src/core/execute.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <assert.h>
  23 #include <dirent.h>
  24 #include <errno.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include <string.h>
  28 #include <signal.h>
  29 #include <sys/socket.h>
  30 #include <sys/un.h>
  31 #include <sys/prctl.h>
  32 #include <linux/sched.h>
  33 #include <sys/types.h>
  34 #include <sys/stat.h>
  35 #include <grp.h>
  36 #include <pwd.h>
  37 #include <sys/mount.h>
  38 #include <linux/fs.h>
  39 #include <linux/oom.h>
  40 #include <sys/poll.h>
  41 #include <glob.h>
  42 #include <sys/personality.h>
  43 #include <libgen.h>
  44 #undef basename
  45
  46 #ifdef HAVE_PAM
  47 #include <security/pam_appl.h>
  48 #endif
  49
  50 #ifdef HAVE_SELINUX
  51 #include <selinux/selinux.h>
  52 #endif
  53
  54 #ifdef HAVE_SECCOMP
  55 #include <seccomp.h>
  56 #endif
  57
  58 #ifdef HAVE_APPARMOR
  59 #include <sys/apparmor.h>
  60 #endif
  61
  62 #include "execute.h"
  63 #include "strv.h"
  64 #include "macro.h"
  65 #include "capability.h"
  66 #include "util.h"
  67 #include "log.h"
  68 #include "sd-messages.h"
  69 #include "ioprio.h"
  70 #include "securebits.h"
  71 #include "namespace.h"
  72 #include "exit-status.h"
  73 #include "missing.h"
  74 #include "utmp-wtmp.h"
  75 #include "def.h"
  76 #include "path-util.h"
  77 #include "env-util.h"
  78 #include "fileio.h"
  79 #include "unit.h"
  80 #include "async.h"
  81 #include "selinux-util.h"
  82 #include "errno-list.h"
  83 #include "af-list.h"
  84 #include "mkdir.h"
  85 #include "apparmor-util.h"
  86
  87 #ifdef HAVE_SECCOMP
  88 #include "seccomp-util.h"
  89 #endif
  90
  91 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
  92 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
  93
  94 /* This assumes there is a 'tty' group */
  95 #define TTY_MODE 0620
  96
  97 #define SNDBUF_SIZE (8*1024*1024)
  98
  99 static int shift_fds(int fds[], unsigned n_fds) {
 100         int start, restart_from;
 101
 102         if (n_fds <= 0)
 103                 return 0;
 104
 105         /* Modifies the fds array! (sorts it) */
 106
 107         assert(fds);
 108
 109         start = 0;
 110         for (;;) {
 111                 int i;
 112
 113                 restart_from = -1;
 114
 115                 for (i = start; i < (int) n_fds; i++) {
 116                         int nfd;
 117
 118                         /* Already at right index? */
 119                         if (fds[i] == i+3)
 120                                 continue;
 121
 122                         if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
 123                                 return -errno;
 124
 125                         safe_close(fds[i]);
 126                         fds[i] = nfd;
 127
 128                         /* Hmm, the fd we wanted isn't free? Then
 129                          * let's remember that and try again from here*/
 130                         if (nfd != i+3 && restart_from < 0)
 131                                 restart_from = i;
 132                 }
 133
 134                 if (restart_from < 0)
 135                         break;
 136
 137                 start = restart_from;
 138         }
 139
 140         return 0;
 141 }
 142
 143 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
 144         unsigned i;
 145         int r;
 146
 147         if (n_fds <= 0)
 148                 return 0;
 149
 150         assert(fds);
 151
 152         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
 153
 154         for (i = 0; i < n_fds; i++) {
 155
 156                 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
 157                         return r;
 158
 159                 /* We unconditionally drop FD_CLOEXEC from the fds,
 160                  * since after all we want to pass these fds to our
 161                  * children */
 162
 163                 if ((r = fd_cloexec(fds[i], false)) < 0)
 164                         return r;
 165         }
 166
 167         return 0;
 168 }
 169
 170 _pure_ static const char *tty_path(const ExecContext *context) {
 171         assert(context);
 172
 173         if (context->tty_path)
 174                 return context->tty_path;
 175
 176         return "/dev/console";
 177 }
 178
 179 static void exec_context_tty_reset(const ExecContext *context) {
 180         assert(context);
 181
 182         if (context->tty_vhangup)
 183                 terminal_vhangup(tty_path(context));
 184
 185         if (context->tty_reset)
 186                 reset_terminal(tty_path(context));
 187
 188         if (context->tty_vt_disallocate && context->tty_path)
 189                 vt_disallocate(context->tty_path);
 190 }
 191
 192 static bool is_terminal_output(ExecOutput o) {
 193         return
 194                 o == EXEC_OUTPUT_TTY ||
 195                 o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
 196                 o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
 197                 o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
 198 }
 199
 200 static int open_null_as(int flags, int nfd) {
 201         int fd, r;
 202
 203         assert(nfd >= 0);
 204
 205         fd = open("/dev/null", flags|O_NOCTTY);
 206         if (fd < 0)
 207                 return -errno;
 208
 209         if (fd != nfd) {
 210                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 211                 safe_close(fd);
 212         } else
 213                 r = nfd;
 214
 215         return r;
 216 }
 217
 218 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
 219         int fd, r;
 220         union sockaddr_union sa = {
 221                 .un.sun_family = AF_UNIX,
 222                 .un.sun_path = "/run/systemd/journal/stdout",
 223         };
 224
 225         assert(context);
 226         assert(output < _EXEC_OUTPUT_MAX);
 227         assert(ident);
 228         assert(nfd >= 0);
 229
 230         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 231         if (fd < 0)
 232                 return -errno;
 233
 234         r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
 235         if (r < 0) {
 236                 safe_close(fd);
 237                 return -errno;
 238         }
 239
 240         if (shutdown(fd, SHUT_RD) < 0) {
 241                 safe_close(fd);
 242                 return -errno;
 243         }
 244
 245         fd_inc_sndbuf(fd, SNDBUF_SIZE);
 246
 247         dprintf(fd,
 248                 "%s\n"
 249                 "%s\n"
 250                 "%i\n"
 251                 "%i\n"
 252                 "%i\n"
 253                 "%i\n"
 254                 "%i\n",
 255                 context->syslog_identifier ? context->syslog_identifier : ident,
 256                 unit_id,
 257                 context->syslog_priority,
 258                 !!context->syslog_level_prefix,
 259                 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
 260                 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
 261                 is_terminal_output(output));
 262
 263         if (fd != nfd) {
 264                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 265                 safe_close(fd);
 266         } else
 267                 r = nfd;
 268
 269         return r;
 270 }
 271 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
 272         int fd, r;
 273
 274         assert(path);
 275         assert(nfd >= 0);
 276
 277         if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
 278                 return fd;
 279
 280         if (fd != nfd) {
 281                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 282                 safe_close(fd);
 283         } else
 284                 r = nfd;
 285
 286         return r;
 287 }
 288
 289 static bool is_terminal_input(ExecInput i) {
 290         return
 291                 i == EXEC_INPUT_TTY ||
 292                 i == EXEC_INPUT_TTY_FORCE ||
 293                 i == EXEC_INPUT_TTY_FAIL;
 294 }
 295
 296 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
 297
 298         if (is_terminal_input(std_input) && !apply_tty_stdin)
 299                 return EXEC_INPUT_NULL;
 300
 301         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 302                 return EXEC_INPUT_NULL;
 303
 304         return std_input;
 305 }
 306
 307 static int fixup_output(ExecOutput std_output, int socket_fd) {
 308
 309         if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 310                 return EXEC_OUTPUT_INHERIT;
 311
 312         return std_output;
 313 }
 314
 315 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
 316         ExecInput i;
 317
 318         assert(context);
 319
 320         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 321
 322         switch (i) {
