlimit-util.h"
  88 #include "rm-rf.h"
  89 #ifdef HAVE_SECCOMP
  90 #include "seccomp-util.h"
  91 #endif
  92 #include "securebits.h"
  93 #include "selinux-util.h"
  94 #include "signal-util.h"
  95 #include "smack-util.h"
  96 #include "special.h"
  97 #include "string-table.h"
  98 #include "string-util.h"
  99 #include "strv.h"
 100 #include "syslog-util.h"
 101 #include "terminal-util.h"
 102 #include "unit.h"
 103 #include "user-util.h"
 104 #include "util.h"
 105 #include "utmp-wtmp.h"
 106
 107 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
 108 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
 109
 110 /* This assumes there is a 'tty' group */
 111 #define TTY_MODE 0620
 112
 113 #define SNDBUF_SIZE (8*1024*1024)
 114
 115 static int shift_fds(int fds[], unsigned n_fds) {
 116         int start, restart_from;
 117
 118         if (n_fds <= 0)
 119                 return 0;
 120
 121         /* Modifies the fds array! (sorts it) */
 122
 123         assert(fds);
 124
 125         start = 0;
 126         for (;;) {
 127                 int i;
 128
 129                 restart_from = -1;
 130
 131                 for (i = start; i < (int) n_fds; i++) {
 132                         int nfd;
 133
 134                         /* Already at right index? */
 135                         if (fds[i] == i+3)
 136                                 continue;
 137
 138                         nfd = fcntl(fds[i], F_DUPFD, i + 3);
 139                         if (nfd < 0)
 140                                 return -errno;
 141
 142                         safe_close(fds[i]);
 143                         fds[i] = nfd;
 144
 145                         /* Hmm, the fd we wanted isn't free? Then
 146                          * let's remember that and try again from here */
 147                         if (nfd != i+3 && restart_from < 0)
 148                                 restart_from = i;
 149                 }
 150
 151                 if (restart_from < 0)
 152                         break;
 153
 154                 start = restart_from;
 155         }
 156
 157         return 0;
 158 }
 159
 160 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
 161         unsigned i;
 162         int r;
 163
 164         if (n_fds <= 0)
 165                 return 0;
 166
 167         assert(fds);
 168
 169         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
 170
 171         for (i = 0; i < n_fds; i++) {
 172
 173                 r = fd_nonblock(fds[i], nonblock);
 174                 if (r < 0)
 175                         return r;
 176
 177                 /* We unconditionally drop FD_CLOEXEC from the fds,
 178                  * since after all we want to pass these fds to our
 179                  * children */
 180
 181                 r = fd_cloexec(fds[i], false);
 182                 if (r < 0)
 183                         return r;
 184         }
 185
 186         return 0;
 187 }
 188
 189 static const char *exec_context_tty_path(const ExecContext *context) {
 190         assert(context);
 191
 192         if (context->stdio_as_fds)
 193                 return NULL;
 194
 195         if (context->tty_path)
 196                 return context->tty_path;
 197
 198         return "/dev/console";
 199 }
 200
 201 static void exec_context_tty_reset(const ExecContext *context, const ExecParameters *p) {
 202         const char *path;
 203
 204         assert(context);
 205
 206         path = exec_context_tty_path(context);
 207
 208         if (context->tty_vhangup) {
 209                 if (p && p->stdin_fd >= 0)
 210                         (void) terminal_vhangup_fd(p->stdin_fd);
 211                 else if (path)
 212                         (void) terminal_vhangup(path);
 213         }
 214
 215         if (context->tty_reset) {
 216                 if (p && p->stdin_fd >= 0)
 217                         (void) reset_terminal_fd(p->stdin_fd, true);
 218                 else if (path)
 219                         (void) reset_terminal(path);
 220         }
 221
 222         if (context->tty_vt_disallocate && path)
 223                 (void) vt_disallocate(path);
 224 }
 225
 226 static bool is_terminal_input(ExecInput i) {
 227         return IN_SET(i,
 228                       EXEC_INPUT_TTY,
 229                       EXEC_INPUT_TTY_FORCE,
 230                       EXEC_INPUT_TTY_FAIL);
 231 }
 232
 233 static bool is_terminal_output(ExecOutput o) {
 234         return IN_SET(o,
 235                       EXEC_OUTPUT_TTY,
 236                       EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
 237                       EXEC_OUTPUT_KMSG_AND_CONSOLE,
 238                       EXEC_OUTPUT_JOURNAL_AND_CONSOLE);
 239 }
 240
 241 static bool exec_context_needs_term(const ExecContext *c) {
 242         assert(c);
 243
 244         /* Return true if the execution context suggests we should set $TERM to something useful. */
 245
 246         if (is_terminal_input(c->std_input))
 247                 return true;
 248
 249         if (is_terminal_output(c->std_output))
 250                 return true;
 251
 252         if (is_terminal_output(c->std_error))
 253                 return true;
 254
 255         return !!c->tty_path;
 256 }
 257
 258 static int open_null_as(int flags, int nfd) {
 259         int fd, r;
 260
 261         assert(nfd >= 0);
 262
 263         fd = open("/dev/null", flags|O_NOCTTY);
 264         if (fd < 0)
 265                 return -errno;
 266
 267         if (fd != nfd) {
 268                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 269                 safe_close(fd);
 270         } else
 271                 r = nfd;
 272
 273         return r;
 274 }
 275
 276 static int connect_journal_socket(int fd, uid_t uid, gid_t gid) {
 277         union sockaddr_union sa = {
 278                 .un.sun_family = AF_UNIX,
 279                 .un.sun_path = "/run/systemd/journal/stdout",
 280         };
 281         uid_t olduid = UID_INVALID;
 282         gid_t oldgid = GID_INVALID;
 283         int r;
 284
 285         if (gid != GID_INVALID) {
 286                 oldgid = getgid();
 287
 288                 r = setegid(gid);
 289                 if (r < 0)
 290                         return -errno;
 291         }
 292
 293         if (uid != UID_INVALID) {
 294                 olduid = getuid();
 295
 296                 r = seteuid(uid);
 297                 if (r < 0) {
 298                         r = -errno;
 299                         goto restore_gid;
 300                 }
 301         }
 302
 303         r = connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 304         if (r < 0)
 305                 r = -errno;
 306
 307         /* If we fail to restore the uid or gid, things will likely
 308            fail later on. This should only happen if an LSM interferes. */
 309
 310         if (uid != UID_INVALID)
 311                 (void) seteuid(olduid);
 312
 313  restore_gid:
 314         if (gid != GID_INVALID)
 315                 (void) setegid(oldgid);
 316
 317         return r;
 318 }
 319
 320 static int connect_logger_as(
 321                 Unit *unit,
 322                 const ExecContext *context,
 323                 ExecOutput output,
 324                 const char *ident,
 325                 int nfd,
 326                 uid_t uid,
 327                 gid_t gid) {
 328
 329         int fd, r;
 330
 331         assert(context);
 332         assert(output < _EXEC_OUTPUT_MAX);
 333         assert(ident);
 334         assert(nfd >= 0);
 335
 336         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 337         if (fd < 0)
 338                 return -errno;
 339
 340         r = connect_journal_socket(fd, uid, gid);
 341         if (r < 0)
 342                 return r;
 343
 344         if (shutdown(fd, SHUT_RD) < 0) {
 345                 safe_close(fd);
 346                 return -errno;
 347         }
 348
 349         (void) fd_inc_sndbuf(fd, SNDBUF_SIZE);
 350
 351         dprintf(fd,
 352                 "%s\n"
 353                 "%s\n"
 354                 "%i\n"
 355                 "%i\n"
 356                 "%i\n"
 357                 "%i\n"
 358                 "%i\n",
 359                 context->syslog_identifier ? context->syslog_identifier : ident,
 360                 unit->id,
 361                 context->syslog_priority,
 362                 !!context->syslog_level_prefix,
 363                 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
 364                 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
 365                 is_terminal_output(output));
 366
 367         if (fd == nfd)
 368                 return nfd;
 369
 370         r = dup2(fd, nfd) < 0 ? -errno : nfd;
 371         safe_close(fd);
 372
 373         return r;
 374 }
 375 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
 376         int fd, r;
 377
 378         assert(path);
 379         assert(nfd >= 0);
 380
 381         fd = open_terminal(path, mode | O_NOCTTY);
 382         if (fd < 0)
 383                 return fd;
 384
 385         if (fd != nfd) {
 386                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 387                 safe_close(fd);
 388         } else
 389                 r = nfd;
 390
 391         return r;
 392 }
 393
 394 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
 395
 396         if (is_terminal_input(std_input) && !apply_tty_stdin)
 397                 return EXEC_INPUT_NULL;
 398
 399         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 400                 return EXEC_INPUT_NULL;
 401
 402         return std_input;
 403 }
 404
 405 static int fixup_output(ExecOutput std_output, int socket_fd) {
 406
 407         if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 408                 return EXEC_OUTPUT_INHERIT;
 409
 410         return std_output;
 411 }
 412
 413 static int setup_input(
 414                 const ExecContext *context,
 415                 const ExecParameters *params,
 416                 int socket_fd,
 417                 int named_iofds[3]) {
 418
 419         ExecInput i;
 420
 421         assert(context);
 422         assert(params);
 423
 424         if (params->stdin_fd >= 0) {
 425                 if (dup2(params->stdin_fd, STDIN_FILENO) < 0)
 426                         return -errno;
 427
 428                 /* Try to make this the controlling tty, if it is a tty, and reset it */
 429                 (void) ioctl(STDIN_FILENO, TIOCSCTTY, context->std_input == EXEC_INPUT_TTY_FORCE);
 430                 (void) reset_terminal_fd(STDIN_FILENO, true);
 431
 432                 return STDIN_FILENO;
 433         }
 434
 435         i = fixup_input(context->std_input, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN);
 436
 437         switch (i) {
 438
 439         case EXEC_INPUT_NULL:
 440                 return open_null_as(O_RDONLY, STDIN_FILENO);
 441
 442         case EXEC_INPUT_TTY:
 443         case EXEC_INPUT_TTY_FORCE:
 444         case EXEC_INPUT_TTY_FAIL: {
 445                 int fd, r;
 446
 447                 fd = acquire_terminal(exec_context_tty_path(context),
 448                                       i == EXEC_INPUT_TTY_FAIL,
 449                                       i == EXEC_INPUT_TTY_FORCE,
 450                                       false,
 451                                       USEC_INFINITY);
 452                 if (fd < 0)
 453                         return fd;
 454
 455                 if (fd != STDIN_FILENO) {
 456                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 457                         safe_close(fd);
 458                 } else
 459                         r = STDIN_FILENO;
 460
 461                 return r;
 462         }
 463
 464         case EXEC_INPUT_SOCKET:
 465                 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 466
 467         case EXEC_INPUT_NAMED_FD:
 468                 (void) fd_nonblock(named_iofds[STDIN_FILENO], false);
 469                 return dup2(named_iofds[STDIN_FILENO], STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 470
 471         default:
 472                 assert_not_reached("Unknown input type");
 473         }
 474 }
 475
 476 static int setup_output(
 477                 Unit *unit,
 478                 const ExecContext *context,
 479                 const ExecParameters *params,
 480                 int fileno,
 481                 int socket_fd,
 482                 int named_iofds[3],
 483                 const char *ident,
 484                 uid_t uid,
 485                 gid_t gid,
 486                 dev_t *journal_stream_dev,
 487                 ino_t *journal_stream_ino) {
 488
 489         ExecOutput o;
 490         ExecInput i;
 491         int r;
 492
 493         assert(unit);
 494         assert(context);
 495         assert(params);
 496         assert(ident);
 497         assert(journal_stream_dev);
 498         assert(journal_stream_ino);
 499
 500         if (fileno == STDOUT_FILENO && params->stdout_fd >= 0) {
 501
 502                 if (dup2(params->stdout_fd, STDOUT_FILENO) < 0)
 503                         return -errno;
 504
 505                 return STDOUT_FILENO;
 506         }
 507
 508         if (fileno == STDERR_FILENO && params->stderr_fd >= 0) {
 509                 if (dup2(params->stderr_fd, STDERR_FILENO) < 0)
 510                         return -errno;
 511
 512                 return STDERR_FILENO;
 513         }
 514
 515         i = fixup_input(context->std_input, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN);
 516         o = fixup_output(context->std_output, socket_fd);
 517
 518         if (fileno == STDERR_FILENO) {
 519                 ExecOutput e;
 520                 e = fixup_output(context->std_error, socket_fd);
 521
 522                 /* This expects the input and output are already set up */
 523
 524                 /* Don't change the stderr file descriptor if we inherit all
 525                  * the way and are not on a tty */
 526                 if (e == EXEC_OUTPUT_INHERIT &&
 527                     o == EXEC_OUTPUT_INHERIT &&
 528                     i == EXEC_INPUT_NULL &&
 529                     !is_terminal_input(context->std_input) &&
 530                     getppid () != 1)
 531                         return fileno;
 532
 533                 /* Duplicate from stdout if possible */
 534                 if ((e == o && e != EXEC_OUTPUT_NAMED_FD) || e == EXEC_OUTPUT_INHERIT)
 535                         return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
 536
 537                 o = e;
 538
 539         } else if (o == EXEC_OUTPUT_INHERIT) {
 540                 /* If input got downgraded, inherit the original value */
 541                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 542                         return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno);
 543
 544                 /* If the input is connected to anything that's not a /dev/null, inherit that... */
 545                 if (i != EXEC_INPUT_NULL)
 546                         return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
 547
 548                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 549                 if (getppid() != 1)
 550                         return fileno;
 551
 552                 /* We need to open /dev/null here anew, to get the right access mode. */
 553                 return open_null_as(O_WRONLY, fileno);
 554         }
 555
 556         switch (o) {
 557
 558         case EXEC_OUTPUT_NULL:
 559                 return open_null_as(O_WRONLY, fileno);
 560
 561         case EXEC_OUTPUT_TTY:
 562                 if (is_terminal_input(i))
 563                         return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
 564
 565                 /* We don't reset the terminal if this is just about output */
 566                 return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno);
 567
 568         case EXEC_OUTPUT_SYSLOG:
 569         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 570         case EXEC_OUTPUT_KMSG:
 571         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 572         case EXEC_OUTPUT_JOURNAL:
 573         case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
 574                 r = connect_logger_as(unit, context, o, ident, fileno, uid, gid);
 575                 if (r < 0) {
 576                         log_unit_error_errno(unit, r, "Failed to connect %s to the journal socket, ignoring: %m", fileno == STDOUT_FILENO ? "stdout" : "stderr");
 577                         r = open_null_as(O_WRONLY, fileno);
 578                 } else {
 579                         struct stat st;
 580
 581                         /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
 582                          * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
 583                          * services to detect whether they are connected to the journal or not. */
 584
 585                         if (fstat(fileno, &st) >= 0) {
 586                                 *journal_stream_dev = st.st_dev;
 587                                 *journal_stream_ino = st.st_ino;
 588                         }
 589                 }
 590                 return r;
 591
 592         case EXEC_OUTPUT_SOCKET:
 593                 assert(socket_fd >= 0);
 594                 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
 595
 596         case EXEC_OUTPUT_NAMED_FD:
 597                 (void) fd_nonblock(named_iofds[fileno], false);
 598                 return dup2(named_iofds[fileno], fileno) < 0 ? -errno : fileno;
 599
 600         default:
 601                 assert_not_reached("Unknown error type");
 602         }
 603 }
 604
 605 static int chown_terminal(int fd, uid_t uid) {
 606         struct stat st;
 607
 608         assert(fd >= 0);
 609
 610         /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