 323
 324         case EXEC_INPUT_NULL:
 325                 return open_null_as(O_RDONLY, STDIN_FILENO);
 326
 327         case EXEC_INPUT_TTY:
 328         case EXEC_INPUT_TTY_FORCE:
 329         case EXEC_INPUT_TTY_FAIL: {
 330                 int fd, r;
 331
 332                 fd = acquire_terminal(tty_path(context),
 333                                       i == EXEC_INPUT_TTY_FAIL,
 334                                       i == EXEC_INPUT_TTY_FORCE,
 335                                       false,
 336                                       USEC_INFINITY);
 337                 if (fd < 0)
 338                         return fd;
 339
 340                 if (fd != STDIN_FILENO) {
 341                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 342                         safe_close(fd);
 343                 } else
 344                         r = STDIN_FILENO;
 345
 346                 return r;
 347         }
 348
 349         case EXEC_INPUT_SOCKET:
 350                 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 351
 352         default:
 353                 assert_not_reached("Unknown input type");
 354         }
 355 }
 356
 357 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
 358         ExecOutput o;
 359         ExecInput i;
 360         int r;
 361
 362         assert(context);
 363         assert(ident);
 364
 365         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 366         o = fixup_output(context->std_output, socket_fd);
 367
 368         if (fileno == STDERR_FILENO) {
 369                 ExecOutput e;
 370                 e = fixup_output(context->std_error, socket_fd);
 371
 372                 /* This expects the input and output are already set up */
 373
 374                 /* Don't change the stderr file descriptor if we inherit all
 375                  * the way and are not on a tty */
 376                 if (e == EXEC_OUTPUT_INHERIT &&
 377                     o == EXEC_OUTPUT_INHERIT &&
 378                     i == EXEC_INPUT_NULL &&
 379                     !is_terminal_input(context->std_input) &&
 380                     getppid () != 1)
 381                         return fileno;
 382
 383                 /* Duplicate from stdout if possible */
 384                 if (e == o || e == EXEC_OUTPUT_INHERIT)
 385                         return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
 386
 387                 o = e;
 388
 389         } else if (o == EXEC_OUTPUT_INHERIT) {
 390                 /* If input got downgraded, inherit the original value */
 391                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 392                         return open_terminal_as(tty_path(context), O_WRONLY, fileno);
 393
 394                 /* If the input is connected to anything that's not a /dev/null, inherit that... */
 395                 if (i != EXEC_INPUT_NULL)
 396                         return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
 397
 398                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 399                 if (getppid() != 1)
 400                         return fileno;
 401
 402                 /* We need to open /dev/null here anew, to get the right access mode. */
 403                 return open_null_as(O_WRONLY, fileno);
 404         }
 405
 406         switch (o) {
 407
 408         case EXEC_OUTPUT_NULL:
 409                 return open_null_as(O_WRONLY, fileno);
 410
 411         case EXEC_OUTPUT_TTY:
 412                 if (is_terminal_input(i))
 413                         return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
 414
 415                 /* We don't reset the terminal if this is just about output */
 416                 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
 417
 418         case EXEC_OUTPUT_SYSLOG:
 419         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 420         case EXEC_OUTPUT_KMSG:
 421         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 422         case EXEC_OUTPUT_JOURNAL:
 423         case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
 424                 r = connect_logger_as(context, o, ident, unit_id, fileno);
 425                 if (r < 0) {
 426                         log_struct_unit(LOG_CRIT, unit_id,
 427                                 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
 428                                 fileno == STDOUT_FILENO ? "out" : "err",
 429                                 unit_id, strerror(-r),
 430                                 "ERRNO=%d", -r,
 431                                 NULL);
 432                         r = open_null_as(O_WRONLY, fileno);
 433                 }
 434                 return r;
 435
 436         case EXEC_OUTPUT_SOCKET:
 437                 assert(socket_fd >= 0);
 438                 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
 439
 440         default:
 441                 assert_not_reached("Unknown error type");
 442         }
 443 }
 444
 445 static int chown_terminal(int fd, uid_t uid) {
 446         struct stat st;
 447
 448         assert(fd >= 0);
 449
 450         /* This might fail. What matters are the results. */
 451         (void) fchown(fd, uid, -1);
 452         (void) fchmod(fd, TTY_MODE);
 453
 454         if (fstat(fd, &st) < 0)
 455                 return -errno;
 456
 457         if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
 458                 return -EPERM;
 459
 460         return 0;
 461 }
 462
 463 static int setup_confirm_stdio(int *_saved_stdin,
 464                                int *_saved_stdout) {
 465         int fd = -1, saved_stdin, saved_stdout = -1, r;
 466
 467         assert(_saved_stdin);
 468         assert(_saved_stdout);
 469
 470         saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
 471         if (saved_stdin < 0)
 472                 return -errno;
 473
 474         saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
 475         if (saved_stdout < 0) {
 476                 r = errno;
 477                 goto fail;
 478         }
 479
 480         fd = acquire_terminal(
 481                         "/dev/console",
 482                         false,
 483                         false,
 484                         false,
 485                         DEFAULT_CONFIRM_USEC);
 486         if (fd < 0) {
 487                 r = fd;
 488                 goto fail;
 489         }
 490
 491         r = chown_terminal(fd, getuid());
 492         if (r < 0)
 493                 goto fail;
 494
 495         if (dup2(fd, STDIN_FILENO) < 0) {
 496                 r = -errno;
 497                 goto fail;
 498         }
 499
 500         if (dup2(fd, STDOUT_FILENO) < 0) {
 501                 r = -errno;
 502                 goto fail;
 503         }
 504
 505         if (fd >= 2)
 506                 safe_close(fd);
 507
 508         *_saved_stdin = saved_stdin;
 509         *_saved_stdout = saved_stdout;
 510
 511         return 0;
 512
 513 fail:
 514         safe_close(saved_stdout);
 515         safe_close(saved_stdin);
 516         safe_close(fd);
 517
 518         return r;
 519 }
 520
 521 _printf_(1, 2) static int write_confirm_message(const char *format, ...) {
 522         _cleanup_close_ int fd = -1;
 523         va_list ap;
 524
 525         assert(format);
 526
 527         fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
 528         if (fd < 0)
 529                 return fd;
 530
 531         va_start(ap, format);
 532         vdprintf(fd, format, ap);
 533         va_end(ap);
 534
 535         return 0;
 536 }
 537
 538 static int restore_confirm_stdio(int *saved_stdin,
 539                                  int *saved_stdout) {
 540
 541         int r = 0;
 542
 543         assert(saved_stdin);
 544         assert(saved_stdout);
 545
 546         release_terminal();
 547
 548         if (*saved_stdin >= 0)
 549                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 550                         r = -errno;
 551
 552         if (*saved_stdout >= 0)
 553                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 554                         r = -errno;
 555
 556         safe_close(*saved_stdin);
 557         safe_close(*saved_stdout);
 558
 559         return r;
 560 }
 561
 562 static int ask_for_confirmation(char *response, char **argv) {
 563         int saved_stdout = -1, saved_stdin = -1, r;
 564         _cleanup_free_ char *line = NULL;
 565
 566         r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
 567         if (r < 0)
 568                 return r;
 569
 570         line = exec_command_line(argv);
 571         if (!line)
 572                 return -ENOMEM;
 573
 574         r = ask_char(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
 575
 576         restore_confirm_stdio(&saved_stdin, &saved_stdout);
 577
 578         return r;
 579 }
 580
 581 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
 582         bool keep_groups = false;
 583         int r;
 584
 585         assert(context);
 586
 587         /* Lookup and set GID and supplementary group list. Here too
 588          * we avoid NSS lookups for gid=0. */
 589
 590         if (context->group || username) {
 591
 592                 if (context->group) {
 593                         const char *g = context->group;
 594
 595                         if ((r = get_group_creds(&g, &gid)) < 0)
 596                                 return r;
 597                 }
 598
 599                 /* First step, initialize groups from /etc/groups */
 600                 if (username && gid != 0) {
 601                         if (initgroups(username, gid) < 0)
 602                                 return -errno;
 603
 604                         keep_groups = true;
 605                 }
 606
 607                 /* Second step, set our gids */
 608                 if (setresgid(gid, gid, gid) < 0)
 609                         return -errno;
 610         }
 611
 612         if (context->supplementary_groups) {
 613                 int ngroups_max, k;
 614                 gid_t *gids;
 615                 char **i;
 616
 617                 /* Final step, initialize any manually set supplementary groups */
 618                 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
 619
 620                 if (!(gids = new(gid_t, ngroups_max)))
 621                         return -ENOMEM;
 622
 623                 if (keep_groups) {
 624                         if ((k = getgroups(ngroups_max, gids)) < 0) {
 625                                 free(gids);
 626                                 return -errno;
 627                         }
 628                 } else
 629                         k = 0;
 630
 631                 STRV_FOREACH(i, context->supplementary_groups) {
 632                         const char *g;
 633
 634                         if (k >= ngroups_max) {
 635                                 free(gids);
 636                                 return -E2BIG;
 637                         }
 638
 639                         g = *i;
 640                         r = get_group_creds(&g, gids+k);
 641                         if (r < 0) {
 642                                 free(gids);
 643                                 return r;
 644                         }
 645
 646                         k++;
 647                 }
 648
 649                 if (setgroups(k, gids) < 0) {
 650                         free(gids);
 651                         return -errno;
 652                 }
 653
 654                 free(gids);
 655         }
 656
 657         return 0;
 658 }
 659
 660 static int enforce_user(const ExecContext *context, uid_t uid) {
 661         assert(context);
 662
 663         /* Sets (but doesn't lookup) the uid and make sure we keep the
 664          * capabilities while doing so. */
 665
 666         if (context->capabilities) {
 667                 _cleanup_cap_free_ cap_t d = NULL;
 668                 static const cap_value_t bits[] = {
 669                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
 670                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
 671                 };
 672
 673                 /* First step: If we need to keep capabilities but
 674                  * drop privileges we need to make sure we keep our
 675                  * caps, while we drop privileges. */
 676                 if (uid != 0) {
 677                         int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
 678
 679                         if (prctl(PR_GET_SECUREBITS) != sb)
 680                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 681                                         return -errno;
 682                 }
 683
 684                 /* Second step: set the capabilities. This will reduce
 685                  * the capabilities to the minimum we need. */
 686
 687                 d = cap_dup(context->capabilities);
 688                 if (!d)
 689                         return -errno;
 690
 691                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
 692                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
 693                         return -errno;
 694
 695                 if (cap_set_proc(d) < 0)
 696                         return -errno;
 697         }
 698
 699         /* Third step: actually set the uids */
 700         if (setresuid(uid, uid, uid) < 0)
 701                 return -errno;
 702
 703         /* At this point we should have all necessary capabilities but
 704            are otherwise a normal user. However, the caps might got
 705            corrupted due to the setresuid() so we need clean them up
 706            later. This is done outside of this call. */
 707
 708         return 0;
 709 }
 710
 711 #ifdef HAVE_PAM
 712
 713 static int null_conv(
 714                 int num_msg,
 715                 const struct pam_message **msg,
 716                 struct pam_response **resp,
 717                 void *appdata_ptr) {
 718
 719         /* We don't support conversations */
 720
 721         return PAM_CONV_ERR;
 722 }
 723
 724 static int setup_pam(
 725                 const char *name,
 726                 const char *user,
 727                 uid_t uid,
 728                 const char *tty,
 729                 char ***pam_env,
 730                 int fds[], unsigned n_fds) {
 731
 732         static const struct pam_conv conv = {
 733                 .conv = null_conv,
 734                 .appdata_ptr = NULL
 735         };
 736
 737         pam_handle_t *handle = NULL;
 738         sigset_t ss, old_ss;
 739         int pam_code = PAM_SUCCESS;
 740         int err;
 741         char **e = NULL;
 742         bool close_session = false;
 743         pid_t pam_pid = 0, parent_pid;
 744         int flags = 0;
 745
 746         assert(name);
 747         assert(user);
 748         assert(pam_env);
 749
 750         /* We set up PAM in the parent process, then fork. The child
 751          * will then stay around until killed via PR_GET_PDEATHSIG or
 752          * systemd via the cgroup logic. It will then remove the PAM
 753          * session again. The parent process will exec() the actual
 754          * daemon. We do things this way to ensure that the main PID
 755          * of the daemon is the one we initially fork()ed. */
 756
 757         if (log_get_max_level() < LOG_PRI(LOG_DEBUG))
 758                 flags |= PAM_SILENT;
 759
 760         pam_code = pam_start(name, user, &conv, &handle);
 761         if (pam_code != PAM_SUCCESS) {
 762                 handle = NULL;
 763                 goto fail;
 764         }
 765
 766         if (tty) {
 767                 pam_code = pam_set_item(handle, PAM_TTY, tty);
 768                 if (pam_code != PAM_SUCCESS)
 769                         goto fail;
 770         }
 771
 772         pam_code = pam_acct_mgmt(handle, flags);
 773         if (pam_code != PAM_SUCCESS)
 774                 goto fail;
 775
 776         pam_code = pam_open_session(handle, flags);
 777         if (pam_code != PAM_SUCCESS)
 778                 goto fail;
 779
 780         close_session = true;
 781
 782         e = pam_getenvlist(handle);
 783         if (!e) {
 784                 pam_code = PAM_BUF_ERR;
 785                 goto fail;
 786         }
 787
 788         /* Block SIGTERM, so that we know that it won't get lost in
 789          * the child */
 790         if (sigemptyset(&ss) < 0 ||
 791             sigaddset(&ss, SIGTERM) < 0 ||
 792             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 793                 goto fail;
 794
 795         parent_pid = getpid();
 796
 797         pam_pid = fork();
 798         if (pam_pid < 0)
 799                 goto fail;
 800
 801         if (pam_pid == 0) {
 802                 int sig;
 803                 int r = EXIT_PAM;
 804
 805                 /* The child's job is to reset the PAM session on
 806                  * termination */
 807
 808                 /* This string must fit in 10 chars (i.e. the length
 809                  * of "/sbin/init"), to look pretty in /bin/ps */
 810                 rename_process("(sd-pam)");
 811
 812                 /* Make sure we don't keep open the passed fds in this
 813                 child. We assume that otherwise only those fds are
 814                 open here that have been opened by PAM. */
 815                 close_many(fds, n_fds);
 816
 817                 /* Drop privileges - we don't need any to pam_close_session
 818                  * and this will make PR_SET_PDEATHSIG work in most cases.