 611         if (isatty(fd) < 1)
 612                 return 0;
 613
 614         /* This might fail. What matters are the results. */
 615         (void) fchown(fd, uid, -1);
 616         (void) fchmod(fd, TTY_MODE);
 617
 618         if (fstat(fd, &st) < 0)
 619                 return -errno;
 620
 621         if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
 622                 return -EPERM;
 623
 624         return 0;
 625 }
 626
 627 static int setup_confirm_stdio(const char *vc, int *_saved_stdin, int *_saved_stdout) {
 628         _cleanup_close_ int fd = -1, saved_stdin = -1, saved_stdout = -1;
 629         int r;
 630
 631         assert(_saved_stdin);
 632         assert(_saved_stdout);
 633
 634         saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
 635         if (saved_stdin < 0)
 636                 return -errno;
 637
 638         saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
 639         if (saved_stdout < 0)
 640                 return -errno;
 641
 642         fd = acquire_terminal(vc, false, false, false, DEFAULT_CONFIRM_USEC);
 643         if (fd < 0)
 644                 return fd;
 645
 646         r = chown_terminal(fd, getuid());
 647         if (r < 0)
 648                 return r;
 649
 650         r = reset_terminal_fd(fd, true);
 651         if (r < 0)
 652                 return r;
 653
 654         if (dup2(fd, STDIN_FILENO) < 0)
 655                 return -errno;
 656
 657         if (dup2(fd, STDOUT_FILENO) < 0)
 658                 return -errno;
 659
 660         if (fd >= 2)
 661                 safe_close(fd);
 662         fd = -1;
 663
 664         *_saved_stdin = saved_stdin;
 665         *_saved_stdout = saved_stdout;
 666
 667         saved_stdin = saved_stdout = -1;
 668
 669         return 0;
 670 }
 671
 672 static void write_confirm_error_fd(int err, int fd, const Unit *u) {
 673         assert(err < 0);
 674
 675         if (err == -ETIMEDOUT)
 676                 dprintf(fd, "Confirmation question timed out for %s, assuming positive response.\n", u->id);
 677         else {
 678                 errno = -err;
 679                 dprintf(fd, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u->id);
 680         }
 681 }
 682
 683 static void write_confirm_error(int err, const char *vc, const Unit *u) {
 684         _cleanup_close_ int fd = -1;
 685
 686         assert(vc);
 687
 688         fd = open_terminal(vc, O_WRONLY|O_NOCTTY|O_CLOEXEC);
 689         if (fd < 0)
 690                 return;
 691
 692         write_confirm_error_fd(err, fd, u);
 693 }
 694
 695 static int restore_confirm_stdio(int *saved_stdin, int *saved_stdout) {
 696         int r = 0;
 697
 698         assert(saved_stdin);
 699         assert(saved_stdout);
 700
 701         release_terminal();
 702
 703         if (*saved_stdin >= 0)
 704                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 705                         r = -errno;
 706
 707         if (*saved_stdout >= 0)
 708                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 709                         r = -errno;
 710
 711         *saved_stdin = safe_close(*saved_stdin);
 712         *saved_stdout = safe_close(*saved_stdout);
 713
 714         return r;
 715 }
 716
 717 enum {
 718         CONFIRM_PRETEND_FAILURE = -1,
 719         CONFIRM_PRETEND_SUCCESS =  0,
 720         CONFIRM_EXECUTE = 1,
 721 };
 722
 723 static int ask_for_confirmation(const char *vc, Unit *u, const char *cmdline) {
 724         int saved_stdout = -1, saved_stdin = -1, r;
 725         _cleanup_free_ char *e = NULL;
 726         char c;
 727
 728         /* For any internal errors, assume a positive response. */
 729         r = setup_confirm_stdio(vc, &saved_stdin, &saved_stdout);
 730         if (r < 0) {
 731                 write_confirm_error(r, vc, u);
 732                 return CONFIRM_EXECUTE;
 733         }
 734
 735         /* confirm_spawn might have been disabled while we were sleeping. */
 736         if (manager_is_confirm_spawn_disabled(u->manager)) {
 737                 r = 1;
 738                 goto restore_stdio;
 739         }
 740
 741         e = ellipsize(cmdline, 60, 100);
 742         if (!e) {
 743                 log_oom();
 744                 r = CONFIRM_EXECUTE;
 745                 goto restore_stdio;
 746         }
 747
 748         for (;;) {
 749                 r = ask_char(&c, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e);
 750                 if (r < 0) {
 751                         write_confirm_error_fd(r, STDOUT_FILENO, u);
 752                         r = CONFIRM_EXECUTE;
 753                         goto restore_stdio;
 754                 }
 755
 756                 switch (c) {
 757                 case 'c':
 758                         printf("Resuming normal execution.\n");
 759                         manager_disable_confirm_spawn();
 760                         r = 1;
 761                         break;
 762                 case 'D':
 763                         unit_dump(u, stdout, "  ");
 764                         continue; /* ask again */
 765                 case 'f':
 766                         printf("Failing execution.\n");
 767                         r = CONFIRM_PRETEND_FAILURE;
 768                         break;
 769                 case 'h':
 770                         printf("  c - continue, proceed without asking anymore\n"
 771                                "  D - dump, show the state of the unit\n"
 772                                "  f - fail, don't execute the command and pretend it failed\n"
 773                                "  h - help\n"
 774                                "  i - info, show a short summary of the unit\n"
 775                                "  j - jobs, show jobs that are in progress\n"
 776                                "  s - skip, don't execute the command and pretend it succeeded\n"
 777                                "  y - yes, execute the command\n");
 778                         continue; /* ask again */
 779                 case 'i':
 780                         printf("  Description: %s\n"
 781                                "  Unit:        %s\n"
 782                                "  Command:     %s\n",
 783                                u->id, u->description, cmdline);
 784                         continue; /* ask again */
 785                 case 'j':
 786                         manager_dump_jobs(u->manager, stdout, "  ");
 787                         continue; /* ask again */
 788                 case 'n':
 789                         /* 'n' was removed in favor of 'f'. */
 790                         printf("Didn't understand 'n', did you mean 'f'?\n");
 791                         continue; /* ask again */
 792                 case 's':
 793                         printf("Skipping execution.\n");
 794                         r = CONFIRM_PRETEND_SUCCESS;
 795                         break;
 796                 case 'y':
 797                         r = CONFIRM_EXECUTE;
 798                         break;
 799                 default:
 800                         assert_not_reached("Unhandled choice");
 801                 }
 802                 break;
 803         }
 804
 805 restore_stdio:
 806         restore_confirm_stdio(&saved_stdin, &saved_stdout);
 807         return r;
 808 }
 809
 810 static int get_fixed_user(const ExecContext *c, const char **user,
 811                           uid_t *uid, gid_t *gid,
 812                           const char **home, const char **shell) {
 813         int r;
 814         const char *name;
 815
 816         assert(c);
 817
 818         /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
 819          * (i.e. are "/" or "/bin/nologin"). */
 820
 821         name = c->user ?: "root";
 822         r = get_user_creds_clean(&name, uid, gid, home, shell);
 823         if (r < 0)
 824                 return r;
 825
 826         *user = name;
 827         return 0;
 828 }
 829
 830 static int get_fixed_group(const ExecContext *c, const char **group, gid_t *gid) {
 831         int r;
 832         const char *name;
 833
 834         assert(c);
 835
 836         if (!c->group)
 837                 return 0;
 838
 839         name = c->group;
 840         r = get_group_creds(&name, gid);
 841         if (r < 0)
 842                 return r;
 843
 844         *group = name;
 845         return 0;
 846 }
 847
 848 static int get_supplementary_groups(const ExecContext *c, const char *user,
 849                                     const char *group, gid_t gid,
 850                                     gid_t **supplementary_gids, int *ngids) {
 851         char **i;
 852         int r, k = 0;
 853         int ngroups_max;
 854         bool keep_groups = false;
 855         gid_t *groups = NULL;
 856         _cleanup_free_ gid_t *l_gids = NULL;
 857
 858         assert(c);
 859
 860         /*
 861          * If user is given, then lookup GID and supplementary groups list.
 862          * We avoid NSS lookups for gid=0. Also we have to initialize groups
 863          * here and as early as possible so we keep the list of supplementary
 864          * groups of the caller.
 865          */
 866         if (user && gid_is_valid(gid) && gid != 0) {
 867                 /* First step, initialize groups from /etc/groups */
 868                 if (initgroups(user, gid) < 0)
 869                         return -errno;
 870
 871                 keep_groups = true;
 872         }
 873
 874         if (!c->supplementary_groups)
 875                 return 0;
 876
 877         /*
 878          * If SupplementaryGroups= was passed then NGROUPS_MAX has to
 879          * be positive, otherwise fail.
 880          */
 881         errno = 0;
 882         ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
 883         if (ngroups_max <= 0) {
 884                 if (errno > 0)
 885                         return -errno;
 886                 else
 887                         return -EOPNOTSUPP; /* For all other values */
 888         }
 889
 890         l_gids = new(gid_t, ngroups_max);
 891         if (!l_gids)
 892                 return -ENOMEM;
 893
 894         if (keep_groups) {
 895                 /*
 896                  * Lookup the list of groups that the user belongs to, we
 897                  * avoid NSS lookups here too for gid=0.
 898                  */
 899                 k = ngroups_max;
 900                 if (getgrouplist(user, gid, l_gids, &k) < 0)
 901                         return -EINVAL;
 902         } else
 903                 k = 0;
 904
 905         STRV_FOREACH(i, c->supplementary_groups) {
 906                 const char *g;
 907
 908                 if (k >= ngroups_max)
 909                         return -E2BIG;
 910
 911                 g = *i;
 912                 r = get_group_creds(&g, l_gids+k);
 913                 if (r < 0)
 914                         return r;
 915
 916                 k++;
 917         }
 918
 919         /*
 920          * Sets ngids to zero to drop all supplementary groups, happens
 921          * when we are under root and SupplementaryGroups= is empty.
 922          */
 923         if (k == 0) {
 924                 *ngids = 0;
 925                 return 0;
 926         }
 927
 928         /* Otherwise get the final list of supplementary groups */
 929         groups = memdup(l_gids, sizeof(gid_t) * k);
 930         if (!groups)
 931                 return -ENOMEM;
 932
 933         *supplementary_gids = groups;
 934         *ngids = k;
 935
 936         groups = NULL;
 937
 938         return 0;
 939 }
 940
 941 static int enforce_groups(const ExecContext *context, gid_t gid,
 942                           gid_t *supplementary_gids, int ngids) {
 943         int r;
 944
 945         assert(context);
 946
 947         /* Handle SupplementaryGroups= even if it is empty */
 948         if (context->supplementary_groups) {
 949                 r = maybe_setgroups(ngids, supplementary_gids);
 950                 if (r < 0)
 951                         return r;
 952         }
 953
 954         if (gid_is_valid(gid)) {
 955                 /* Then set our gids */
 956                 if (setresgid(gid, gid, gid) < 0)
 957                         return -errno;
 958         }
 959
 960         return 0;
 961 }
 962
 963 static int enforce_user(const ExecContext *context, uid_t uid) {
 964         assert(context);
 965
 966         if (!uid_is_valid(uid))
 967                 return 0;
 968
 969         /* Sets (but doesn't look up) the uid and make sure we keep the
 970          * capabilities while doing so. */
 971
 972         if (context->capability_ambient_set != 0) {
 973
 974                 /* First step: If we need to keep capabilities but
 975                  * drop privileges we need to make sure we keep our
 976                  * caps, while we drop privileges. */
 977                 if (uid != 0) {
 978                         int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
 979
 980                         if (prctl(PR_GET_SECUREBITS) != sb)
 981                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 982                                         return -errno;
 983                 }
 984         }
 985
 986         /* Second step: actually set the uids */
 987         if (setresuid(uid, uid, uid) < 0)
 988                 return -errno;
 989
 990         /* At this point we should have all necessary capabilities but
 991            are otherwise a normal user. However, the caps might got
 992            corrupted due to the setresuid() so we need clean them up
 993            later. This is done outside of this call. */
 994
 995         return 0;
 996 }
 997
 998 #ifdef HAVE_PAM
 999
1000 static int null_conv(
1001                 int num_msg,
1002                 const struct pam_message **msg,
1003                 struct pam_response **resp,
1004                 void *appdata_ptr) {
1005
1006         /* We don't support conversations */
1007
1008         return PAM_CONV_ERR;
1009 }
1010
1011 #endif
1012
1013 static int setup_pam(
1014                 const char *name,
1015                 const char *user,
1016                 uid_t uid,
1017                 gid_t gid,
1018                 const char *tty,
1019                 char ***env,
1020                 int fds[], unsigned n_fds) {
1021
1022 #ifdef HAVE_PAM
1023
1024         static const struct pam_conv conv = {
1025                 .conv = null_conv,
1026                 .appdata_ptr = NULL
1027         };
1028
1029         _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
1030         pam_handle_t *handle = NULL;
1031         sigset_t old_ss;
1032         int pam_code = PAM_SUCCESS, r;
1033         char **nv, **e = NULL;
1034         bool close_session = false;
1035         pid_t pam_pid = 0, parent_pid;
1036         int flags = 0;
1037
1038         assert(name);
1039         assert(user);
1040         assert(env);
1041
1042         /* We set up PAM in the parent process, then fork. The child
1043          * will then stay around until killed via PR_GET_PDEATHSIG or
1044          * systemd via the cgroup logic. It will then remove the PAM
1045          * session again. The parent process will exec() the actual
1046          * daemon. We do things this way to ensure that the main PID
1047          * of the daemon is the one we initially fork()ed. */
1048
1049         r = barrier_create(&barrier);
1050         if (r < 0)
1051                 goto fail;
1052
1053         if (log_get_max_level() < LOG_DEBUG)
1054                 flags |= PAM_SILENT;
1055
1056         pam_code = pam_start(name, user, &conv, &handle);
1057         if (pam_code != PAM_SUCCESS) {
1058                 handle = NULL;
1059                 goto fail;
1060         }
1061
1062         if (tty) {
1063                 pam_code = pam_set_item(handle, PAM_TTY, tty);
1064                 if (pam_code != PAM_SUCCESS)
1065                         goto fail;
1066         }
1067
1068         STRV_FOREACH(nv, *env) {
1069                 pam_code = pam_putenv(handle, *nv);
1070                 if (pam_code != PAM_SUCCESS)
1071                         goto fail;
1072         }
1073
1074         pam_code = pam_acct_mgmt(handle, flags);
1075         if (pam_code != PAM_SUCCESS)
1076                 goto fail;
1077
1078         pam_code = pam_open_session(handle, flags);
1079         if (pam_code != PAM_SUCCESS)
1080                 goto fail;
1081
1082         close_session = true;
1083
1084         e = pam_getenvlist(handle);
1085         if (!e) {
1086                 pam_code = PAM_BUF_ERR;
1087                 goto fail;
1088         }
1089
1090         /* Block SIGTERM, so that we know that it won't get lost in
1091          * the child */
1092
1093         assert_se(sigprocmask_many(SIG_BLOCK, &old_ss, SIGTERM, -1) >= 0);
1094
1095         parent_pid = getpid();
1096
1097         pam_pid = fork();
1098         if (pam_pid < 0) {
1099                 r = -errno;
1100                 goto fail;
1101         }
1102
1103         if (pam_pid == 0) {
1104                 int sig, ret = EXIT_PAM;
1105
1106                 /* The child's job is to reset the PAM session on
1107                  * termination */
1108                 barrier_set_role(&barrier, BARRIER_CHILD);
1109
1110                 /* This string must fit in 10 chars (i.e. the length
1111                  * of "/sbin/init"), to look pretty in /bin/ps */
1112                 rename_process("(sd-pam)");
1113
1114                 /* Make sure we don't keep open the passed fds in this
1115                 child. We assume that otherwise only those fds are
1116                 open here that have been opened by PAM. */
1117                 close_many(fds, n_fds);
1118
1119                 /* Drop privileges - we don't need any to pam_close_session
1120                  * and this will make PR_SET_PDEATHSIG work in most cases.