 819                  * If this fails, ignore the error - but expect sd-pam threads
 820                  * to fail to exit normally */
 821                 if (setresuid(uid, uid, uid) < 0)
 822                         log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r));
 823
 824                 /* Wait until our parent died. This will only work if
 825                  * the above setresuid() succeeds, otherwise the kernel
 826                  * will not allow unprivileged parents kill their privileged
 827                  * children this way. We rely on the control groups kill logic
 828                  * to do the rest for us. */
 829                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 830                         goto child_finish;
 831
 832                 /* Check if our parent process might already have
 833                  * died? */
 834                 if (getppid() == parent_pid) {
 835                         for (;;) {
 836                                 if (sigwait(&ss, &sig) < 0) {
 837                                         if (errno == EINTR)
 838                                                 continue;
 839
 840                                         goto child_finish;
 841                                 }
 842
 843                                 assert(sig == SIGTERM);
 844                                 break;
 845                         }
 846                 }
 847
 848                 /* If our parent died we'll end the session */
 849                 if (getppid() != parent_pid) {
 850                         pam_code = pam_close_session(handle, flags);
 851                         if (pam_code != PAM_SUCCESS)
 852                                 goto child_finish;
 853                 }
 854
 855                 r = 0;
 856
 857         child_finish:
 858                 pam_end(handle, pam_code | flags);
 859                 _exit(r);
 860         }
 861
 862         /* If the child was forked off successfully it will do all the
 863          * cleanups, so forget about the handle here. */
 864         handle = NULL;
 865
 866         /* Unblock SIGTERM again in the parent */
 867         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 868                 goto fail;
 869
 870         /* We close the log explicitly here, since the PAM modules
 871          * might have opened it, but we don't want this fd around. */
 872         closelog();
 873
 874         *pam_env = e;
 875         e = NULL;
 876
 877         return 0;
 878
 879 fail:
 880         if (pam_code != PAM_SUCCESS) {
 881                 log_error("PAM failed: %s", pam_strerror(handle, pam_code));
 882                 err = -EPERM;  /* PAM errors do not map to errno */
 883         } else {
 884                 log_error("PAM failed: %m");
 885                 err = -errno;
 886         }
 887
 888         if (handle) {
 889                 if (close_session)
 890                         pam_code = pam_close_session(handle, flags);
 891
 892                 pam_end(handle, pam_code | flags);
 893         }
 894
 895         strv_free(e);
 896
 897         closelog();
 898
 899         if (pam_pid > 1) {
 900                 kill(pam_pid, SIGTERM);
 901                 kill(pam_pid, SIGCONT);
 902         }
 903
 904         return err;
 905 }
 906 #endif
 907
 908 static void rename_process_from_path(const char *path) {
 909         char process_name[11];
 910         const char *p;
 911         size_t l;
 912
 913         /* This resulting string must fit in 10 chars (i.e. the length
 914          * of "/sbin/init") to look pretty in /bin/ps */
 915
 916         p = basename(path);
 917         if (isempty(p)) {
 918                 rename_process("(...)");
 919                 return;
 920         }
 921
 922         l = strlen(p);
 923         if (l > 8) {
 924                 /* The end of the process name is usually more
 925                  * interesting, since the first bit might just be
 926                  * "systemd-" */
 927                 p = p + l - 8;
 928                 l = 8;
 929         }
 930
 931         process_name[0] = '(';
 932         memcpy(process_name+1, p, l);
 933         process_name[1+l] = ')';
 934         process_name[1+l+1] = 0;
 935
 936         rename_process(process_name);
 937 }
 938
 939 #ifdef HAVE_SECCOMP
 940
 941 static int apply_seccomp(ExecContext *c) {
 942         uint32_t negative_action, action;
 943         scmp_filter_ctx *seccomp;
 944         Iterator i;
 945         void *id;
 946         int r;
 947
 948         assert(c);
 949
 950         negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
 951
 952         seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW);
 953         if (!seccomp)
 954                 return -ENOMEM;
 955
 956         if (c->syscall_archs) {
 957
 958                 SET_FOREACH(id, c->syscall_archs, i) {
 959                         r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
 960                         if (r == -EEXIST)
 961                                 continue;
 962                         if (r < 0)
 963                                 goto finish;
 964                 }
 965
 966         } else {
 967                 r = seccomp_add_secondary_archs(seccomp);
 968                 if (r < 0)
 969                         goto finish;
 970         }
 971
 972         action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action;
 973         SET_FOREACH(id, c->syscall_filter, i) {
 974                 r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0);
 975                 if (r < 0)
 976                         goto finish;
 977         }
 978
 979         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
 980         if (r < 0)
 981                 goto finish;
 982
 983         r = seccomp_load(seccomp);
 984
 985 finish:
 986         seccomp_release(seccomp);
 987         return r;
 988 }
 989
 990 static int apply_address_families(ExecContext *c) {
 991         scmp_filter_ctx *seccomp;
 992         Iterator i;
 993         int r;
 994
 995         assert(c);
 996
 997         seccomp = seccomp_init(SCMP_ACT_ALLOW);
 998         if (!seccomp)
 999                 return -ENOMEM;
1000
1001         r = seccomp_add_secondary_archs(seccomp);
1002         if (r < 0)
1003                 goto finish;
1004
1005         if (c->address_families_whitelist) {
1006                 int af, first = 0, last = 0;
1007                 void *afp;
1008
1009                 /* If this is a whitelist, we first block the address
1010                  * families that are out of range and then everything
1011                  * that is not in the set. First, we find the lowest
1012                  * and highest address family in the set. */
1013
1014                 SET_FOREACH(afp, c->address_families, i) {
1015                         af = PTR_TO_INT(afp);
1016
1017                         if (af <= 0 || af >= af_max())
1018                                 continue;
1019
1020                         if (first == 0 || af < first)
1021                                 first = af;
1022
1023                         if (last == 0 || af > last)
1024                                 last = af;
1025                 }
1026
1027                 assert((first == 0) == (last == 0));
1028
1029                 if (first == 0) {
1030
1031                         /* No entries in the valid range, block everything */
1032                         r = seccomp_rule_add(
1033                                         seccomp,
1034                                         SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1035                                         SCMP_SYS(socket),
1036                                         0);
1037                         if (r < 0)
1038                                 goto finish;
1039
1040                 } else {
1041
1042                         /* Block everything below the first entry */
1043                         r = seccomp_rule_add(
1044                                         seccomp,
1045                                         SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1046                                         SCMP_SYS(socket),
1047                                         1,
1048                                         SCMP_A0(SCMP_CMP_LT, first));
1049                         if (r < 0)
1050                                 goto finish;
1051
1052                         /* Block everything above the last entry */
1053                         r = seccomp_rule_add(
1054                                         seccomp,
1055                                         SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1056                                         SCMP_SYS(socket),
1057                                         1,
1058                                         SCMP_A0(SCMP_CMP_GT, last));
1059                         if (r < 0)
1060                                 goto finish;
1061
1062                         /* Block everything between the first and last
1063                          * entry */
1064                         for (af = 1; af < af_max(); af++) {
1065
1066                                 if (set_contains(c->address_families, INT_TO_PTR(af)))
1067                                         continue;
1068
1069                                 r = seccomp_rule_add(
1070                                                 seccomp,
1071                                                 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1072                                                 SCMP_SYS(socket),
1073                                                 1,
1074                                                 SCMP_A0(SCMP_CMP_EQ, af));
1075                                 if (r < 0)
1076                                         goto finish;
1077                         }
1078                 }
1079
1080         } else {
1081                 void *af;
1082
1083                 /* If this is a blacklist, then generate one rule for
1084                  * each address family that are then combined in OR
1085                  * checks. */
1086
1087                 SET_FOREACH(af, c->address_families, i) {
1088
1089                         r = seccomp_rule_add(
1090                                         seccomp,