1121                  * If this fails, ignore the error - but expect sd-pam threads
1122                  * to fail to exit normally */
1123
1124                 r = maybe_setgroups(0, NULL);
1125                 if (r < 0)
1126                         log_warning_errno(r, "Failed to setgroups() in sd-pam: %m");
1127                 if (setresgid(gid, gid, gid) < 0)
1128                         log_warning_errno(errno, "Failed to setresgid() in sd-pam: %m");
1129                 if (setresuid(uid, uid, uid) < 0)
1130                         log_warning_errno(errno, "Failed to setresuid() in sd-pam: %m");
1131
1132                 (void) ignore_signals(SIGPIPE, -1);
1133
1134                 /* Wait until our parent died. This will only work if
1135                  * the above setresuid() succeeds, otherwise the kernel
1136                  * will not allow unprivileged parents kill their privileged
1137                  * children this way. We rely on the control groups kill logic
1138                  * to do the rest for us. */
1139                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
1140                         goto child_finish;
1141
1142                 /* Tell the parent that our setup is done. This is especially
1143                  * important regarding dropping privileges. Otherwise, unit
1144                  * setup might race against our setresuid(2) call. */
1145                 barrier_place(&barrier);
1146
1147                 /* Check if our parent process might already have
1148                  * died? */
1149                 if (getppid() == parent_pid) {
1150                         sigset_t ss;
1151
1152                         assert_se(sigemptyset(&ss) >= 0);
1153                         assert_se(sigaddset(&ss, SIGTERM) >= 0);
1154
1155                         for (;;) {
1156                                 if (sigwait(&ss, &sig) < 0) {
1157                                         if (errno == EINTR)
1158                                                 continue;
1159
1160                                         goto child_finish;
1161                                 }
1162
1163                                 assert(sig == SIGTERM);
1164                                 break;
1165                         }
1166                 }
1167
1168                 /* If our parent died we'll end the session */
1169                 if (getppid() != parent_pid) {
1170                         pam_code = pam_close_session(handle, flags);
1171                         if (pam_code != PAM_SUCCESS)
1172                                 goto child_finish;
1173                 }
1174
1175                 ret = 0;
1176
1177         child_finish:
1178                 pam_end(handle, pam_code | flags);
1179                 _exit(ret);
1180         }
1181
1182         barrier_set_role(&barrier, BARRIER_PARENT);
1183
1184         /* If the child was forked off successfully it will do all the
1185          * cleanups, so forget about the handle here. */
1186         handle = NULL;
1187
1188         /* Unblock SIGTERM again in the parent */
1189         assert_se(sigprocmask(SIG_SETMASK, &old_ss, NULL) >= 0);
1190
1191         /* We close the log explicitly here, since the PAM modules
1192          * might have opened it, but we don't want this fd around. */
1193         closelog();
1194
1195         /* Synchronously wait for the child to initialize. We don't care for
1196          * errors as we cannot recover. However, warn loudly if it happens. */
1197         if (!barrier_place_and_sync(&barrier))
1198                 log_error("PAM initialization failed");
1199
1200         strv_free(*env);
1201         *env = e;
1202
1203         return 0;
1204
1205 fail:
1206         if (pam_code != PAM_SUCCESS) {
1207                 log_error("PAM failed: %s", pam_strerror(handle, pam_code));
1208                 r = -EPERM;  /* PAM errors do not map to errno */
1209         } else
1210                 log_error_errno(r, "PAM failed: %m");
1211
1212         if (handle) {
1213                 if (close_session)
1214                         pam_code = pam_close_session(handle, flags);
1215
1216                 pam_end(handle, pam_code | flags);
1217         }
1218
1219         strv_free(e);
1220         closelog();
1221
1222         return r;
1223 #else
1224         return 0;
1225 #endif
1226 }
1227
1228 static void rename_process_from_path(const char *path) {
1229         char process_name[11];
1230         const char *p;
1231         size_t l;
1232
1233         /* This resulting string must fit in 10 chars (i.e. the length
1234          * of "/sbin/init") to look pretty in /bin/ps */
1235
1236         p = basename(path);
1237         if (isempty(p)) {
1238                 rename_process("(...)");
1239                 return;
1240         }
1241
1242         l = strlen(p);
1243         if (l > 8) {
1244                 /* The end of the process name is usually more
1245                  * interesting, since the first bit might just be
1246                  * "systemd-" */
1247                 p = p + l - 8;
1248                 l = 8;
1249         }
1250
1251         process_name[0] = '(';
1252         memcpy(process_name+1, p, l);
1253         process_name[1+l] = ')';
1254         process_name[1+l+1] = 0;
1255
1256         rename_process(process_name);
1257 }
1258
1259 static bool context_has_address_families(const ExecContext *c) {
1260         assert(c);
1261
1262         return c->address_families_whitelist ||
1263                 !set_isempty(c->address_families);
1264 }
1265
1266 static bool context_has_syscall_filters(const ExecContext *c) {
1267         assert(c);
1268
1269         return c->syscall_whitelist ||
1270                 !set_isempty(c->syscall_filter);
1271 }
1272
1273 static bool context_has_no_new_privileges(const ExecContext *c) {
1274         assert(c);
1275
1276         if (c->no_new_privileges)
1277                 return true;
1278
1279         if (have_effective_cap(CAP_SYS_ADMIN)) /* if we are privileged, we don't need NNP */
1280                 return false;
1281
1282         /* We need NNP if we have any form of seccomp and are unprivileged */
1283         return context_has_address_families(c) ||
1284                 c->memory_deny_write_execute ||
1285                 c->restrict_realtime ||
1286                 exec_context_restrict_namespaces_set(c) ||
1287                 c->protect_kernel_tunables ||
1288                 c->protect_kernel_modules ||
1289                 c->private_devices ||
1290                 context_has_syscall_filters(c) ||
1291                 !set_isempty(c->syscall_archs);
1292 }
1293
1294 #ifdef HAVE_SECCOMP
1295
1296 static bool skip_seccomp_unavailable(const Unit* u, const char* msg) {
1297
1298         if (is_seccomp_available())
1299                 return false;
1300
1301         log_open();
1302         log_unit_debug(u, "SECCOMP features not detected in the kernel, skipping %s", msg);
1303         log_close();
1304         return true;
1305 }
1306
1307 static int apply_syscall_filter(const Unit* u, const ExecContext *c) {
1308         uint32_t negative_action, default_action, action;
1309
1310         assert(u);
1311         assert(c);
1312
1313         if (!context_has_syscall_filters(c))
1314                 return 0;
1315
1316         if (skip_seccomp_unavailable(u, "SystemCallFilter="))
1317                 return 0;
1318
1319         negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
1320
1321         if (c->syscall_whitelist) {
1322                 default_action = negative_action;
1323                 action = SCMP_ACT_ALLOW;
1324         } else {
1325                 default_action = SCMP_ACT_ALLOW;
1326                 action = negative_action;
1327         }
1328
1329         return seccomp_load_syscall_filter_set_raw(default_action, c->syscall_filter, action);
1330 }
1331
1332 static int apply_syscall_archs(const Unit *u, const ExecContext *c) {
1333         assert(u);
1334         assert(c);
1335
1336         if (set_isempty(c->syscall_archs))
1337                 return 0;
1338
1339         if (skip_seccomp_unavailable(u, "SystemCallArchitectures="))
1340                 return 0;
1341
1342         return seccomp_restrict_archs(c->syscall_archs);
1343 }
1344
1345 static int apply_address_families(const Unit* u, const ExecContext *c) {
1346         assert(u);
1347         assert(c);
1348
1349         if (!context_has_address_families(c))
1350                 return 0;
1351
1352         if (skip_seccomp_unavailable(u, "RestrictAddressFamilies="))
1353                 return 0;
1354
1355         return seccomp_restrict_address_families(c->address_families, c->address_families_whitelist);
1356 }
1357
1358 static int apply_memory_deny_write_execute(const Unit* u, const ExecContext *c) {
1359         assert(u);
1360         assert(c);
1361
1362         if (!c->memory_deny_write_execute)
1363                 return 0;
1364
1365         if (skip_seccomp_unavailable(u, "MemoryDenyWriteExecute="))
1366                 return 0;
1367
1368         return seccomp_memory_deny_write_execute();
1369 }
1370
1371 static int apply_restrict_realtime(const Unit* u, const ExecContext *c) {
1372         assert(u);
1373         assert(c);
1374
1375         if (!c->restrict_realtime)
1376                 return 0;
1377
1378         if (skip_seccomp_unavailable(u, "RestrictRealtime="))
1379                 return 0;
1380
1381         return seccomp_restrict_realtime();
1382 }
1383
1384 static int apply_protect_sysctl(const Unit *u, const ExecContext *c) {
1385         assert(u);
1386         assert(c);
1387
1388         /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1389          * let's protect even those systems where this is left on in the kernel. */
1390
1391         if (!c->protect_kernel_tunables)
1392                 return 0;
1393
1394         if (skip_seccomp_unavailable(u, "ProtectKernelTunables="))
1395                 return 0;
1396
1397         return seccomp_protect_sysctl();
1398 }
1399
1400 static int apply_protect_kernel_modules(const Unit *u, const ExecContext *c) {
1401         assert(u);
1402         assert(c);
1403
1404         /* Turn off module syscalls on ProtectKernelModules=yes */
1405
1406         if (!c->protect_kernel_modules)
1407                 return 0;
1408
1409         if (skip_seccomp_unavailable(u, "ProtectKernelModules="))
1410                 return 0;
1411
1412         return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_MODULE, SCMP_ACT_ERRNO(EPERM));
1413 }
1414
1415 static int apply_private_devices(const Unit *u, const ExecContext *c) {
1416         assert(u);
1417         assert(c);
1418
1419         /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1420
1421         if (!c->private_devices)
1422                 return 0;
1423
1424         if (skip_seccomp_unavailable(u, "PrivateDevices="))
1425                 return 0;
1426
1427         return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_RAW_IO, SCMP_ACT_ERRNO(EPERM));
1428 }
1429
1430 static int apply_restrict_namespaces(Unit *u, const ExecContext *c) {
1431         assert(u);
1432         assert(c);
1433
1434         if (!exec_context_restrict_namespaces_set(c))
1435                 return 0;
1436
1437         if (skip_seccomp_unavailable(u, "RestrictNamespaces="))
1438                 return 0;
1439
1440         return seccomp_restrict_namespaces(c->restrict_namespaces);
1441 }
1442
1443 #endif
1444
1445 static void do_idle_pipe_dance(int idle_pipe[4]) {
1446         assert(idle_pipe);
1447
1448         idle_pipe[1] = safe_close(idle_pipe[1]);
1449         idle_pipe[2] = safe_close(idle_pipe[2]);
1450
1451         if (idle_pipe[0] >= 0) {
1452                 int r;
1453
1454                 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1455
1456                 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
1457                         ssize_t n;
1458
1459                         /* Signal systemd that we are bored and want to continue. */
1460                         n = write(idle_pipe[3], "x", 1);
1461                         if (n > 0)
1462                                 /* Wait for systemd to react to the signal above. */
1463                                 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
1464                 }
1465
1466                 idle_pipe[0] = safe_close(idle_pipe[0]);
1467
1468         }
1469
1470         idle_pipe[3] = safe_close(idle_pipe[3]);
1471 }
1472
1473 static int build_environment(
1474                 Unit *u,
1475                 const ExecContext *c,
1476                 const ExecParameters *p,
1477                 unsigned n_fds,
1478                 const char *home,
1479                 const char *username,
1480                 const char *shell,
1481                 dev_t journal_stream_dev,
1482                 ino_t journal_stream_ino,
1483                 char ***ret) {
1484
1485         _cleanup_strv_free_ char **our_env = NULL;
1486         unsigned n_env = 0;
1487         char *x;
1488
1489         assert(u);
1490         assert(c);
1491         assert(ret);
1492
1493         our_env = new0(char*, 14);
1494         if (!our_env)
1495                 return -ENOMEM;
1496
1497         if (n_fds > 0) {
1498                 _cleanup_free_ char *joined = NULL;
1499
1500                 if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
1501                         return -ENOMEM;
1502                 our_env[n_env++] = x;
1503
1504                 if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
1505                         return -ENOMEM;
1506                 our_env[n_env++] = x;
1507
1508                 joined = strv_join(p->fd_names, ":");
1509                 if (!joined)
1510                         return -ENOMEM;
1511
1512                 x = strjoin("LISTEN_FDNAMES=", joined);
1513                 if (!x)
1514                         return -ENOMEM;
1515                 our_env[n_env++] = x;
1516         }
1517
1518         if ((p->flags & EXEC_SET_WATCHDOG) && p->watchdog_usec > 0) {
1519                 if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
1520                         return -ENOMEM;
1521                 our_env[n_env++] = x;
1522
1523                 if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, p->watchdog_usec) < 0)
1524                         return -ENOMEM;
1525                 our_env[n_env++] = x;
1526         }
1527
1528         /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1529          * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1530          * check the database directly. */
1531         if (unit_has_name(u, SPECIAL_DBUS_SERVICE)) {
1532                 x = strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1533                 if (!x)
1534                         return -ENOMEM;
1535                 our_env[n_env++] = x;
1536         }
1537
1538         if (home) {
1539                 x = strappend("HOME=", home);
1540                 if (!x)
1541                         return -ENOMEM;
1542                 our_env[n_env++] = x;
1543         }
1544
1545         if (username) {
1546                 x = strappend("LOGNAME=", username);
1547                 if (!x)
1548                         return -ENOMEM;
1549                 our_env[n_env++] = x;
1550
1551                 x = strappend("USER=", username);
1552                 if (!x)
1553                         return -ENOMEM;
1554                 our_env[n_env++] = x;
1555         }
1556
1557         if (shell) {
1558                 x = strappend("SHELL=", shell);
1559                 if (!x)
1560                         return -ENOMEM;
1561                 our_env[n_env++] = x;
1562         }
1563
1564         if (!sd_id128_is_null(u->invocation_id)) {
1565                 if (asprintf(&x, "INVOCATION_ID=" SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(u->invocation_id)) < 0)
1566                         return -ENOMEM;
1567
1568                 our_env[n_env++] = x;
1569         }
1570
1571         if (exec_context_needs_term(c)) {
1572                 const char *tty_path, *term = NULL;
1573
1574                 tty_path = exec_context_tty_path(c);
1575
1576                 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1577                  * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1578                  * passes to PID 1 ends up all the way in the console login shown. */
1579
1580                 if (path_equal(tty_path, "/dev/console") && getppid() == 1)
1581                         term = getenv("TERM");
1582                 if (!term)
1583                         term = default_term_for_tty(tty_path);
1584
1585                 x = strappend("TERM=", term);
1586                 if (!x)
1587                         return -ENOMEM;
1588                 our_env[n_env++] = x;
1589         }
1590
1591         if (journal_stream_dev != 0 && journal_stream_ino != 0) {
1592                 if (asprintf(&x, "JOURNAL_STREAM=" DEV_FMT ":" INO_FMT, journal_stream_dev, journal_stream_ino) < 0)
1593                         return -ENOMEM;
1594
1595                 our_env[n_env++] = x;
1596         }
1597
1598         our_env[n_env++] = NULL;
1599         assert(n_env <= 12);
1600
1601         *ret = our_env;
1602         our_env = NULL;
1603
1604         return 0;
1605 }
1606
1607 static int build_pass_environment(const ExecContext *c, char ***ret) {
1608         _cleanup_strv_free_ char **pass_env = NULL;
1609         size_t n_env = 0, n_bufsize = 0;
1610         char **i;
1611
1612         STRV_FOREACH(i, c->pass_environment) {
1613                 _cleanup_free_ char *x = NULL;
1614                 char *v;
1615
1616                 v = getenv(*i);
1617                 if (!v)
1618                         continue;
1619                 x = strjoin(*i, "=", v);
1620                 if (!x)
1621                         return -ENOMEM;
1622                 if (!GREEDY_REALLOC(pass_env, n_bufsize, n_env + 2))
1623                         return -ENOMEM;
1624                 pass_env[n_env++] = x;
1625                 pass_env[n_env] = NULL;
1626                 x = NULL;
1627         }
1628
1629         *ret = pass_env;
1630         pass_env = NULL;
1631
1632         return 0;
1633 }
1634
1635 static bool exec_needs_mount_namespace(
1636                 const ExecContext *context,
1637                 const ExecParameters *params,
1638                 ExecRuntime *runtime) {
1639
1640         assert(context);
1641         assert(params);
1642
1643         if (!strv_isempty(context->read_write_paths) ||
1644             !strv_isempty(context->read_only_paths) ||
1645             !strv_isempty(context->inaccessible_paths))
1646                 return true;
1647
1648         if (context->n_bind_mounts > 0)
1649                 return true;
1650