1091                                         SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1092                                         SCMP_SYS(socket),
1093                                         1,
1094                                         SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af)));
1095                         if (r < 0)
1096                                 goto finish;
1097                 }
1098         }
1099
1100         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1101         if (r < 0)
1102                 goto finish;
1103
1104         r = seccomp_load(seccomp);
1105
1106 finish:
1107         seccomp_release(seccomp);
1108         return r;
1109 }
1110
1111 #endif
1112
1113 static void do_idle_pipe_dance(int idle_pipe[4]) {
1114         assert(idle_pipe);
1115
1116
1117         safe_close(idle_pipe[1]);
1118         safe_close(idle_pipe[2]);
1119
1120         if (idle_pipe[0] >= 0) {
1121                 int r;
1122
1123                 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1124
1125                 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
1126                         /* Signal systemd that we are bored and want to continue. */
1127                         write(idle_pipe[3], "x", 1);
1128
1129                         /* Wait for systemd to react to the signal above. */
1130                         fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
1131                 }
1132
1133                 safe_close(idle_pipe[0]);
1134
1135         }
1136
1137         safe_close(idle_pipe[3]);
1138 }
1139
1140 static int build_environment(
1141                 const ExecContext *c,
1142                 unsigned n_fds,
1143                 usec_t watchdog_usec,
1144                 const char *home,
1145                 const char *username,
1146                 const char *shell,
1147                 char ***ret) {
1148
1149         _cleanup_strv_free_ char **our_env = NULL;
1150         unsigned n_env = 0;
1151         char *x;
1152
1153         assert(c);
1154         assert(ret);
1155
1156         our_env = new0(char*, 10);
1157         if (!our_env)
1158                 return -ENOMEM;
1159
1160         if (n_fds > 0) {
1161                 if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
1162                         return -ENOMEM;
1163                 our_env[n_env++] = x;
1164
1165                 if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
1166                         return -ENOMEM;
1167                 our_env[n_env++] = x;
1168         }
1169
1170         if (watchdog_usec > 0) {
1171                 if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
1172                         return -ENOMEM;
1173                 our_env[n_env++] = x;
1174
1175                 if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, watchdog_usec) < 0)
1176                         return -ENOMEM;
1177                 our_env[n_env++] = x;
1178         }
1179
1180         if (home) {
1181                 x = strappend("HOME=", home);
1182                 if (!x)
1183                         return -ENOMEM;
1184                 our_env[n_env++] = x;
1185         }
1186
1187         if (username) {
1188                 x = strappend("LOGNAME=", username);
1189                 if (!x)
1190                         return -ENOMEM;
1191                 our_env[n_env++] = x;
1192
1193                 x = strappend("USER=", username);
1194                 if (!x)
1195                         return -ENOMEM;
1196                 our_env[n_env++] = x;
1197         }
1198
1199         if (shell) {
1200                 x = strappend("SHELL=", shell);
1201                 if (!x)
1202                         return -ENOMEM;
1203                 our_env[n_env++] = x;
1204         }
1205
1206         if (is_terminal_input(c->std_input) ||
1207             c->std_output == EXEC_OUTPUT_TTY ||
1208             c->std_error == EXEC_OUTPUT_TTY ||
1209             c->tty_path) {
1210
1211                 x = strdup(default_term_for_tty(tty_path(c)));
1212                 if (!x)
1213                         return -ENOMEM;
1214                 our_env[n_env++] = x;
1215         }
1216
1217         our_env[n_env++] = NULL;
1218         assert(n_env <= 10);
1219
1220         *ret = our_env;
1221         our_env = NULL;
1222
1223         return 0;
1224 }
1225
1226 static int exec_child(ExecCommand *command,
1227                       const ExecContext *context,
1228                       const ExecParameters *params,
1229                       ExecRuntime *runtime,
1230                       char **argv,
1231                       int socket_fd,
1232                       int *fds, unsigned n_fds,
1233                       char **files_env,
1234                       int *error) {
1235
1236         _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1237         const char *username = NULL, *home = NULL, *shell = NULL;
1238         unsigned n_dont_close = 0;
1239         int dont_close[n_fds + 3];
1240         uid_t uid = (uid_t) -1;
1241         gid_t gid = (gid_t) -1;
1242         int i, err;
1243
1244         assert(command);
1245         assert(context);
1246         assert(params);
1247         assert(error);
1248
1249         rename_process_from_path(command->path);
1250
1251         /* We reset exactly these signals, since they are the
1252          * only ones we set to SIG_IGN in the main daemon. All
1253          * others we leave untouched because we set them to
1254          * SIG_DFL or a valid handler initially, both of which
1255          * will be demoted to SIG_DFL. */
1256         default_signals(SIGNALS_CRASH_HANDLER,
1257                         SIGNALS_IGNORE, -1);
1258
1259         if (context->ignore_sigpipe)
1260                 ignore_signals(SIGPIPE, -1);
1261
1262         err = reset_signal_mask();
1263         if (err < 0) {
1264                 *error = EXIT_SIGNAL_MASK;
1265                 return err;
1266         }
1267
1268         if (params->idle_pipe)
1269                 do_idle_pipe_dance(params->idle_pipe);
1270
1271         /* Close sockets very early to make sure we don't
1272          * block init reexecution because it cannot bind its
1273          * sockets */
1274         log_forget_fds();
1275
1276         if (socket_fd >= 0)
1277                 dont_close[n_dont_close++] = socket_fd;
1278         if (n_fds > 0) {
1279                 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
1280                 n_dont_close += n_fds;
1281         }
1282         if (runtime) {
1283                 if (runtime->netns_storage_socket[0] >= 0)
1284                         dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
1285                 if (runtime->netns_storage_socket[1] >= 0)
1286                         dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
1287         }
1288
1289         err = close_all_fds(dont_close, n_dont_close);
1290         if (err < 0) {
1291                 *error = EXIT_FDS;
1292                 return err;
1293         }
1294
1295         if (!context->same_pgrp)
1296                 if (setsid() < 0) {
1297                         *error = EXIT_SETSID;
1298                         return -errno;
1299                 }
1300
1301         exec_context_tty_reset(context);
1302
1303         if (params->confirm_spawn) {
1304                 char response;
1305
1306                 err = ask_for_confirmation(&response, argv);
1307                 if (err == -ETIMEDOUT)
1308                         write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1309                 else if (err < 0)
1310                         write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1311                 else if (response == 's') {
1312                         write_confirm_message("Skipping execution.\n");
1313                         *error = EXIT_CONFIRM;
1314                         return -ECANCELED;
1315                 } else if (response == 'n') {
1316                         write_confirm_message("Failing execution.\n");
1317                         *error = 0;
1318                         return 0;
1319                 }
1320         }
1321
1322         /* If a socket is connected to STDIN/STDOUT/STDERR, we
1323          * must sure to drop O_NONBLOCK */
1324         if (socket_fd >= 0)
1325                 fd_nonblock(socket_fd, false);
1326
1327         err = setup_input(context, socket_fd, params->apply_tty_stdin);
1328         if (err < 0) {
1329                 *error = EXIT_STDIN;
1330                 return err;
1331         }
1332
1333         err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin);
1334         if (err < 0) {
1335                 *error = EXIT_STDOUT;
1336                 return err;
1337         }
1338
1339         err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin);
1340         if (err < 0) {
1341                 *error = EXIT_STDERR;
1342                 return err;
1343         }
1344
1345         if (params->cgroup_path) {
1346                 err = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0);
1347                 if (err < 0) {
1348                         *error = EXIT_CGROUP;
1349                         return err;
1350                 }
1351         }
1352
1353         if (context->oom_score_adjust_set) {
1354                 char t[16];
1355
1356                 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1357                 char_array_0(t);
1358
1359                 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1360                         *error = EXIT_OOM_ADJUST;
1361                         return -errno;
1362                 }
1363         }
1364
1365         if (context->nice_set)
1366                 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1367                         *error = EXIT_NICE;
1368                         return -errno;
1369                 }
1370
1371         if (context->cpu_sched_set) {
1372                 struct sched_param param = {
1373                         .sched_priority = context->cpu_sched_priority,
1374                 };
1375
1376                 err = sched_setscheduler(0,
1377                                          context->cpu_sched_policy |
1378                                          (context->cpu_sched_reset_on_fork ?