1651         if (context->mount_flags != 0)
1652                 return true;
1653
1654         if (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir))
1655                 return true;
1656
1657         if (context->private_devices ||
1658             context->protect_system != PROTECT_SYSTEM_NO ||
1659             context->protect_home != PROTECT_HOME_NO ||
1660             context->protect_kernel_tunables ||
1661             context->protect_kernel_modules ||
1662             context->protect_control_groups)
1663                 return true;
1664
1665         return false;
1666 }
1667
1668 static int setup_private_users(uid_t uid, gid_t gid) {
1669         _cleanup_free_ char *uid_map = NULL, *gid_map = NULL;
1670         _cleanup_close_pair_ int errno_pipe[2] = { -1, -1 };
1671         _cleanup_close_ int unshare_ready_fd = -1;
1672         _cleanup_(sigkill_waitp) pid_t pid = 0;
1673         uint64_t c = 1;
1674         siginfo_t si;
1675         ssize_t n;
1676         int r;
1677
1678         /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1679          * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1680          * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1681          * which waits for the parent to create the new user namespace while staying in the original namespace. The
1682          * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1683          * continues execution normally. */
1684
1685         if (uid != 0 && uid_is_valid(uid)) {
1686                 r = asprintf(&uid_map,
1687                              "0 0 1\n"                      /* Map root → root */
1688                              UID_FMT " " UID_FMT " 1\n",    /* Map $UID → $UID */
1689                              uid, uid);
1690                 if (r < 0)
1691                         return -ENOMEM;
1692         } else {
1693                 uid_map = strdup("0 0 1\n");            /* The case where the above is the same */
1694                 if (!uid_map)
1695                         return -ENOMEM;
1696         }
1697
1698         if (gid != 0 && gid_is_valid(gid)) {
1699                 r = asprintf(&gid_map,
1700                              "0 0 1\n"                      /* Map root → root */
1701                              GID_FMT " " GID_FMT " 1\n",    /* Map $GID → $GID */
1702                              gid, gid);
1703                 if (r < 0)
1704                         return -ENOMEM;
1705         } else {
1706                 gid_map = strdup("0 0 1\n");            /* The case where the above is the same */
1707                 if (!gid_map)
1708                         return -ENOMEM;
1709         }
1710
1711         /* Create a communication channel so that the parent can tell the child when it finished creating the user
1712          * namespace. */
1713         unshare_ready_fd = eventfd(0, EFD_CLOEXEC);
1714         if (unshare_ready_fd < 0)
1715                 return -errno;
1716
1717         /* Create a communication channel so that the child can tell the parent a proper error code in case it
1718          * failed. */
1719         if (pipe2(errno_pipe, O_CLOEXEC) < 0)
1720                 return -errno;
1721
1722         pid = fork();
1723         if (pid < 0)
1724                 return -errno;
1725
1726         if (pid == 0) {
1727                 _cleanup_close_ int fd = -1;
1728                 const char *a;
1729                 pid_t ppid;
1730
1731                 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1732                  * here, after the parent opened its own user namespace. */
1733
1734                 ppid = getppid();
1735                 errno_pipe[0] = safe_close(errno_pipe[0]);
1736
1737                 /* Wait until the parent unshared the user namespace */
1738                 if (read(unshare_ready_fd, &c, sizeof(c)) < 0) {
1739                         r = -errno;
1740                         goto child_fail;
1741                 }
1742
1743                 /* Disable the setgroups() system call in the child user namespace, for good. */
1744                 a = procfs_file_alloca(ppid, "setgroups");
1745                 fd = open(a, O_WRONLY|O_CLOEXEC);
1746                 if (fd < 0) {
1747                         if (errno != ENOENT) {
1748                                 r = -errno;
1749                                 goto child_fail;
1750                         }
1751
1752                         /* If the file is missing the kernel is too old, let's continue anyway. */
1753                 } else {
1754                         if (write(fd, "deny\n", 5) < 0) {
1755                                 r = -errno;
1756                                 goto child_fail;
1757                         }
1758
1759                         fd = safe_close(fd);
1760                 }
1761
1762                 /* First write the GID map */
1763                 a = procfs_file_alloca(ppid, "gid_map");
1764                 fd = open(a, O_WRONLY|O_CLOEXEC);
1765                 if (fd < 0) {
1766                         r = -errno;
1767                         goto child_fail;
1768                 }
1769                 if (write(fd, gid_map, strlen(gid_map)) < 0) {
1770                         r = -errno;
1771                         goto child_fail;
1772                 }
1773                 fd = safe_close(fd);
1774
1775                 /* The write the UID map */
1776                 a = procfs_file_alloca(ppid, "uid_map");
1777                 fd = open(a, O_WRONLY|O_CLOEXEC);
1778                 if (fd < 0) {
1779                         r = -errno;
1780                         goto child_fail;
1781                 }
1782                 if (write(fd, uid_map, strlen(uid_map)) < 0) {
1783                         r = -errno;
1784                         goto child_fail;
1785                 }
1786
1787                 _exit(EXIT_SUCCESS);
1788
1789         child_fail:
1790                 (void) write(errno_pipe[1], &r, sizeof(r));
1791                 _exit(EXIT_FAILURE);
1792         }
1793
1794         errno_pipe[1] = safe_close(errno_pipe[1]);
1795
1796         if (unshare(CLONE_NEWUSER) < 0)
1797                 return -errno;
1798
1799         /* Let the child know that the namespace is ready now */
1800         if (write(unshare_ready_fd, &c, sizeof(c)) < 0)
1801                 return -errno;
1802
1803         /* Try to read an error code from the child */
1804         n = read(errno_pipe[0], &r, sizeof(r));
1805         if (n < 0)
1806                 return -errno;
1807         if (n == sizeof(r)) { /* an error code was sent to us */
1808                 if (r < 0)
1809                         return r;
1810                 return -EIO;
1811         }
1812         if (n != 0) /* on success we should have read 0 bytes */
1813                 return -EIO;
1814
1815         r = wait_for_terminate(pid, &si);
1816         if (r < 0)
1817                 return r;
1818         pid = 0;
1819
1820         /* If something strange happened with the child, let's consider this fatal, too */
1821         if (si.si_code != CLD_EXITED || si.si_status != 0)
1822                 return -EIO;
1823
1824         return 0;
1825 }
1826
1827 static int setup_runtime_directory(
1828                 const ExecContext *context,
1829                 const ExecParameters *params,
1830                 uid_t uid,
1831                 gid_t gid) {
1832
1833         char **rt;
1834         int r;
1835
1836         assert(context);
1837         assert(params);
1838
1839         STRV_FOREACH(rt, context->runtime_directory) {
1840                 _cleanup_free_ char *p;
1841
1842                 p = strjoin(params->runtime_prefix, "/", *rt);
1843                 if (!p)
1844                         return -ENOMEM;
1845
1846                 r = mkdir_p_label(p, context->runtime_directory_mode);
1847                 if (r < 0)
1848                         return r;
1849
1850                 r = chmod_and_chown(p, context->runtime_directory_mode, uid, gid);
1851                 if (r < 0)
1852                         return r;
1853         }
1854
1855         return 0;
1856 }
1857
1858 static int setup_smack(
1859                 const ExecContext *context,
1860                 const ExecCommand *command) {
1861
1862 #ifdef HAVE_SMACK
1863         int r;
1864
1865         assert(context);
1866         assert(command);
1867
1868         if (!mac_smack_use())
1869                 return 0;
1870
1871         if (context->smack_process_label) {
1872                 r = mac_smack_apply_pid(0, context->smack_process_label);
1873                 if (r < 0)
1874                         return r;
1875         }
1876 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1877         else {
1878                 _cleanup_free_ char *exec_label = NULL;
1879
1880                 r = mac_smack_read(command->path, SMACK_ATTR_EXEC, &exec_label);
1881                 if (r < 0 && r != -ENODATA && r != -EOPNOTSUPP)
1882                         return r;
1883
1884                 r = mac_smack_apply_pid(0, exec_label ? : SMACK_DEFAULT_PROCESS_LABEL);
1885                 if (r < 0)
1886                         return r;
1887         }
1888 #endif
1889 #endif
1890
1891         return 0;
1892 }
1893
1894 static int compile_read_write_paths(
1895                 const ExecContext *context,
1896                 const ExecParameters *params,
1897                 char ***ret) {
1898
1899         _cleanup_strv_free_ char **l = NULL;
1900         char **rt;
1901
1902         /* Compile the list of writable paths. This is the combination of
1903          * the explicitly configured paths, plus all runtime directories. */
1904
1905         if (strv_isempty(context->read_write_paths) &&
1906             strv_isempty(context->runtime_directory)) {
1907                 *ret = NULL; /* NOP if neither is set */
1908                 return 0;
1909         }
1910
1911         l = strv_copy(context->read_write_paths);
1912         if (!l)
1913                 return -ENOMEM;
1914
1915         STRV_FOREACH(rt, context->runtime_directory) {
1916                 char *s;
1917
1918                 s = strjoin(params->runtime_prefix, "/", *rt);
1919                 if (!s)
1920                         return -ENOMEM;
1921
1922                 if (strv_consume(&l, s) < 0)
1923                         return -ENOMEM;
1924         }
1925
1926         *ret = l;
1927         l = NULL;
1928
1929         return 0;
1930 }
1931
1932 static int apply_mount_namespace(Unit *u, const ExecContext *context,
1933                                  const ExecParameters *params,
1934                                  ExecRuntime *runtime) {
1935         int r;
1936         _cleanup_strv_free_ char **rw = NULL;
1937         char *tmp = NULL, *var = NULL;
1938         const char *root_dir = NULL;
1939         NameSpaceInfo ns_info = {
1940                 .ignore_protect_paths = false,
1941                 .private_dev = context->private_devices,
1942                 .protect_control_groups = context->protect_control_groups,
1943                 .protect_kernel_tunables = context->protect_kernel_tunables,
1944                 .protect_kernel_modules = context->protect_kernel_modules,
1945         };
1946
1947         assert(context);
1948
1949         /* The runtime struct only contains the parent of the private /tmp,
1950          * which is non-accessible to world users. Inside of it there's a /tmp
1951          * that is sticky, and that's the one we want to use here. */
1952
1953         if (context->private_tmp && runtime) {
1954                 if (runtime->tmp_dir)
1955                         tmp = strjoina(runtime->tmp_dir, "/tmp");
1956                 if (runtime->var_tmp_dir)
1957                         var = strjoina(runtime->var_tmp_dir, "/tmp");
1958         }
1959
1960         r = compile_read_write_paths(context, params, &rw);
1961         if (r < 0)
1962                 return r;
1963
1964         if (params->flags & EXEC_APPLY_CHROOT)
1965                 root_dir = context->root_directory;
1966
1967         /*
1968          * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
1969          * sandbox info, otherwise enforce it, don't ignore protected paths and
1970          * fail if we are enable to apply the sandbox inside the mount namespace.
1971          */
1972         if (!context->dynamic_user && root_dir)
1973                 ns_info.ignore_protect_paths = true;
1974
1975         r = setup_namespace(root_dir, &ns_info, rw,
1976                             context->read_only_paths,
1977                             context->inaccessible_paths,
1978                             context->bind_mounts,
1979                             context->n_bind_mounts,
1980                             tmp,
1981                             var,
1982                             context->protect_home,
1983                             context->protect_system,
1984                             context->mount_flags);
1985
1986         /* If we couldn't set up the namespace this is probably due to a
1987          * missing capability. In this case, silently proceeed. */
1988         if (IN_SET(r, -EPERM, -EACCES)) {
1989                 log_open();
1990                 log_unit_debug_errno(u, r, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
1991                 log_close();
1992                 r = 0;
1993         }
1994
1995         return r;
1996 }
1997
1998 static int apply_working_directory(const ExecContext *context,
1999                                    const ExecParameters *params,
2000                                    const char *home,
2001                                    const bool needs_mount_ns) {
2002         const char *d;
2003         const char *wd;
2004
2005         assert(context);
2006
2007         if (context->working_directory_home)
2008                 wd = home;
2009         else if (context->working_directory)
2010                 wd = context->working_directory;
2011         else
2012                 wd = "/";
2013
2014         if (params->flags & EXEC_APPLY_CHROOT) {
2015                 if (!needs_mount_ns && context->root_directory)
2016                         if (chroot(context->root_directory) < 0)
2017                                 return -errno;
2018
2019                 d = wd;
2020         } else
2021                 d = strjoina(strempty(context->root_directory), "/", strempty(wd));
2022
2023         if (chdir(d) < 0 && !context->working_directory_missing_ok)
2024                 return -errno;
2025
2026         return 0;
2027 }
2028
2029 static int setup_keyring(Unit *u, const ExecParameters *p, uid_t uid, gid_t gid) {
2030         key_serial_t keyring;
2031
2032         assert(u);
2033         assert(p);
2034
2035         /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2036          * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2037          * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2038          * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2039          * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2040          * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2041
2042         if (!(p->flags & EXEC_NEW_KEYRING))
2043                 return 0;
2044
2045         keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
2046         if (keyring == -1) {
2047                 if (errno == ENOSYS)
2048                         log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
2049                 else if (IN_SET(errno, EACCES, EPERM))
2050                         log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
2051                 else if (errno == EDQUOT)
2052                         log_debug_errno(errno, "Out of kernel keyrings to allocate, ignoring.");
2053                 else
2054                         return log_error_errno(errno, "Setting up kernel keyring failed: %m");
2055
2056                 return 0;
2057         }
2058
2059         /* Populate they keyring with the invocation ID by default. */
2060         if (!sd_id128_is_null(u->invocation_id)) {
2061                 key_serial_t key;
2062
2063                 key = add_key("user", "invocation_id", &u->invocation_id, sizeof(u->invocation_id), KEY_SPEC_SESSION_KEYRING);
2064                 if (key == -1)
2065                         log_debug_errno(errno, "Failed to add invocation ID to keyring, ignoring: %m");
2066                 else {
2067                         if (keyctl(KEYCTL_SETPERM, key,
2068                                    KEY_POS_VIEW|KEY_POS_READ|KEY_POS_SEARCH|
2069                                    KEY_USR_VIEW|KEY_USR_READ|KEY_USR_SEARCH, 0, 0) < 0)
2070                                 return log_error_errno(errno, "Failed to restrict invocation ID permission: %m");
2071                 }
2072         }
2073
2074         /* And now, make the keyring owned by the service's user */
2075         if (uid_is_valid(uid) || gid_is_valid(gid))
2076                 if (keyctl(KEYCTL_CHOWN, keyring, uid, gid, 0) < 0)
2077                         return log_error_errno(errno, "Failed to change ownership of session keyring: %m");
2078
2079         return 0;
2080 }
2081
2082 static void append_socket_pair(int *array, unsigned *n, int pair[2]) {
2083         assert(array);
2084         assert(n);
2085
2086         if (!pair)
2087                 return;
2088
2089         if (pair[0] >= 0)
2090                 array[(*n)++] = pair[0];
2091         if (pair[1] >= 0)
2092                 array[(*n)++] = pair[1];
2093 }
2094
2095 static int close_remaining_fds(
2096                 const ExecParameters *params,
2097                 ExecRuntime *runtime,
2098                 DynamicCreds *dcreds,
2099                 int user_lookup_fd,
2100                 int socket_fd,
2101                 int *fds, unsigned n_fds) {
2102
2103         unsigned n_dont_close = 0;
2104         int dont_close[n_fds + 12];
2105
2106         assert(params);
2107
2108         if (params->stdin_fd >= 0)
2109                 dont_close[n_dont_close++] = params->stdin_fd;
2110         if (params->stdout_fd >= 0)
2111                 dont_close[n_dont_close++] = params->stdout_fd;
2112         if (params->stderr_fd >= 0)
2113                 dont_close[n_dont_close++] = params->stderr_fd;
2114
2115         if (socket_fd >= 0)
2116                 dont_close[n_dont_close++] = socket_fd;
2117         if (n_fds > 0) {
2118                 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
2119                 n_dont_close += n_fds;
2120         }
2121
2122         if (runtime)
2123                 append_socket_pair(dont_close, &n_dont_close, runtime->netns_storage_socket);
2124
2125         if (dcreds) {
2126                 if (dcreds->user)
2127                         append_socket_pair(dont_close, &n_dont_close, dcreds->user->storage_socket);