1379                                           SCHED_RESET_ON_FORK : 0),
1380                                          &param);
1381                 if (err < 0) {
1382                         *error = EXIT_SETSCHEDULER;
1383                         return -errno;
1384                 }
1385         }
1386
1387         if (context->cpuset)
1388                 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1389                         *error = EXIT_CPUAFFINITY;
1390                         return -errno;
1391                 }
1392
1393         if (context->ioprio_set)
1394                 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1395                         *error = EXIT_IOPRIO;
1396                         return -errno;
1397                 }
1398
1399         if (context->timer_slack_nsec != NSEC_INFINITY)
1400                 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1401                         *error = EXIT_TIMERSLACK;
1402                         return -errno;
1403                 }
1404
1405         if (context->personality != 0xffffffffUL)
1406                 if (personality(context->personality) < 0) {
1407                         *error = EXIT_PERSONALITY;
1408                         return -errno;
1409                 }
1410
1411         if (context->utmp_id)
1412                 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1413
1414         if (context->user) {
1415                 username = context->user;
1416                 err = get_user_creds(&username, &uid, &gid, &home, &shell);
1417                 if (err < 0) {
1418                         *error = EXIT_USER;
1419                         return err;
1420                 }
1421
1422                 if (is_terminal_input(context->std_input)) {
1423                         err = chown_terminal(STDIN_FILENO, uid);
1424                         if (err < 0) {
1425                                 *error = EXIT_STDIN;
1426                                 return err;
1427                         }
1428                 }
1429         }
1430
1431 #ifdef HAVE_PAM
1432         if (params->cgroup_path && context->user && context->pam_name) {
1433                 err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid);
1434                 if (err < 0) {
1435                         *error = EXIT_CGROUP;
1436                         return err;
1437                 }
1438
1439
1440                 err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid);
1441                 if (err < 0) {
1442                         *error = EXIT_CGROUP;
1443                         return err;
1444                 }
1445         }
1446 #endif
1447
1448         if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) {
1449                 char **rt;
1450
1451                 STRV_FOREACH(rt, context->runtime_directory) {
1452                         _cleanup_free_ char *p;
1453
1454                         p = strjoin(params->runtime_prefix, "/", *rt, NULL);
1455                         if (!p) {
1456                                 *error = EXIT_RUNTIME_DIRECTORY;
1457                                 return -ENOMEM;
1458                         }
1459
1460                         err = mkdir_safe(p, context->runtime_directory_mode, uid, gid);
1461                         if (err < 0) {
1462                                 *error = EXIT_RUNTIME_DIRECTORY;
1463                                 return err;
1464                         }
1465                 }
1466         }
1467
1468         if (params->apply_permissions) {
1469                 err = enforce_groups(context, username, gid);
1470                 if (err < 0) {
1471                         *error = EXIT_GROUP;
1472                         return err;
1473                 }
1474         }
1475
1476         umask(context->umask);
1477
1478 #ifdef HAVE_PAM
1479         if (params->apply_permissions && context->pam_name && username) {
1480                 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1481                 if (err < 0) {
1482                         *error = EXIT_PAM;
1483                         return err;
1484                 }
1485         }
1486 #endif
1487
1488         if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
1489                 err = setup_netns(runtime->netns_storage_socket);
1490                 if (err < 0) {
1491                         *error = EXIT_NETWORK;
1492                         return err;
1493                 }
1494         }
1495
1496         if (!strv_isempty(context->read_write_dirs) ||
1497             !strv_isempty(context->read_only_dirs) ||
1498             !strv_isempty(context->inaccessible_dirs) ||
1499             context->mount_flags != 0 ||
1500             (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
1501             context->private_devices ||
1502             context->protect_system != PROTECT_SYSTEM_NO ||
1503             context->protect_home != PROTECT_HOME_NO) {
1504
1505                 char *tmp = NULL, *var = NULL;
1506
1507                 /* The runtime struct only contains the parent
1508                  * of the private /tmp, which is
1509                  * non-accessible to world users. Inside of it
1510                  * there's a /tmp that is sticky, and that's
1511                  * the one we want to use here. */
1512
1513                 if (context->private_tmp && runtime) {
1514                         if (runtime->tmp_dir)
1515                                 tmp = strappenda(runtime->tmp_dir, "/tmp");
1516                         if (runtime->var_tmp_dir)
1517                                 var = strappenda(runtime->var_tmp_dir, "/tmp");
1518                 }
1519
1520                 err = setup_namespace(
1521                                 context->read_write_dirs,
1522                                 context->read_only_dirs,
1523                                 context->inaccessible_dirs,
1524                                 tmp,
1525                                 var,
1526                                 context->private_devices,
1527                                 context->protect_home,
1528                                 context->protect_system,
1529                                 context->mount_flags);
1530                 if (err < 0) {
1531                         *error = EXIT_NAMESPACE;
1532                         return err;
1533                 }
1534         }
1535
1536         if (params->apply_chroot) {
1537                 if (context->root_directory)
1538                         if (chroot(context->root_directory) < 0) {
1539                                 *error = EXIT_CHROOT;
1540                                 return -errno;
1541                         }
1542
1543                 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1544                         *error = EXIT_CHDIR;
1545                         return -errno;
1546                 }
1547         } else {
1548                 _cleanup_free_ char *d = NULL;
1549
1550                 if (asprintf(&d, "%s/%s",
1551                              context->root_directory ? context->root_directory : "",
1552                              context->working_directory ? context->working_directory : "") < 0) {
1553                         *error = EXIT_MEMORY;
1554                         return -ENOMEM;
1555                 }
1556
1557                 if (chdir(d) < 0) {
1558                         *error = EXIT_CHDIR;
1559                         return -errno;
1560                 }
1561         }
1562
1563         /* We repeat the fd closing here, to make sure that
1564          * nothing is leaked from the PAM modules. Note that
1565          * we are more aggressive this time since socket_fd
1566          * and the netns fds we don#t need anymore. */
1567         err = close_all_fds(fds, n_fds);
1568         if (err >= 0)
1569                 err = shift_fds(fds, n_fds);
1570         if (err >= 0)
1571                 err = flags_fds(fds, n_fds, context->non_blocking);
1572         if (err < 0) {
1573                 *error = EXIT_FDS;
1574                 return err;
1575         }
1576
1577         if (params->apply_permissions) {
1578
1579                 for (i = 0; i < _RLIMIT_MAX; i++) {
1580                         if (!context->rlimit[i])
1581                                 continue;
1582
1583                         if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1584                                 *error = EXIT_LIMITS;
1585                                 return -errno;
1586                         }
1587                 }
1588
1589                 if (context->capability_bounding_set_drop) {
1590                         err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1591                         if (err < 0) {
1592                                 *error = EXIT_CAPABILITIES;
1593                                 return err;
1594                         }
1595                 }
1596
1597                 if (context->user) {
1598                         err = enforce_user(context, uid);
1599                         if (err < 0) {
1600                                 *error = EXIT_USER;
1601                                 return err;
1602                         }
1603                 }
1604
1605                 /* PR_GET_SECUREBITS is not privileged, while
1606                  * PR_SET_SECUREBITS is. So to suppress
1607                  * potential EPERMs we'll try not to call
1608                  * PR_SET_SECUREBITS unless necessary. */
1609                 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1610                         if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1611                                 *error = EXIT_SECUREBITS;
1612                                 return -errno;
1613                         }
1614
1615                 if (context->capabilities)
1616                         if (cap_set_proc(context->capabilities) < 0) {
1617                                 *error = EXIT_CAPABILITIES;
1618                                 return -errno;
1619                         }
1620
1621                 if (context->no_new_privileges)
1622                         if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1623                                 *error = EXIT_NO_NEW_PRIVILEGES;
1624                                 return -errno;