2128                 if (dcreds->group)
2129                         append_socket_pair(dont_close, &n_dont_close, dcreds->group->storage_socket);
2130         }
2131
2132         if (user_lookup_fd >= 0)
2133                 dont_close[n_dont_close++] = user_lookup_fd;
2134
2135         return close_all_fds(dont_close, n_dont_close);
2136 }
2137
2138 static int send_user_lookup(
2139                 Unit *unit,
2140                 int user_lookup_fd,
2141                 uid_t uid,
2142                 gid_t gid) {
2143
2144         assert(unit);
2145
2146         /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2147          * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2148          * specified. */
2149
2150         if (user_lookup_fd < 0)
2151                 return 0;
2152
2153         if (!uid_is_valid(uid) && !gid_is_valid(gid))
2154                 return 0;
2155
2156         if (writev(user_lookup_fd,
2157                (struct iovec[]) {
2158                            { .iov_base = &uid, .iov_len = sizeof(uid) },
2159                            { .iov_base = &gid, .iov_len = sizeof(gid) },
2160                            { .iov_base = unit->id, .iov_len = strlen(unit->id) }}, 3) < 0)
2161                 return -errno;
2162
2163         return 0;
2164 }
2165
2166 static int exec_child(
2167                 Unit *unit,
2168                 ExecCommand *command,
2169                 const ExecContext *context,
2170                 const ExecParameters *params,
2171                 ExecRuntime *runtime,
2172                 DynamicCreds *dcreds,
2173                 char **argv,
2174                 int socket_fd,
2175                 int named_iofds[3],
2176                 int *fds, unsigned n_fds,
2177                 char **files_env,
2178                 int user_lookup_fd,
2179                 int *exit_status,
2180                 char **error_message) {
2181
2182         _cleanup_strv_free_ char **our_env = NULL, **pass_env = NULL, **accum_env = NULL, **final_argv = NULL;
2183         _cleanup_free_ char *mac_selinux_context_net = NULL;
2184         _cleanup_free_ gid_t *supplementary_gids = NULL;
2185         const char *username = NULL, *groupname = NULL;
2186         const char *home = NULL, *shell = NULL;
2187         dev_t journal_stream_dev = 0;
2188         ino_t journal_stream_ino = 0;
2189         bool needs_mount_namespace;
2190         uid_t uid = UID_INVALID;
2191         gid_t gid = GID_INVALID;
2192         int i, r, ngids = 0;
2193
2194         assert(unit);
2195         assert(command);
2196         assert(context);
2197         assert(params);
2198         assert(exit_status);
2199         assert(error_message);
2200         /* We don't always set error_message, hence it must be initialized */
2201         assert(*error_message == NULL);
2202
2203         rename_process_from_path(command->path);
2204
2205         /* We reset exactly these signals, since they are the
2206          * only ones we set to SIG_IGN in the main daemon. All
2207          * others we leave untouched because we set them to
2208          * SIG_DFL or a valid handler initially, both of which
2209          * will be demoted to SIG_DFL. */
2210         (void) default_signals(SIGNALS_CRASH_HANDLER,
2211                                SIGNALS_IGNORE, -1);
2212
2213         if (context->ignore_sigpipe)
2214                 (void) ignore_signals(SIGPIPE, -1);
2215
2216         r = reset_signal_mask();
2217         if (r < 0) {
2218                 *exit_status = EXIT_SIGNAL_MASK;
2219                 *error_message = strdup("Failed to reset signal mask");
2220                 /* If strdup fails, here and below, we will just print the generic error message. */
2221                 return r;
2222         }
2223
2224         if (params->idle_pipe)
2225                 do_idle_pipe_dance(params->idle_pipe);
2226
2227         /* Close sockets very early to make sure we don't
2228          * block init reexecution because it cannot bind its
2229          * sockets */
2230
2231         log_forget_fds();
2232
2233         r = close_remaining_fds(params, runtime, dcreds, user_lookup_fd, socket_fd, fds, n_fds);
2234         if (r < 0) {
2235                 *exit_status = EXIT_FDS;
2236                 *error_message = strdup("Failed to close remaining fds");
2237                 return r;
2238         }
2239
2240         if (!context->same_pgrp)
2241                 if (setsid() < 0) {
2242                         *exit_status = EXIT_SETSID;
2243                         return -errno;
2244                 }
2245
2246         exec_context_tty_reset(context, params);
2247
2248         if (unit_shall_confirm_spawn(unit)) {
2249                 const char *vc = params->confirm_spawn;
2250                 _cleanup_free_ char *cmdline = NULL;
2251
2252                 cmdline = exec_command_line(argv);
2253                 if (!cmdline) {
2254                         *exit_status = EXIT_CONFIRM;
2255                         return -ENOMEM;
2256                 }
2257
2258                 r = ask_for_confirmation(vc, unit, cmdline);
2259                 if (r != CONFIRM_EXECUTE) {
2260                         if (r == CONFIRM_PRETEND_SUCCESS) {
2261                                 *exit_status = EXIT_SUCCESS;
2262                                 return 0;
2263                         }
2264                         *exit_status = EXIT_CONFIRM;
2265                         *error_message = strdup("Execution cancelled");
2266                         return -ECANCELED;
2267                 }
2268         }
2269
2270         if (context->dynamic_user && dcreds) {
2271
2272                 /* Make sure we bypass our own NSS module for any NSS checks */
2273                 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2274                         *exit_status = EXIT_USER;
2275                         *error_message = strdup("Failed to update environment");
2276                         return -errno;
2277                 }
2278
2279                 r = dynamic_creds_realize(dcreds, &uid, &gid);
2280                 if (r < 0) {
2281                         *exit_status = EXIT_USER;
2282                         *error_message = strdup("Failed to update dynamic user credentials");
2283                         return r;
2284                 }
2285
2286                 if (!uid_is_valid(uid)) {
2287                         *exit_status = EXIT_USER;
2288                         (void) asprintf(error_message, "UID validation failed for \""UID_FMT"\"", uid);
2289                         /* If asprintf fails, here and below, we will just print the generic error message. */
2290                         return -ESRCH;
2291                 }
2292
2293                 if (!gid_is_valid(gid)) {
2294                         *exit_status = EXIT_USER;
2295                         (void) asprintf(error_message, "GID validation failed for \""GID_FMT"\"", gid);
2296                         return -ESRCH;
2297                 }
2298
2299                 if (dcreds->user)
2300                         username = dcreds->user->name;
2301
2302         } else {
2303                 r = get_fixed_user(context, &username, &uid, &gid, &home, &shell);
2304                 if (r < 0) {
2305                         *exit_status = EXIT_USER;
2306                         *error_message = strdup("Failed to determine user credentials");
2307                         return r;
2308                 }
2309
2310                 r = get_fixed_group(context, &groupname, &gid);
2311                 if (r < 0) {
2312                         *exit_status = EXIT_GROUP;
2313                         *error_message = strdup("Failed to determine group credentials");
2314                         return r;
2315                 }
2316         }
2317
2318         /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2319         r = get_supplementary_groups(context, username, groupname, gid,
2320                                      &supplementary_gids, &ngids);
2321         if (r < 0) {
2322                 *exit_status = EXIT_GROUP;
2323                 *error_message = strdup("Failed to determine supplementary groups");
2324                 return r;
2325         }
2326
2327         r = send_user_lookup(unit, user_lookup_fd, uid, gid);
2328         if (r < 0) {
2329                 *exit_status = EXIT_USER;
2330                 *error_message = strdup("Failed to send user credentials to PID1");
2331                 return r;
2332         }
2333
2334         user_lookup_fd = safe_close(user_lookup_fd);
2335
2336         /* If a socket is connected to STDIN/STDOUT/STDERR, we
2337          * must sure to drop O_NONBLOCK */
2338         if (socket_fd >= 0)
2339                 (void) fd_nonblock(socket_fd, false);
2340
2341         r = setup_input(context, params, socket_fd, named_iofds);
2342         if (r < 0) {
2343                 *exit_status = EXIT_STDIN;
2344                 *error_message = strdup("Failed to set up stdin");
2345                 return r;
2346         }
2347
2348         r = setup_output(unit, context, params, STDOUT_FILENO, socket_fd, named_iofds, basename(command->path), uid, gid, &journal_stream_dev, &journal_stream_ino);
2349         if (r < 0) {
2350                 *exit_status = EXIT_STDOUT;
2351                 *error_message = strdup("Failed to set up stdout");
2352                 return r;
2353         }
2354
2355         r = setup_output(unit, context, params, STDERR_FILENO, socket_fd, named_iofds, basename(command->path), uid, gid, &journal_stream_dev, &journal_stream_ino);
2356         if (r < 0) {
2357                 *exit_status = EXIT_STDERR;
2358                 *error_message = strdup("Failed to set up stderr");
2359                 return r;
2360         }
2361
2362         if (params->cgroup_path) {
2363                 r = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0, NULL, NULL);
2364                 if (r < 0) {
2365                         *exit_status = EXIT_CGROUP;
2366                         (void) asprintf(error_message, "Failed to attach to cgroup %s", params->cgroup_path);
2367                         return r;
2368                 }
2369         }
2370
2371         if (context->oom_score_adjust_set) {
2372                 char t[DECIMAL_STR_MAX(context->oom_score_adjust)];
2373
2374                 /* When we can't make this change due to EPERM, then
2375                  * let's silently skip over it. User namespaces
2376                  * prohibit write access to this file, and we
2377                  * shouldn't trip up over that. */
2378
2379                 sprintf(t, "%i", context->oom_score_adjust);
2380                 r = write_string_file("/proc/self/oom_score_adj", t, 0);
2381                 if (r == -EPERM || r == -EACCES) {
2382                         log_open();
2383                         log_unit_debug_errno(unit, r, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2384                         log_close();
2385                 } else if (r < 0) {
2386                         *exit_status = EXIT_OOM_ADJUST;
2387                         *error_message = strdup("Failed to write /proc/self/oom_score_adj");
2388                         return -errno;
2389                 }
2390         }
2391
2392         if (context->nice_set)
2393                 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
2394                         *exit_status = EXIT_NICE;
2395                         return -errno;
2396                 }
2397
2398         if (context->cpu_sched_set) {
2399                 struct sched_param param = {
2400                         .sched_priority = context->cpu_sched_priority,
2401                 };
2402
2403                 r = sched_setscheduler(0,
2404                                        context->cpu_sched_policy |
2405                                        (context->cpu_sched_reset_on_fork ?
2406                                         SCHED_RESET_ON_FORK : 0),
2407                                        &param);
2408                 if (r < 0) {
2409                         *exit_status = EXIT_SETSCHEDULER;
2410                         return -errno;
2411                 }
2412         }
2413
2414         if (context->cpuset)
2415                 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
2416                         *exit_status = EXIT_CPUAFFINITY;
2417                         return -errno;
2418                 }
2419
2420         if (context->ioprio_set)
2421                 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
2422                         *exit_status = EXIT_IOPRIO;
2423                         return -errno;
2424                 }
2425
2426         if (context->timer_slack_nsec != NSEC_INFINITY)
2427                 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
2428                         *exit_status = EXIT_TIMERSLACK;
2429                         return -errno;
2430                 }
2431
2432         if (context->personality != PERSONALITY_INVALID)
2433                 if (personality(context->personality) < 0) {
2434                         *exit_status = EXIT_PERSONALITY;
2435                         return -errno;
2436                 }
2437
2438         if (context->utmp_id)
2439                 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path,
2440                                       context->utmp_mode == EXEC_UTMP_INIT  ? INIT_PROCESS :
2441                                       context->utmp_mode == EXEC_UTMP_LOGIN ? LOGIN_PROCESS :
2442                                       USER_PROCESS,
2443                                       username ? "root" : context->user);
2444
2445         if (context->user) {
2446                 r = chown_terminal(STDIN_FILENO, uid);
2447                 if (r < 0) {
2448                         *exit_status = EXIT_STDIN;
2449                         return r;
2450                 }
2451         }
2452
2453         /* If delegation is enabled we'll pass ownership of the cgroup
2454          * (but only in systemd's own controller hierarchy!) to the
2455          * user of the new process. */
2456         if (params->cgroup_path && context->user && params->cgroup_delegate) {
2457                 r = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid);
2458                 if (r < 0) {
2459                         *exit_status = EXIT_CGROUP;
2460                         return r;
2461                 }
2462
2463
2464                 r = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid);
2465                 if (r < 0) {
2466                         *exit_status = EXIT_CGROUP;
2467                         return r;
2468                 }
2469         }
2470
2471         if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) {
2472                 r = setup_runtime_directory(context, params, uid, gid);
2473                 if (r < 0) {
2474                         *exit_status = EXIT_RUNTIME_DIRECTORY;
2475                         return r;
2476                 }
2477         }
2478
2479         r = build_environment(
2480                         unit,
2481                         context,
2482                         params,
2483                         n_fds,
2484                         home,
2485                         username,
2486                         shell,
2487                         journal_stream_dev,
2488                         journal_stream_ino,
2489                         &our_env);
2490         if (r < 0) {
2491                 *exit_status = EXIT_MEMORY;
2492                 return r;
2493         }
2494
2495         r = build_pass_environment(context, &pass_env);
2496         if (r < 0) {
2497                 *exit_status = EXIT_MEMORY;
2498                 return r;
2499         }
2500
2501         accum_env = strv_env_merge(5,
2502                                    params->environment,
2503                                    our_env,
2504                                    pass_env,
2505                                    context->environment,
2506                                    files_env,
2507                                    NULL);
2508         if (!accum_env) {
2509                 *exit_status = EXIT_MEMORY;
2510                 return -ENOMEM;
2511         }
2512         accum_env = strv_env_clean(accum_env);
2513
2514         (void) umask(context->umask);
2515
2516         r = setup_keyring(unit, params, uid, gid);
2517         if (r < 0) {
2518                 *exit_status = EXIT_KEYRING;
2519                 return r;
2520         }
2521
2522         if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) {
2523                 if (context->pam_name && username) {
2524                         r = setup_pam(context->pam_name, username, uid, gid, context->tty_path, &accum_env, fds, n_fds);
2525                         if (r < 0) {
2526                                 *exit_status = EXIT_PAM;
2527                                 return r;
2528                         }
2529                 }
2530         }
2531
2532         if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
2533                 r = setup_netns(runtime->netns_storage_socket);
2534                 if (r < 0) {
2535                         *exit_status = EXIT_NETWORK;
2536                         return r;
2537                 }
2538         }
2539
2540         needs_mount_namespace = exec_needs_mount_namespace(context, params, runtime);
2541         if (needs_mount_namespace) {
2542                 r = apply_mount_namespace(unit, context, params, runtime);
2543                 if (r < 0) {
2544                         *exit_status = EXIT_NAMESPACE;
2545                         return r;
2546                 }
2547         }
2548
2549         /* Apply just after mount namespace setup */
2550         r = apply_working_directory(context, params, home, needs_mount_namespace);
2551         if (r < 0) {
2552                 *exit_status = EXIT_CHROOT;
2553                 return r;
2554         }
2555
2556         /* Drop groups as early as possbile */
2557         if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) {
2558                 r = enforce_groups(context, gid, supplementary_gids, ngids);
2559                 if (r < 0) {
2560                         *exit_status = EXIT_GROUP;
2561                         return r;
2562                 }
2563         }
2564
2565 #ifdef HAVE_SELINUX
2566         if ((params->flags & EXEC_APPLY_PERMISSIONS) &&
2567             mac_selinux_use() &&
2568             params->selinux_context_net &&
2569             socket_fd >= 0 &&
2570             !command->privileged) {
2571
2572                 r = mac_selinux_get_child_mls_label(socket_fd, command->path, context->selinux_context, &mac_selinux_context_net);
2573                 if (r < 0) {