1625                         }
1626
1627 #ifdef HAVE_SECCOMP
1628                 if (context->address_families_whitelist ||
1629                     !set_isempty(context->address_families)) {
1630                         err = apply_address_families(context);
1631                         if (err < 0) {
1632                                 *error = EXIT_ADDRESS_FAMILIES;
1633                                 return err;
1634                         }
1635                 }
1636
1637                 if (context->syscall_whitelist ||
1638                     !set_isempty(context->syscall_filter) ||
1639                     !set_isempty(context->syscall_archs)) {
1640                         err = apply_seccomp(context);
1641                         if (err < 0) {
1642                                 *error = EXIT_SECCOMP;
1643                                 return err;
1644                         }
1645                 }
1646 #endif
1647
1648 #ifdef HAVE_SELINUX
1649                 if (context->selinux_context && use_selinux()) {
1650                         err = setexeccon(context->selinux_context);
1651                         if (err < 0 && !context->selinux_context_ignore) {
1652                                 *error = EXIT_SELINUX_CONTEXT;
1653                                 return err;
1654                         }
1655                 }
1656 #endif
1657
1658 #ifdef HAVE_APPARMOR
1659                 if (context->apparmor_profile && use_apparmor()) {
1660                         err = aa_change_onexec(context->apparmor_profile);
1661                         if (err < 0 && !context->apparmor_profile_ignore) {
1662                                 *error = EXIT_APPARMOR_PROFILE;
1663                                 return err;
1664                         }
1665                 }
1666 #endif
1667         }
1668
1669         err = build_environment(context, n_fds, params->watchdog_usec, home, username, shell, &our_env);
1670         if (err < 0) {
1671                 *error = EXIT_MEMORY;
1672                 return err;
1673         }
1674
1675         final_env = strv_env_merge(5,
1676                                    params->environment,
1677                                    our_env,
1678                                    context->environment,
1679                                    files_env,
1680                                    pam_env,
1681                                    NULL);
1682         if (!final_env) {
1683                 *error = EXIT_MEMORY;
1684                 return -ENOMEM;
1685         }
1686
1687         final_argv = replace_env_argv(argv, final_env);
1688         if (!final_argv) {
1689                 *error = EXIT_MEMORY;
1690                 return -ENOMEM;
1691         }
1692
1693         final_env = strv_env_clean(final_env);
1694
1695         if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
1696                 _cleanup_free_ char *line;
1697
1698                 line = exec_command_line(final_argv);
1699                 if (line) {
1700                         log_open();
1701                         log_struct_unit(LOG_DEBUG,
1702                                         params->unit_id,
1703                                         "EXECUTABLE=%s", command->path,
1704                                         "MESSAGE=Executing: %s", line,
1705                                         NULL);
1706                         log_close();
1707                 }
1708         }
1709         execve(command->path, final_argv, final_env);
1710         *error = EXIT_EXEC;
1711         return -errno;
1712 }
1713
1714 int exec_spawn(ExecCommand *command,
1715                const ExecContext *context,
1716                const ExecParameters *params,
1717                ExecRuntime *runtime,
1718                pid_t *ret) {
1719
1720         _cleanup_strv_free_ char **files_env = NULL;
1721         int *fds = NULL; unsigned n_fds = 0;
1722         char *line, **argv;
1723         int socket_fd;
1724         pid_t pid;
1725         int err;
1726
1727         assert(command);
1728         assert(context);
1729         assert(ret);
1730         assert(params);
1731         assert(params->fds || params->n_fds <= 0);
1732
1733         if (context->std_input == EXEC_INPUT_SOCKET ||
1734             context->std_output == EXEC_OUTPUT_SOCKET ||
1735             context->std_error == EXEC_OUTPUT_SOCKET) {
1736
1737                 if (params->n_fds != 1)
1738                         return -EINVAL;
1739
1740                 socket_fd = params->fds[0];
1741         } else {
1742                 socket_fd = -1;
1743                 fds = params->fds;
1744                 n_fds = params->n_fds;
1745         }
1746
1747         err = exec_context_load_environment(context, &files_env);
1748         if (err < 0) {
1749                 log_struct_unit(LOG_ERR,
1750                            params->unit_id,
1751                            "MESSAGE=Failed to load environment files: %s", strerror(-err),
1752                            "ERRNO=%d", -err,
1753                            NULL);
1754                 return err;
1755         }
1756
1757         argv = params->argv ?: command->argv;
1758
1759         line = exec_command_line(argv);
1760         if (!line)
1761                 return log_oom();
1762
1763         log_struct_unit(LOG_DEBUG,
1764                         params->unit_id,
1765                         "EXECUTABLE=%s", command->path,
1766                         "MESSAGE=About to execute: %s", line,
1767                         NULL);
1768         free(line);
1769
1770         pid = fork();
1771         if (pid < 0)
1772                 return -errno;
1773
1774         if (pid == 0) {
1775                 int r;
1776
1777                 err = exec_child(command,
1778                                  context,
1779                                  params,
1780                                  runtime,
1781                                  argv,
1782                                  socket_fd,
1783                                  fds, n_fds,
1784                                  files_env,
1785                                  &r);
1786                 if (r != 0) {
1787                         log_open();
1788                         log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1789                                    "EXECUTABLE=%s", command->path,
1790                                    "MESSAGE=Failed at step %s spawning %s: %s",
1791                                           exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1792                                           command->path, strerror(-err),
1793                                    "ERRNO=%d", -err,
1794                                    NULL);
1795                         log_close();
1796                 }
1797
1798                 _exit(r);
1799         }
1800
1801         log_struct_unit(LOG_DEBUG,
1802                         params->unit_id,
1803                         "MESSAGE=Forked %s as "PID_FMT,
1804                         command->path, pid,
1805                         NULL);
1806
1807         /* We add the new process to the cgroup both in the child (so
1808          * that we can be sure that no user code is ever executed
1809          * outside of the cgroup) and in the parent (so that we can be
1810          * sure that when we kill the cgroup the process will be
1811          * killed too). */
1812         if (params->cgroup_path)
1813                 cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid);
1814
1815         exec_status_start(&command->exec_status, pid);
1816
1817         *ret = pid;
1818         return 0;
1819 }
1820
1821 void exec_context_init(ExecContext *c) {
1822         assert(c);
1823
1824         c->umask = 0022;
1825         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1826         c->cpu_sched_policy = SCHED_OTHER;
1827         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1828         c->syslog_level_prefix = true;
1829         c->ignore_sigpipe = true;
1830         c->timer_slack_nsec = NSEC_INFINITY;
1831         c->personality = 0xffffffffUL;
1832         c->runtime_directory_mode = 0755;
1833 }
1834
1835 void exec_context_done(ExecContext *c) {
1836         unsigned l;
1837
1838         assert(c);
1839
1840         strv_free(c->environment);
1841         c->environment = NULL;
1842
1843         strv_free(c->environment_files);
1844         c->environment_files = NULL;
1845
1846         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1847                 free(c->rlimit[l]);
1848                 c->rlimit[l] = NULL;
1849         }
1850
1851         free(c->working_directory);
1852         c->working_directory = NULL;
1853         free(c->root_directory);
1854         c->root_directory = NULL;
1855
1856         free(c->tty_path);
1857         c->tty_path = NULL;
1858
1859         free(c->syslog_identifier);
1860         c->syslog_identifier = NULL;
1861
1862         free(c->user);
1863         c->user = NULL;
1864
1865         free(c->group);
1866         c->group = NULL;
1867
1868         strv_free(c->supplementary_groups);
1869         c->supplementary_groups = NULL;
1870
1871         free(c->pam_name);
1872         c->pam_name = NULL;
1873
1874         if (c->capabilities) {
1875                 cap_free(c->capabilities);
1876                 c->capabilities = NULL;
1877         }
1878
1879         strv_free(c->read_only_dirs);
1880         c->read_only_dirs = NULL;
1881
1882         strv_free(c->read_write_dirs);
1883         c->read_write_dirs = NULL;
1884
1885         strv_free(c->inaccessible_dirs);
1886         c->inaccessible_dirs = NULL;
1887
1888         if (c->cpuset)
1889                 CPU_FREE(c->cpuset);
1890
1891         free(c->utmp_id);
1892         c->utmp_id = NULL;
1893
1894         free(c->selinux_context);
1895         c->selinux_context = NULL;
1896
1897         free(c->apparmor_profile);
1898         c->apparmor_profile = NULL;
1899
1900         set_free(c->syscall_filter);
1901         c->syscall_filter = NULL;
1902
1903         set_free(c->syscall_archs);
1904         c->syscall_archs = NULL;
1905
1906         set_free(c->address_families);
1907         c->address_families = NULL;
1908
1909         strv_free(c->runtime_directory);
1910         c->runtime_directory = NULL;
1911
1912         bus_endpoint_free(c->bus_endpoint);