2574                         *exit_status = EXIT_SELINUX_CONTEXT;
2575                         return r;
2576                 }
2577         }
2578 #endif
2579
2580         if ((params->flags & EXEC_APPLY_PERMISSIONS) && context->private_users) {
2581                 r = setup_private_users(uid, gid);
2582                 if (r < 0) {
2583                         *exit_status = EXIT_USER;
2584                         return r;
2585                 }
2586         }
2587
2588         /* We repeat the fd closing here, to make sure that
2589          * nothing is leaked from the PAM modules. Note that
2590          * we are more aggressive this time since socket_fd
2591          * and the netns fds we don't need anymore. The custom
2592          * endpoint fd was needed to upload the policy and can
2593          * now be closed as well. */
2594         r = close_all_fds(fds, n_fds);
2595         if (r >= 0)
2596                 r = shift_fds(fds, n_fds);
2597         if (r >= 0)
2598                 r = flags_fds(fds, n_fds, context->non_blocking);
2599         if (r < 0) {
2600                 *exit_status = EXIT_FDS;
2601                 return r;
2602         }
2603
2604         if ((params->flags & EXEC_APPLY_PERMISSIONS) && !command->privileged) {
2605
2606                 int secure_bits = context->secure_bits;
2607
2608                 for (i = 0; i < _RLIMIT_MAX; i++) {
2609
2610                         if (!context->rlimit[i])
2611                                 continue;
2612
2613                         r = setrlimit_closest(i, context->rlimit[i]);
2614                         if (r < 0) {
2615                                 *exit_status = EXIT_LIMITS;
2616                                 return r;
2617                         }
2618                 }
2619
2620                 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2621                 if (context->restrict_realtime && !context->rlimit[RLIMIT_RTPRIO]) {
2622                         if (setrlimit(RLIMIT_RTPRIO, &RLIMIT_MAKE_CONST(0)) < 0) {
2623                                 *exit_status = EXIT_LIMITS;
2624                                 return -errno;
2625                         }
2626                 }
2627
2628                 if (!cap_test_all(context->capability_bounding_set)) {
2629                         r = capability_bounding_set_drop(context->capability_bounding_set, false);
2630                         if (r < 0) {
2631                                 *exit_status = EXIT_CAPABILITIES;
2632                                 *error_message = strdup("Failed to drop capabilities");
2633                                 return r;
2634                         }
2635                 }
2636
2637                 /* This is done before enforce_user, but ambient set
2638                  * does not survive over setresuid() if keep_caps is not set. */
2639                 if (context->capability_ambient_set != 0) {
2640                         r = capability_ambient_set_apply(context->capability_ambient_set, true);
2641                         if (r < 0) {
2642                                 *exit_status = EXIT_CAPABILITIES;
2643                                 *error_message = strdup("Failed to apply ambient capabilities (before UID change)");
2644                                 return r;
2645                         }
2646                 }
2647
2648                 if (context->user) {
2649                         r = enforce_user(context, uid);
2650                         if (r < 0) {
2651                                 *exit_status = EXIT_USER;
2652                                 (void) asprintf(error_message, "Failed to change UID to "UID_FMT, uid);
2653                                 return r;
2654                         }
2655                         if (context->capability_ambient_set != 0) {
2656
2657                                 /* Fix the ambient capabilities after user change. */
2658                                 r = capability_ambient_set_apply(context->capability_ambient_set, false);
2659                                 if (r < 0) {
2660                                         *exit_status = EXIT_CAPABILITIES;
2661                                         *error_message = strdup("Failed to apply ambient capabilities (after UID change)");
2662                                         return r;
2663                                 }
2664
2665                                 /* If we were asked to change user and ambient capabilities
2666                                  * were requested, we had to add keep-caps to the securebits
2667                                  * so that we would maintain the inherited capability set
2668                                  * through the setresuid(). Make sure that the bit is added
2669                                  * also to the context secure_bits so that we don't try to
2670                                  * drop the bit away next. */
2671
2672                                 secure_bits |= 1<<SECURE_KEEP_CAPS;
2673                         }
2674                 }
2675
2676                 /* Apply the MAC contexts late, but before seccomp syscall filtering, as those should really be last to
2677                  * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
2678                  * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
2679                  * are restricted. */
2680
2681 #ifdef HAVE_SELINUX
2682                 if (mac_selinux_use()) {
2683                         char *exec_context = mac_selinux_context_net ?: context->selinux_context;
2684
2685                         if (exec_context) {
2686                                 r = setexeccon(exec_context);
2687                                 if (r < 0) {
2688                                         *exit_status = EXIT_SELINUX_CONTEXT;
2689                                         (void) asprintf(error_message, "Failed to set SELinux context to %s", exec_context);
2690                                         return r;
2691                                 }
2692                         }
2693                 }
2694 #endif
2695
2696                 r = setup_smack(context, command);
2697                 if (r < 0) {
2698                         *exit_status = EXIT_SMACK_PROCESS_LABEL;
2699                         *error_message = strdup("Failed to set SMACK process label");
2700                         return r;
2701                 }
2702
2703 #ifdef HAVE_APPARMOR
2704                 if (context->apparmor_profile && mac_apparmor_use()) {
2705                         r = aa_change_onexec(context->apparmor_profile);
2706                         if (r < 0 && !context->apparmor_profile_ignore) {
2707                                 *exit_status = EXIT_APPARMOR_PROFILE;
2708                                 (void) asprintf(error_message,
2709                                                 "Failed to prepare AppArmor profile change to %s",
2710                                                 context->apparmor_profile);
2711                                 return -errno;
2712                         }
2713                 }
2714 #endif
2715
2716                 /* PR_GET_SECUREBITS is not privileged, while
2717                  * PR_SET_SECUREBITS is. So to suppress
2718                  * potential EPERMs we'll try not to call
2719                  * PR_SET_SECUREBITS unless necessary. */
2720                 if (prctl(PR_GET_SECUREBITS) != secure_bits)
2721                         if (prctl(PR_SET_SECUREBITS, secure_bits) < 0) {
2722                                 *exit_status = EXIT_SECUREBITS;
2723                                 *error_message = strdup("Failed to set secure bits");
2724                                 return -errno;
2725                         }
2726
2727                 if (context_has_no_new_privileges(context))
2728                         if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
2729                                 *exit_status = EXIT_NO_NEW_PRIVILEGES;
2730                                 *error_message = strdup("Failed to disable new privileges");
2731                                 return -errno;
2732                         }
2733
2734 #ifdef HAVE_SECCOMP
2735                 r = apply_address_families(unit, context);
2736                 if (r < 0) {
2737                         *exit_status = EXIT_ADDRESS_FAMILIES;
2738                         *error_message = strdup("Failed to restrict address families");
2739                         return r;
2740                 }
2741
2742                 r = apply_memory_deny_write_execute(unit, context);
2743                 if (r < 0) {
2744                         *exit_status = EXIT_SECCOMP;
2745                         *error_message = strdup("Failed to disable writing to executable memory");
2746                         return r;
2747                 }
2748
2749                 r = apply_restrict_realtime(unit, context);
2750                 if (r < 0) {
2751                         *exit_status = EXIT_SECCOMP;
2752                         *error_message = strdup("Failed to apply realtime restrictions");
2753                         return r;
2754                 }
2755
2756                 r = apply_restrict_namespaces(unit, context);
2757                 if (r < 0) {
2758                         *exit_status = EXIT_SECCOMP;
2759                         *error_message = strdup("Failed to apply namespace restrictions");
2760                         return r;
2761                 }
2762
2763                 r = apply_protect_sysctl(unit, context);
2764                 if (r < 0) {
2765                         *exit_status = EXIT_SECCOMP;
2766                         *error_message = strdup("Failed to apply sysctl restrictions");
2767                         return r;
2768                 }
2769
2770                 r = apply_protect_kernel_modules(unit, context);
2771                 if (r < 0) {
2772                         *exit_status = EXIT_SECCOMP;
2773                         *error_message = strdup("Failed to apply module loading restrictions");
2774                         return r;
2775                 }
2776
2777                 r = apply_private_devices(unit, context);
2778                 if (r < 0) {
2779                         *exit_status = EXIT_SECCOMP;
2780                         *error_message = strdup("Failed to set up private devices");
2781                         return r;
2782                 }
2783
2784                 r = apply_syscall_archs(unit, context);
2785                 if (r < 0) {
2786                         *exit_status = EXIT_SECCOMP;
2787                         *error_message = strdup("Failed to apply syscall architecture restrictions");
2788                         return r;
2789                 }
2790
2791                 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
2792                  * by the filter as little as possible. */
2793                 r = apply_syscall_filter(unit, context);
2794                 if (r < 0) {
2795                         *exit_status = EXIT_SECCOMP;
2796                         *error_message = strdup("Failed to apply syscall filters");
2797                         return r;
2798                 }
2799 #endif
2800         }
2801
2802         final_argv = replace_env_argv(argv, accum_env);
2803         if (!final_argv) {
2804                 *exit_status = EXIT_MEMORY;
2805                 *error_message = strdup("Failed to prepare process arguments");
2806                 return -ENOMEM;
2807         }
2808
2809         if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
2810                 _cleanup_free_ char *line;
2811
2812                 line = exec_command_line(final_argv);
2813                 if (line) {
2814                         log_open();
2815                         log_struct(LOG_DEBUG,
2816                                    LOG_UNIT_ID(unit),
2817                                    "EXECUTABLE=%s", command->path,
2818                                    LOG_UNIT_MESSAGE(unit, "Executing: %s", line),
2819                                    NULL);
2820                         log_close();
2821                 }
2822         }
2823
2824         execve(command->path, final_argv, accum_env);
2825         *exit_status = EXIT_EXEC;
2826         return -errno;
2827 }
2828
2829 int exec_spawn(Unit *unit,
2830                ExecCommand *command,
2831                const ExecContext *context,
2832                const ExecParameters *params,
2833                ExecRuntime *runtime,
2834                DynamicCreds *dcreds,
2835                pid_t *ret) {
2836
2837         _cleanup_strv_free_ char **files_env = NULL;
2838         int *fds = NULL; unsigned n_fds = 0;
2839         _cleanup_free_ char *line = NULL;
2840         int socket_fd, r;
2841         int named_iofds[3] = { -1, -1, -1 };
2842         char **argv;
2843         pid_t pid;
2844
2845         assert(unit);
2846         assert(command);
2847         assert(context);
2848         assert(ret);
2849         assert(params);
2850         assert(params->fds || params->n_fds <= 0);
2851
2852         if (context->std_input == EXEC_INPUT_SOCKET ||
2853             context->std_output == EXEC_OUTPUT_SOCKET ||
2854             context->std_error == EXEC_OUTPUT_SOCKET) {
2855
2856                 if (params->n_fds != 1) {
2857                         log_unit_error(unit, "Got more than one socket.");
2858                         return -EINVAL;
2859                 }
2860
2861                 socket_fd = params->fds[0];
2862         } else {
2863                 socket_fd = -1;
2864                 fds = params->fds;
2865                 n_fds = params->n_fds;
2866         }
2867
2868         r = exec_context_named_iofds(unit, context, params, named_iofds);
2869         if (r < 0)
2870                 return log_unit_error_errno(unit, r, "Failed to load a named file descriptor: %m");
2871
2872         r = exec_context_load_environment(unit, context, &files_env);
2873         if (r < 0)
2874                 return log_unit_error_errno(unit, r, "Failed to load environment files: %m");
2875
2876         argv = params->argv ?: command->argv;
2877         line = exec_command_line(argv);
2878         if (!line)
2879                 return log_oom();
2880
2881         log_struct(LOG_DEBUG,
2882                    LOG_UNIT_ID(unit),
2883                    LOG_UNIT_MESSAGE(unit, "About to execute: %s", line),
2884                    "EXECUTABLE=%s", command->path,
2885                    NULL);
2886         pid = fork();
2887         if (pid < 0)
2888                 return log_unit_error_errno(unit, errno, "Failed to fork: %m");
2889
2890         if (pid == 0) {
2891                 int exit_status;
2892                 _cleanup_free_ char *error_message = NULL;
2893
2894                 r = exec_child(unit,
2895                                command,
2896                                context,
2897                                params,
2898                                runtime,
2899                                dcreds,
2900                                argv,
2901                                socket_fd,
2902                                named_iofds,
2903                                fds, n_fds,
2904                                files_env,
2905                                unit->manager->user_lookup_fds[1],
2906                                &exit_status,
2907                                &error_message);
2908                 if (r < 0) {
2909                         log_open();
2910                         if (error_message)
2911                                 log_struct_errno(LOG_ERR, r,
2912                                                  LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
2913                                                  LOG_UNIT_ID(unit),
2914                                                  LOG_UNIT_MESSAGE(unit, "%s: %m",
2915                                                                   error_message),
2916                                                  "EXECUTABLE=%s", command->path,
2917                                                  NULL);
2918                         else
2919                                 log_struct_errno(LOG_ERR, r,
2920                                                  LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
2921                                                  LOG_UNIT_ID(unit),
2922                                                  LOG_UNIT_MESSAGE(unit, "Failed at step %s spawning %s: %m",
2923                                                                   exit_status_to_string(exit_status, EXIT_STATUS_SYSTEMD),
2924                                                                   command->path),
2925                                                  "EXECUTABLE=%s", command->path,
2926                                                  NULL);
2927                 }
2928
2929                 _exit(exit_status);
2930         }
2931
2932         log_unit_debug(unit, "Forked %s as "PID_FMT, command->path, pid);
2933
2934         /* We add the new process to the cgroup both in the child (so
2935          * that we can be sure that no user code is ever executed
2936          * outside of the cgroup) and in the parent (so that we can be
2937          * sure that when we kill the cgroup the process will be
2938          * killed too). */
2939         if (params->cgroup_path)
2940                 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid);
2941
2942         exec_status_start(&command->exec_status, pid);
2943
2944         *ret = pid;
2945         return 0;
2946 }
2947
2948 void exec_context_init(ExecContext *c) {
2949         assert(c);
2950
2951         c->umask = 0022;
2952         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
2953         c->cpu_sched_policy = SCHED_OTHER;
2954         c->syslog_priority = LOG_DAEMON|LOG_INFO;
2955         c->syslog_level_prefix = true;
2956         c->ignore_sigpipe = true;
2957         c->timer_slack_nsec = NSEC_INFINITY;
2958         c->personality = PERSONALITY_INVALID;
2959         c->runtime_directory_mode = 0755;
2960         c->capability_bounding_set = CAP_ALL;
2961         c->restrict_namespaces = NAMESPACE_FLAGS_ALL;
2962 }
2963
2964 void exec_context_done(ExecContext *c) {
2965         unsigned l;
2966
2967         assert(c);
2968
2969         c->environment = strv_free(c->environment);
2970         c->environment_files = strv_free(c->environment_files);
2971         c->pass_environment = strv_free(c->pass_environment);
2972
2973         for (l = 0; l < ELEMENTSOF(c->rlimit); l++)
2974                 c->rlimit[l] = mfree(c->rlimit[l]);
2975
2976         for (l = 0; l < 3; l++)
2977                 c->stdio_fdname[l] = mfree(c->stdio_fdname[l]);
2978
2979         c->working_directory = mfree(c->working_directory);
2980         c->root_directory = mfree(c->root_directory);
2981         c->tty_path = mfree(c->tty_path);
2982         c->syslog_identifier = mfree(c->syslog_identifier);
2983         c->user = mfree(c->user);
2984         c->group = mfree(c->group);
2985
2986         c->supplementary_groups = strv_free(c->supplementary_groups);
2987
2988         c->pam_name = mfree(c->pam_name);
2989
2990         c->read_only_paths = strv_free(c->read_only_paths);
2991         c->read_write_paths = strv_free(c->read_write_paths);
2992         c->inaccessible_paths = strv_free(c->inaccessible_paths);
2993
2994         bind_mount_free_many(c->bind_mounts, c->n_bind_mounts);
2995
2996         if (c->cpuset)
2997                 CPU_FREE(c->cpuset);
2998
2999         c->utmp_id = mfree(c->utmp_id);
3000         c->selinux_context = mfree(c->selinux_context);
3001         c->apparmor_profile = mfree(c->apparmor_profile);
3002
3003         c->syscall_filter = set_free(c->syscall_filter);
3004         c->syscall_archs = set_free(c->syscall_archs);
3005         c->address_families = set_free(c->address_families);
3006
3007         c->runtime_directory = strv_free(c->runtime_directory);
3008 }
3009
3010 int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
3011         char **i;
3012
3013         assert(c);
3014
3015         if (!runtime_prefix)
3016                 return 0;
3017
3018         STRV_FOREACH(i, c->runtime_directory) {
3019                 _cleanup_free_ char *p;
3020
3021                 p = strjoin(runtime_prefix, "/", *i);
3022                 if (!p)
3023                         return -ENOMEM;
3024
3025                 /* We execute this synchronously, since we need to be
3026                  * sure this is gone when we start the service
3027                  * next. */
3028                 (void) rm_rf(p, REMOVE_ROOT);
3029         }
3030
3031         return 0;
3032 }
3033
3034 void exec_command_done(ExecCommand *c) {
3035         assert(c);
3036
3037         c->path = mfree(c->path);
3038
3039         c->argv = strv_free(c->argv);
3040 }
3041
3042 void exec_command_done_array(ExecCommand *c, unsigned n) {
3043         unsigned i;
3044
3045         for (i = 0; i < n; i++)
3046                 exec_command_done(c+i);
3047 }
3048
3049 ExecCommand* exec_command_free_list(ExecCommand *c) {
3050         ExecCommand *i;
3051
3052         while ((i = c)) {
3053                 LIST_REMOVE(command, c, i);
3054                 exec_command_done(i);
3055                 free(i);
3056         }
3057
3058         return NULL;
3059 }
3060
3061 void exec_command_free_array(ExecCommand **c, unsigned n) {
3062         unsigned i;
3063
3064         for (i = 0; i < n; i++)
3065                 c[i] = exec_command_free_list(c[i]);
3066 }
3067
3068 typedef struct InvalidEnvInfo {
3069         Unit *unit;
3070         const char *path;
3071 } InvalidEnvInfo;
3072
3073 static void invalid_env(const char *p, void *userdata) {
3074         InvalidEnvInfo *info = userdata;
3075
3076         log_unit_error(info->unit, "Ignoring invalid environment assignment '%s': %s", p, info->path);
3077 }
3078
3079 const char* exec_context_fdname(const ExecContext *c, int fd_index) {
3080         assert(c);
3081
3082         switch (fd_index) {
3083         case STDIN_FILENO:
3084                 if (c->std_input != EXEC_INPUT_NAMED_FD)
3085                         return NULL;
3086                 return c->stdio_fdname[STDIN_FILENO] ?: "stdin";
3087         case STDOUT_FILENO:
3088                 if (c->std_output != EXEC_OUTPUT_NAMED_FD)
3089                         return NULL;
3090                 return c->stdio_fdname[STDOUT_FILENO] ?: "stdout";
3091         case STDERR_FILENO:
3092                 if (c->std_error != EXEC_OUTPUT_NAMED_FD)
3093                         return NULL;
3094                 return c->stdio_fdname[STDERR_FILENO] ?: "stderr";
3095         default:
3096                 return NULL;
3097         }
3098 }
3099
3100 int exec_context_named_iofds(Unit *unit, const ExecContext *c, const ExecParameters *p, int named_iofds[3]) {
3101         unsigned i, targets;
3102         const char* stdio_fdname[3];
3103
3104         assert(c);
3105         assert(p);
3106
3107         targets = (c->std_input == EXEC_INPUT_NAMED_FD) +
3108                   (c->std_output == EXEC_OUTPUT_NAMED_FD) +
3109                   (c->std_error == EXEC_OUTPUT_NAMED_FD);
3110
3111         for (i = 0; i < 3; i++)
3112                 stdio_fdname[i] = exec_context_fdname(c, i);
3113
3114         for (i = 0; i < p->n_fds && targets > 0; i++)
3115                 if (named_iofds[STDIN_FILENO] < 0 &&
3116                     c->std_input == EXEC_INPUT_NAMED_FD &&
3117                     stdio_fdname[STDIN_FILENO] &&
3118                     streq(p->fd_names[i], stdio_fdname[STDIN_FILENO])) {
3119
3120                         named_iofds[STDIN_FILENO] = p->fds[i];
3121                         targets--;
3122
3123                 } else if (named_iofds[STDOUT_FILENO] < 0 &&
3124                            c->std_output == EXEC_OUTPUT_NAMED_FD &&
3125                            stdio_fdname[STDOUT_FILENO] &&
3126                            streq(p->fd_names[i], stdio_fdname[STDOUT_FILENO])) {
3127
3128                         named_iofds[STDOUT_FILENO] = p->fds[i];
3129                         targets--;
3130
3131                 } else if (named_iofds[STDERR_FILENO] < 0 &&
3132                            c->std_error == EXEC_OUTPUT_NAMED_FD &&
3133                            stdio_fdname[STDERR_FILENO] &&
3134                            streq(p->fd_names[i], stdio_fdname[STDERR_FILENO])) {
3135
3136                         named_iofds[STDERR_FILENO] = p->fds[i];
3137                         targets--;
3138                 }
3139
3140         return targets == 0 ? 0 : -ENOENT;
3141 }
3142
3143 int exec_context_load_environment(Unit *unit, const ExecContext *c, char ***l) {
3144         char **i, **r = NULL;
3145
3146         assert(c);
3147         assert(l);
3148
3149         STRV_FOREACH(i, c->environment_files) {
3150                 char *fn;
3151                 int k;
3152                 bool ignore = false;
3153                 char **p;
3154                 _cleanup_globfree_ glob_t pglob = {};
3155                 int count, n;
3156
3157                 fn = *i;
3158
3159                 if (fn[0] == '-') {
3160                         ignore = true;
3161                         fn++;
3162                 }
3163
3164                 if (!path_is_absolute(fn)) {
3165                         if (ignore)
3166                                 continue;
3167
3168                         strv_free(r);
3169                         return -EINVAL;
3170                 }
3171
3172                 /* Filename supports globbing, take all matching files */
3173                 errno = 0;
3174                 if (glob(fn, 0, NULL, &pglob) != 0) {
3175                         if (ignore)
3176                                 continue;
3177
3178                         strv_free(r);
3179                         return errno > 0 ? -errno : -EINVAL;
3180                 }
3181                 count = pglob.gl_pathc;
3182                 if (count == 0) {
3183                         if (ignore)
3184                                 continue;
3185
3186                         strv_free(r);
3187                         return -EINVAL;
3188                 }
3189                 for (n = 0; n < count; n++) {
3190                         k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p);
3191                         if (k < 0) {
3192                                 if (ignore)
3193                                         continue;
3194
3195                                 strv_free(r);
3196                                 return k;
3197                         }
3198                         /* Log invalid environment variables with filename */
3199                         if (p) {
3200                                 InvalidEnvInfo info = {
3201                                         .unit = unit,
3202                                         .path = pglob.gl_pathv[n]
3203                                 };
3204
3205                                 p = strv_env_clean_with_callback(p, invalid_env, &info);
3206                         }
3207
3208                         if (r == NULL)
3209                                 r = p;
3210                         else {
3211                                 char **m;
3212
3213                                 m = strv_env_merge(2, r, p);
3214                                 strv_free(r);
3215                                 strv_free(p);
3216                                 if (!m)
3217                                         return -ENOMEM;
3218
3219                                 r = m;
3220                         }
3221                 }
3222         }
3223
3224         *l = r;
3225
3226         return 0;
3227 }
3228
3229 static bool tty_may_match_dev_console(const char *tty) {
3230         _cleanup_free_ char *active = NULL;
3231         char *console;
3232
3233         if (!tty)
3234                 return true;
3235
3236         if (startswith(tty, "/dev/"))
3237                 tty += 5;
3238
3239         /* trivial identity? */
3240         if (streq(tty, "console"))
3241                 return true;
3242
3243         console = resolve_dev_console(&active);
3244         /* if we could not resolve, assume it may */
3245         if (!console)
3246                 return true;
3247
3248         /* "tty0" means the active VC, so it may be the same sometimes */
3249         return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
3250 }
3251
3252 bool exec_context_may_touch_console(ExecContext *ec) {
3253
3254         return (ec->tty_reset ||
3255                 ec->tty_vhangup ||
3256                 ec->tty_vt_disallocate ||
3257                 is_terminal_input(ec->std_input) ||
3258                 is_terminal_output(ec->std_output) ||
3259                 is_terminal_output(ec->std_error)) &&
3260                tty_may_match_dev_console(exec_context_tty_path(ec));
3261 }
3262
3263 static void strv_fprintf(FILE *f, char **l) {
3264         char **g;
3265
3266         assert(f);
3267
3268         STRV_FOREACH(g, l)
3269                 fprintf(f, " %s", *g);
3270 }
3271
3272 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
3273         char **e, **d;
3274         unsigned i;
3275         int r;
3276
3277         assert(c);
3278         assert(f);
3279
3280         prefix = strempty(prefix);
3281
3282         fprintf(f,
3283                 "%sUMask: %04o\n"
3284                 "%sWorkingDirectory: %s\n"
3285                 "%sRootDirectory: %s\n"
3286                 "%sNonBlocking: %s\n"
3287                 "%sPrivateTmp: %s\n"
3288                 "%sPrivateDevices: %s\n"
3289                 "%sProtectKernelTunables: %s\n"
3290                 "%sProtectKernelModules: %s\n"
3291                 "%sProtectControlGroups: %s\n"
3292                 "%sPrivateNetwork: %s\n"
3293                 "%sPrivateUsers: %s\n"
3294                 "%sProtectHome: %s\n"
3295                 "%sProtectSystem: %s\n"
3296                 "%sIgnoreSIGPIPE: %s\n"
3297                 "%sMemoryDenyWriteExecute: %s\n"
3298                 "%sRestrictRealtime: %s\n",
3299                 prefix, c->umask,
3300                 prefix, c->working_directory ? c->working_directory : "/",
3301                 prefix, c->root_directory ? c->root_directory : "/",
3302                 prefix, yes_no(c->non_blocking),
3303                 prefix, yes_no(c->private_tmp),
3304                 prefix, yes_no(c->private_devices),
3305                 prefix, yes_no(c->protect_kernel_tunables),
3306                 prefix, yes_no(c->protect_kernel_modules),
3307                 prefix, yes_no(c->protect_control_groups),
3308                 prefix, yes_no(c->private_network),
3309                 prefix, yes_no(c->private_users),
3310                 prefix, protect_home_to_string(c->protect_home),
3311                 prefix, protect_system_to_string(c->protect_system),
3312                 prefix, yes_no(c->ignore_sigpipe),
3313                 prefix, yes_no(c->memory_deny_write_execute),
3314                 prefix, yes_no(c->restrict_realtime));
3315
3316         STRV_FOREACH(e, c->environment)
3317                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
3318
3319         STRV_FOREACH(e, c->environment_files)
3320                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
3321
3322         STRV_FOREACH(e, c->pass_environment)
3323                 fprintf(f, "%sPassEnvironment: %s\n", prefix, *e);
3324
3325         fprintf(f, "%sRuntimeDirectoryMode: %04o\n", prefix, c->runtime_directory_mode);
3326
3327         STRV_FOREACH(d, c->runtime_directory)
3328                 fprintf(f, "%sRuntimeDirectory: %s\n", prefix, *d);
3329
3330         if (c->nice_set)
3331                 fprintf(f,
3332                         "%sNice: %i\n",
3333                         prefix, c->nice);
3334
3335         if (c->oom_score_adjust_set)
3336                 fprintf(f,
3337                         "%sOOMScoreAdjust: %i\n",
3338                         prefix, c->oom_score_adjust);
3339
3340         for (i = 0; i < RLIM_NLIMITS; i++)
3341                 if (c->rlimit[i]) {
3342                         fprintf(f, "%s%s: " RLIM_FMT "\n",
3343                                 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
3344                         fprintf(f, "%s%sSoft: " RLIM_FMT "\n",
3345                                 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_cur);
3346                 }
3347
3348         if (c->ioprio_set) {
3349                 _cleanup_free_ char *class_str = NULL;
3350
3351                 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
3352                 fprintf(f,
3353                         "%sIOSchedulingClass: %s\n"
3354                         "%sIOPriority: %i\n",
3355                         prefix, strna(class_str),
3356                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
3357         }
3358
3359         if (c->cpu_sched_set) {
3360                 _cleanup_free_ char *policy_str = NULL;
3361
3362                 sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
3363                 fprintf(f,
3364                         "%sCPUSchedulingPolicy: %s\n"
3365                         "%sCPUSchedulingPriority: %i\n"
3366                         "%sCPUSchedulingResetOnFork: %s\n",
3367                         prefix, strna(policy_str),
3368                         prefix, c->cpu_sched_priority,
3369                         prefix, yes_no(c->cpu_sched_reset_on_fork));
3370         }
3371
3372         if (c->cpuset) {
3373                 fprintf(f, "%sCPUAffinity:", prefix);
3374                 for (i = 0; i < c->cpuset_ncpus; i++)
3375                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
3376                                 fprintf(f, " %u", i);
3377                 fputs("\n", f);
3378         }
3379
3380         if (c->timer_slack_nsec != NSEC_INFINITY)
3381                 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
3382
3383         fprintf(f,
3384                 "%sStandardInput: %s\n"
3385                 "%sStandardOutput: %s\n"
3386                 "%sStandardError: %s\n",
3387                 prefix, exec_input_to_string(c->std_input),
3388                 prefix, exec_output_to_string(c->std_output),
3389                 prefix, exec_output_to_string(c->std_error));
3390
3391         if (c->tty_path)
3392                 fprintf(f,
3393                         "%sTTYPath: %s\n"
3394                         "%sTTYReset: %s\n"
3395                         "%sTTYVHangup: %s\n"
3396                         "%sTTYVTDisallocate: %s\n",
3397                         prefix, c->tty_path,
3398                         prefix, yes_no(c->tty_reset),
3399                         prefix, yes_no(c->tty_vhangup),
3400                         prefix, yes_no(c->tty_vt_disallocate));
3401
3402         if (c->std_output == EXEC_OUTPUT_SYSLOG ||
3403             c->std_output == EXEC_OUTPUT_KMSG ||
3404             c->std_output == EXEC_OUTPUT_JOURNAL ||
3405             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
3406             c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
3407             c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
3408             c->std_error == EXEC_OUTPUT_SYSLOG ||
3409             c->std_error == EXEC_OUTPUT_KMSG ||
3410             c->std_error == EXEC_OUTPUT_JOURNAL ||
3411             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
3412             c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
3413             c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
3414
3415                 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
3416
3417                 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
3418                 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
3419
3420                 fprintf(f,
3421                         "%sSyslogFacility: %s\n"
3422                         "%sSyslogLevel: %s\n",
3423                         prefix, strna(fac_str),
3424                         prefix, strna(lvl_str));
3425         }
3426
3427         if (c->secure_bits)
3428                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
3429                         prefix,
3430                         (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
3431                         (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
3432                         (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
3433                         (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
3434                         (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
3435                         (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
3436
3437         if (c->capability_bounding_set != CAP_ALL) {
3438                 unsigned long l;
3439                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
3440
3441                 for (l = 0; l <= cap_last_cap(); l++)
3442                         if (c->capability_bounding_set & (UINT64_C(1) << l))
3443                                 fprintf(f, " %s", strna(capability_to_name(l)));
3444
3445                 fputs("\n", f);
3446         }
3447
3448         if (c->capability_ambient_set != 0) {
3449                 unsigned long l;
3450                 fprintf(f, "%sAmbientCapabilities:", prefix);
3451
3452                 for (l = 0; l <= cap_last_cap(); l++)
3453                         if (c->capability_ambient_set & (UINT64_C(1) << l))
3454                                 fprintf(f, " %s", strna(capability_to_name(l)));
3455
3456                 fputs("\n", f);
3457         }
3458
3459         if (c->user)
3460                 fprintf(f, "%sUser: %s\n", prefix, c->user);
3461         if (c->group)
3462                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
3463
3464         fprintf(f, "%sDynamicUser: %s\n", prefix, yes_no(c->dynamic_user));
3465
3466         if (strv_length(c->supplementary_groups) > 0) {
3467                 fprintf(f, "%sSupplementaryGroups:", prefix);
3468                 strv_fprintf(f, c->supplementary_groups);
3469                 fputs("\n", f);
3470         }
3471
3472         if (c->pam_name)
3473                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
3474
3475         if (strv_length(c->read_write_paths) > 0) {
3476                 fprintf(f, "%sReadWritePaths:", prefix);
3477                 strv_fprintf(f, c->read_write_paths);
3478                 fputs("\n", f);
3479         }
3480
3481         if (strv_length(c->read_only_paths) > 0) {
3482                 fprintf(f, "%sReadOnlyPaths:", prefix);
3483                 strv_fprintf(f, c->read_only_paths);
3484                 fputs("\n", f);
3485         }
3486
3487         if (strv_length(c->inaccessible_paths) > 0) {
3488                 fprintf(f, "%sInaccessiblePaths:", prefix);
3489                 strv_fprintf(f, c->inaccessible_paths);
3490                 fputs("\n", f);
3491         }
3492
3493         if (c->n_bind_mounts > 0)
3494                 for (i = 0; i < c->n_bind_mounts; i++) {
3495                         fprintf(f, "%s%s: %s:%s:%s\n", prefix,
3496                                 c->bind_mounts[i].read_only ? "BindReadOnlyPaths" : "BindPaths",
3497                                 c->bind_mounts[i].source,
3498                                 c->bind_mounts[i].destination,
3499                                 c->bind_mounts[i].recursive ? "rbind" : "norbind");
3500                 }
3501
3502         if (c->utmp_id)
3503                 fprintf(f,
3504                         "%sUtmpIdentifier: %s\n",
3505                         prefix, c->utmp_id);
3506
3507         if (c->selinux_context)
3508                 fprintf(f,
3509                         "%sSELinuxContext: %s%s\n",
3510                         prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
3511
3512         if (c->personality != PERSONALITY_INVALID)
3513                 fprintf(f,
3514                         "%sPersonality: %s\n",
3515                         prefix, strna(personality_to_string(c->personality)));
3516
3517         if (c->syscall_filter) {
3518 #ifdef HAVE_SECCOMP
3519                 Iterator j;
3520                 void *id;
3521                 bool first = true;
3522 #endif
3523
3524                 fprintf(f,
3525                         "%sSystemCallFilter: ",
3526                         prefix);
3527
3528                 if (!c->syscall_whitelist)
3529                         fputc('~', f);
3530
3531 #ifdef HAVE_SECCOMP
3532                 SET_FOREACH(id, c->syscall_filter, j) {
3533                         _cleanup_free_ char *name = NULL;
3534
3535                         if (first)
3536                                 first = false;
3537                         else
3538                                 fputc(' ', f);
3539
3540                         name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
3541                         fputs(strna(name), f);
3542                 }
3543 #endif
3544
3545                 fputc('\n', f);
3546         }
3547
3548         if (c->syscall_archs) {
3549 #ifdef HAVE_SECCOMP
3550                 Iterator j;
3551                 void *id;
3552 #endif
3553
3554                 fprintf(f,
3555                         "%sSystemCallArchitectures:",
3556                         prefix);
3557
3558 #ifdef HAVE_SECCOMP
3559                 SET_FOREACH(id, c->syscall_archs, j)
3560                         fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
3561 #endif
3562                 fputc('\n', f);
3563         }
3564
3565         if (exec_context_restrict_namespaces_set(c)) {
3566                 _cleanup_free_ char *s = NULL;
3567
3568                 r = namespace_flag_to_string_many(c->restrict_namespaces, &s);
3569                 if (r >= 0)
3570                         fprintf(f, "%sRestrictNamespaces: %s\n",
3571                                 prefix, s);
3572         }
3573
3574         if (c->syscall_errno > 0)
3575                 fprintf(f,
3576                         "%sSystemCallErrorNumber: %s\n",
3577                         prefix, strna(errno_to_name(c->syscall_errno)));
3578
3579         if (c->apparmor_profile)
3580                 fprintf(f,
3581                         "%sAppArmorProfile: %s%s\n",
3582                         prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
3583 }
3584
3585 bool exec_context_maintains_privileges(ExecContext *c) {
3586         assert(c);
3587
3588         /* Returns true if the process forked off would run under
3589          * an unchanged UID or as root. */
3590
3591         if (!c->user)
3592                 return true;
3593
3594         if (streq(c->user, "root") || streq(c->user, "0"))
3595                 return true;
3596
3597         return false;
3598 }
3599
3600 void exec_status_start(ExecStatus *s, pid_t pid) {
3601         assert(s);
3602
3603         zero(*s);
3604         s->pid = pid;
3605         dual_timestamp_get(&s->start_timestamp);
3606 }
3607
3608 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
3609         assert(s);
3610
3611         if (s->pid && s->pid != pid)
3612                 zero(*s);
3613
3614         s->pid = pid;
3615         dual_timestamp_get(&s->exit_timestamp);
3616
3617         s->code = code;
3618         s->status = status;
3619
3620         if (context) {
3621                 if (context->utmp_id)
3622                         utmp_put_dead_process(context->utmp_id, pid, code, status);
3623
3624                 exec_context_tty_reset(context, NULL);
3625         }
3626 }
3627
3628 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
3629         char buf[FORMAT_TIMESTAMP_MAX];
3630
3631         assert(s);
3632         assert(f);
3633
3634         if (s->pid <= 0)
3635                 return;
3636
3637         prefix = strempty(prefix);
3638
3639         fprintf(f,
3640                 "%sPID: "PID_FMT"\n",
3641                 prefix, s->pid);
3642
3643         if (dual_timestamp_is_set(&s->start_timestamp))
3644                 fprintf(f,
3645                         "%sStart Timestamp: %s\n",
3646                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
3647
3648         if (dual_timestamp_is_set(&s->exit_timestamp))
3649                 fprintf(f,
3650                         "%sExit Timestamp: %s\n"
3651                         "%sExit Code: %s\n"
3652                         "%sExit Status: %i\n",
3653                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
3654                         prefix, sigchld_code_to_string(s->code),
3655                         prefix, s->status);
3656 }
3657
3658 char *exec_command_line(char **argv) {
3659         size_t k;
3660         char *n, *p, **a;
3661         bool first = true;
3662
3663         assert(argv);
3664
3665         k = 1;
3666         STRV_FOREACH(a, argv)
3667                 k += strlen(*a)+3;
3668
3669         n = new(char, k);
3670         if (!n)
3671                 return NULL;
3672
3673         p = n;
3674         STRV_FOREACH(a, argv) {
3675
3676                 if (!first)
3677                         *(p++) = ' ';
3678                 else
3679                         first = false;
3680
3681                 if (strpbrk(*a, WHITESPACE)) {
3682                         *(p++) = '\'';
3683                         p = stpcpy(p, *a);
3684                         *(p++) = '\'';
3685                 } else
3686                         p = stpcpy(p, *a);
3687
3688         }
3689
3690         *p = 0;
3691
3692         /* FIXME: this doesn't really handle arguments that have
3693          * spaces and ticks in them */
3694
3695         return n;
3696 }
3697
3698 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
3699         _cleanup_free_ char *cmd = NULL;
3700         const char *prefix2;
3701
3702         assert(c);
3703         assert(f);
3704
3705         prefix = strempty(prefix);
3706         prefix2 = strjoina(prefix, "\t");
3707
3708         cmd = exec_command_line(c->argv);
3709         fprintf(f,
3710                 "%sCommand Line: %s\n",
3711                 prefix, cmd ? cmd : strerror(ENOMEM));
3712
3713         exec_status_dump(&c->exec_status, f, prefix2);
3714 }
3715
3716 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
3717         assert(f);
3718
3719         prefix = strempty(prefix);
3720
3721         LIST_FOREACH(command, c, c)
3722                 exec_command_dump(c, f, prefix);
3723 }
3724
3725 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
3726         ExecCommand *end;
3727
3728         assert(l);
3729         assert(e);
3730
3731         if (*l) {
3732                 /* It's kind of important, that we keep the order here */
3733                 LIST_FIND_TAIL(command, *l, end);
3734                 LIST_INSERT_AFTER(command, *l, end, e);
3735         } else
3736               *l = e;
3737 }
3738
3739 int exec_command_set(ExecCommand *c, const char *path, ...) {
3740         va_list ap;
3741         char **l, *p;
3742
3743         assert(c);
3744         assert(path);
3745
3746         va_start(ap, path);
3747         l = strv_new_ap(path, ap);
3748         va_end(ap);
3749
3750         if (!l)
3751                 return -ENOMEM;
3752
3753         p = strdup(path);
3754         if (!p) {
3755                 strv_free(l);
3756                 return -ENOMEM;
3757         }
3758
3759         free(c->path);
3760         c->path = p;
3761
3762         strv_free(c->argv);
3763         c->argv = l;
3764
3765         return 0;
3766 }
3767
3768 int exec_command_append(ExecCommand *c, const char *path, ...) {
3769         _cleanup_strv_free_ char **l = NULL;
3770         va_list ap;
3771         int r;
3772
3773         assert(c);
3774         assert(path);
3775
3776         va_start(ap, path);
3777         l = strv_new_ap(path, ap);
3778         va_end(ap);
3779
3780         if (!l)
3781                 return -ENOMEM;
3782
3783         r = strv_extend_strv(&c->argv, l, false);
3784         if (r < 0)
3785                 return r;
3786
3787         return 0;
3788 }
3789
3790
3791 static int exec_runtime_allocate(ExecRuntime **rt) {
3792
3793         if (*rt)
3794                 return 0;
3795
3796         *rt = new0(ExecRuntime, 1);
3797         if (!*rt)
3798                 return -ENOMEM;
3799
3800         (*rt)->n_ref = 1;
3801         (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
3802
3803         return 0;
3804 }
3805
3806 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
3807         int r;
3808
3809         assert(rt);
3810         assert(c);
3811         assert(id);
3812
3813         if (*rt)
3814                 return 1;
3815
3816         if (!c->private_network && !c->private_tmp)
3817                 return 0;
3818
3819         r = exec_runtime_allocate(rt);
3820         if (r < 0)
3821                 return r;
3822
3823         if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
3824                 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, (*rt)->netns_storage_socket) < 0)
3825                         return -errno;
3826         }
3827
3828         if (c->private_tmp && !(*rt)->tmp_dir) {
3829                 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
3830                 if (r < 0)
3831                         return r;
3832         }
3833
3834         return 1;
3835 }
3836
3837 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
3838         assert(r);
3839         assert(r->n_ref > 0);
3840
3841         r->n_ref++;
3842         return r;
3843 }
3844
3845 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
3846
3847         if (!r)
3848                 return NULL;
3849
3850         assert(r->n_ref > 0);
3851
3852         r->n_ref--;
3853         if (r->n_ref > 0)
3854                 return NULL;
3855
3856         free(r->tmp_dir);
3857         free(r->var_tmp_dir);
3858         safe_close_pair(r->netns_storage_socket);
3859         return mfree(r);
3860 }
3861
3862 int exec_runtime_serialize(Unit *u, ExecRuntime *rt, FILE *f, FDSet *fds) {
3863         assert(u);
3864         assert(f);
3865         assert(fds);
3866
3867         if (!rt)
3868                 return 0;
3869
3870         if (rt->tmp_dir)
3871                 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
3872
3873         if (rt->var_tmp_dir)
3874                 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
3875
3876         if (rt->netns_storage_socket[0] >= 0) {
3877                 int copy;
3878
3879                 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
3880                 if (copy < 0)
3881                         return copy;
3882
3883                 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
3884         }
3885
3886         if (rt->netns_storage_socket[1] >= 0) {
3887                 int copy;
3888
3889                 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
3890                 if (copy < 0)
3891                         return copy;
3892
3893                 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
3894         }
3895
3896         return 0;
3897 }
3898
3899 int exec_runtime_deserialize_item(Unit *u, ExecRuntime **rt, const char *key, const char *value, FDSet *fds) {
3900         int r;
3901
3902         assert(rt);
3903         assert(key);
3904         assert(value);
3905
3906         if (streq(key, "tmp-dir")) {
3907                 char *copy;
3908
3909                 r = exec_runtime_allocate(rt);
3910                 if (r < 0)
3911                         return log_oom();
3912
3913                 copy = strdup(value);
3914                 if (!copy)
3915                         return log_oom();
3916
3917                 free((*rt)->tmp_dir);
3918                 (*rt)->tmp_dir = copy;
3919
3920         } else if (streq(key, "var-tmp-dir")) {
3921                 char *copy;
3922
3923                 r = exec_runtime_allocate(rt);
3924                 if (r < 0)
3925                         return log_oom();
3926
3927                 copy = strdup(value);
3928                 if (!copy)
3929                         return log_oom();
3930
3931                 free((*rt)->var_tmp_dir);
3932                 (*rt)->var_tmp_dir = copy;
3933
3934         } else if (streq(key, "netns-socket-0")) {
3935                 int fd;
3936
3937                 r = exec_runtime_allocate(rt);
3938                 if (r < 0)
3939                         return log_oom();
3940
3941                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
3942                         log_unit_debug(u, "Failed to parse netns socket value: %s", value);
3943                 else {
3944                         safe_close((*rt)->netns_storage_socket[0]);
3945                         (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
3946                 }
3947         } else if (streq(key, "netns-socket-1")) {
3948                 int fd;
3949
3950                 r = exec_runtime_allocate(rt);
3951                 if (r < 0)
3952                         return log_oom();
3953
3954                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
3955                         log_unit_debug(u, "Failed to parse netns socket value: %s", value);
3956                 else {
3957                         safe_close((*rt)->netns_storage_socket[1]);
3958                         (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
3959                 }
3960         } else
3961                 return 0;
3962
3963         return 1;
3964 }
3965
3966 static void *remove_tmpdir_thread(void *p) {
3967         _cleanup_free_ char *path = p;
3968
3969         (void) rm_rf(path, REMOVE_ROOT|REMOVE_PHYSICAL);
3970         return NULL;
3971 }
3972
3973 void exec_runtime_destroy(ExecRuntime *rt) {
3974         int r;
3975
3976         if (!rt)
3977                 return;
3978
3979         /* If there are multiple users of this, let's leave the stuff around */
3980         if (rt->n_ref > 1)
3981                 return;
3982
3983         if (rt->tmp_dir) {
3984                 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
3985
3986                 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
3987                 if (r < 0) {
3988                         log_warning_errno(r, "Failed to nuke %s: %m", rt->tmp_dir);
3989                         free(rt->tmp_dir);
3990                 }
3991
3992                 rt->tmp_dir = NULL;
3993         }
3994
3995         if (rt->var_tmp_dir) {
3996                 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
3997
3998                 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
3999                 if (r < 0) {
4000                         log_warning_errno(r, "Failed to nuke %s: %m", rt->var_tmp_dir);
4001                         free(rt->var_tmp_dir);
4002                 }
4003
4004                 rt->var_tmp_dir = NULL;
4005         }
4006
4007         safe_close_pair(rt->netns_storage_socket);
4008 }
4009
4010 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
4011         [EXEC_INPUT_NULL] = "null",
4012         [EXEC_INPUT_TTY] = "tty",
4013         [EXEC_INPUT_TTY_FORCE] = "tty-force",
4014         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
4015         [EXEC_INPUT_SOCKET] = "socket",
4016         [EXEC_INPUT_NAMED_FD] = "fd",
4017 };
4018
4019 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
4020
4021 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
4022         [EXEC_OUTPUT_INHERIT] = "inherit",
4023         [EXEC_OUTPUT_NULL] = "null",
4024         [EXEC_OUTPUT_TTY] = "tty",
4025         [EXEC_OUTPUT_SYSLOG] = "syslog",
4026         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
4027         [EXEC_OUTPUT_KMSG] = "kmsg",
4028         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
4029         [EXEC_OUTPUT_JOURNAL] = "journal",
4030         [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
4031         [EXEC_OUTPUT_SOCKET] = "socket",
4032         [EXEC_OUTPUT_NAMED_FD] = "fd",
4033 };
4034
4035 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
4036
4037 static const char* const exec_utmp_mode_table[_EXEC_UTMP_MODE_MAX] = {
4038         [EXEC_UTMP_INIT] = "init",
4039         [EXEC_UTMP_LOGIN] = "login",
4040         [EXEC_UTMP_USER] = "user",
4041 };
4042
4043 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode, ExecUtmpMode);