1913         c->bus_endpoint = NULL;
1914 }
1915
1916 int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
1917         char **i;
1918
1919         assert(c);
1920
1921         if (!runtime_prefix)
1922                 return 0;
1923
1924         STRV_FOREACH(i, c->runtime_directory) {
1925                 _cleanup_free_ char *p;
1926
1927                 p = strjoin(runtime_prefix, "/", *i, NULL);
1928                 if (!p)
1929                         return -ENOMEM;
1930
1931                 /* We execute this synchronously, since we need to be
1932                  * sure this is gone when we start the service
1933                  * next. */
1934                 rm_rf_dangerous(p, false, true, false);
1935         }
1936
1937         return 0;
1938 }
1939
1940 void exec_command_done(ExecCommand *c) {
1941         assert(c);
1942
1943         free(c->path);
1944         c->path = NULL;
1945
1946         strv_free(c->argv);
1947         c->argv = NULL;
1948 }
1949
1950 void exec_command_done_array(ExecCommand *c, unsigned n) {
1951         unsigned i;
1952
1953         for (i = 0; i < n; i++)
1954                 exec_command_done(c+i);
1955 }
1956
1957 void exec_command_free_list(ExecCommand *c) {
1958         ExecCommand *i;
1959
1960         while ((i = c)) {
1961                 LIST_REMOVE(command, c, i);
1962                 exec_command_done(i);
1963                 free(i);
1964         }
1965 }
1966
1967 void exec_command_free_array(ExecCommand **c, unsigned n) {
1968         unsigned i;
1969
1970         for (i = 0; i < n; i++) {
1971                 exec_command_free_list(c[i]);
1972                 c[i] = NULL;
1973         }
1974 }
1975
1976 int exec_context_load_environment(const ExecContext *c, char ***l) {
1977         char **i, **r = NULL;
1978
1979         assert(c);
1980         assert(l);
1981
1982         STRV_FOREACH(i, c->environment_files) {
1983                 char *fn;
1984                 int k;
1985                 bool ignore = false;
1986                 char **p;
1987                 _cleanup_globfree_ glob_t pglob = {};
1988                 int count, n;
1989
1990                 fn = *i;
1991
1992                 if (fn[0] == '-') {
1993                         ignore = true;
1994                         fn ++;
1995                 }
1996
1997                 if (!path_is_absolute(fn)) {
1998                         if (ignore)
1999                                 continue;
2000
2001                         strv_free(r);
2002                         return -EINVAL;
2003                 }
2004
2005                 /* Filename supports globbing, take all matching files */
2006                 errno = 0;
2007                 if (glob(fn, 0, NULL, &pglob) != 0) {
2008                         if (ignore)
2009                                 continue;
2010
2011                         strv_free(r);
2012                         return errno ? -errno : -EINVAL;
2013                 }
2014                 count = pglob.gl_pathc;
2015                 if (count == 0) {
2016                         if (ignore)
2017                                 continue;
2018
2019                         strv_free(r);
2020                         return -EINVAL;
2021                 }
2022                 for (n = 0; n < count; n++) {
2023                         k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p);
2024                         if (k < 0) {
2025                                 if (ignore)
2026                                         continue;
2027
2028                                 strv_free(r);
2029                                 return k;
2030                         }
2031                         /* Log invalid environment variables with filename */
2032                         if (p)
2033                                 p = strv_env_clean_log(p, pglob.gl_pathv[n]);
2034
2035                         if (r == NULL)
2036                                 r = p;
2037                         else {
2038                                 char **m;
2039
2040                                 m = strv_env_merge(2, r, p);
2041                                 strv_free(r);
2042                                 strv_free(p);
2043                                 if (!m)
2044                                         return -ENOMEM;
2045
2046                                 r = m;
2047                         }
2048                 }
2049         }
2050
2051         *l = r;
2052
2053         return 0;
2054 }
2055
2056 static bool tty_may_match_dev_console(const char *tty) {
2057         _cleanup_free_ char *active = NULL;
2058        char *console;
2059
2060         if (startswith(tty, "/dev/"))
2061                 tty += 5;
2062
2063         /* trivial identity? */
2064         if (streq(tty, "console"))
2065                 return true;
2066
2067         console = resolve_dev_console(&active);
2068         /* if we could not resolve, assume it may */
2069         if (!console)
2070                 return true;
2071
2072         /* "tty0" means the active VC, so it may be the same sometimes */
2073         return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
2074 }
2075
2076 bool exec_context_may_touch_console(ExecContext *ec) {
2077         return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
2078                 is_terminal_input(ec->std_input) ||
2079                 is_terminal_output(ec->std_output) ||
2080                 is_terminal_output(ec->std_error)) &&
2081                tty_may_match_dev_console(tty_path(ec));
2082 }
2083
2084 static void strv_fprintf(FILE *f, char **l) {
2085         char **g;
2086
2087         assert(f);
2088
2089         STRV_FOREACH(g, l)
2090                 fprintf(f, " %s", *g);
2091 }
2092
2093 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
2094         char **e;
2095         unsigned i;
2096
2097         assert(c);
2098         assert(f);
2099
2100         prefix = strempty(prefix);
2101
2102         fprintf(f,
2103                 "%sUMask: %04o\n"
2104                 "%sWorkingDirectory: %s\n"
2105                 "%sRootDirectory: %s\n"
2106                 "%sNonBlocking: %s\n"
2107                 "%sPrivateTmp: %s\n"
2108                 "%sPrivateNetwork: %s\n"
2109                 "%sPrivateDevices: %s\n"
2110                 "%sProtectHome: %s\n"
2111                 "%sProtectSystem: %s\n"
2112                 "%sIgnoreSIGPIPE: %s\n",
2113                 prefix, c->umask,
2114                 prefix, c->working_directory ? c->working_directory : "/",
2115                 prefix, c->root_directory ? c->root_directory : "/",
2116                 prefix, yes_no(c->non_blocking),
2117                 prefix, yes_no(c->private_tmp),
2118                 prefix, yes_no(c->private_network),
2119                 prefix, yes_no(c->private_devices),
2120                 prefix, protect_home_to_string(c->protect_home),
2121                 prefix, protect_system_to_string(c->protect_system),
2122                 prefix, yes_no(c->ignore_sigpipe));
2123
2124         STRV_FOREACH(e, c->environment)
2125                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
2126
2127         STRV_FOREACH(e, c->environment_files)
2128                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
2129
2130         if (c->nice_set)
2131                 fprintf(f,
2132                         "%sNice: %i\n",
2133                         prefix, c->nice);
2134
2135         if (c->oom_score_adjust_set)
2136                 fprintf(f,
2137                         "%sOOMScoreAdjust: %i\n",
2138                         prefix, c->oom_score_adjust);
2139
2140         for (i = 0; i < RLIM_NLIMITS; i++)
2141                 if (c->rlimit[i])
2142                         fprintf(f, "%s%s: "RLIM_FMT"\n",
2143                                 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
2144
2145         if (c->ioprio_set) {
2146                 _cleanup_free_ char *class_str = NULL;
2147
2148                 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
2149                 fprintf(f,
2150                         "%sIOSchedulingClass: %s\n"
2151                         "%sIOPriority: %i\n",
2152                         prefix, strna(class_str),
2153                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
2154         }
2155
2156         if (c->cpu_sched_set) {
2157                 _cleanup_free_ char *policy_str = NULL;
2158
2159                 sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
2160                 fprintf(f,
2161                         "%sCPUSchedulingPolicy: %s\n"
2162                         "%sCPUSchedulingPriority: %i\n"
2163                         "%sCPUSchedulingResetOnFork: %s\n",
2164                         prefix, strna(policy_str),
2165                         prefix, c->cpu_sched_priority,
2166                         prefix, yes_no(c->cpu_sched_reset_on_fork));
2167         }
2168
2169         if (c->cpuset) {
2170                 fprintf(f, "%sCPUAffinity:", prefix);
2171                 for (i = 0; i < c->cpuset_ncpus; i++)
2172                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
2173                                 fprintf(f, " %u", i);
2174                 fputs("\n", f);
2175         }
2176
2177         if (c->timer_slack_nsec != NSEC_INFINITY)
2178                 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
2179
2180         fprintf(f,
2181                 "%sStandardInput: %s\n"
2182                 "%sStandardOutput: %s\n"
2183                 "%sStandardError: %s\n",
2184                 prefix, exec_input_to_string(c->std_input),
2185                 prefix, exec_output_to_string(c->std_output),
2186                 prefix, exec_output_to_string(c->std_error));
2187
2188         if (c->tty_path)
2189                 fprintf(f,
2190                         "%sTTYPath: %s\n"
2191                         "%sTTYReset: %s\n"
2192                         "%sTTYVHangup: %s\n"
2193                         "%sTTYVTDisallocate: %s\n",
2194                         prefix, c->tty_path,
2195                         prefix, yes_no(c->tty_reset),
2196                         prefix, yes_no(c->tty_vhangup),
2197                         prefix, yes_no(c->tty_vt_disallocate));
2198
2199         if (c->std_output == EXEC_OUTPUT_SYSLOG ||
2200             c->std_output == EXEC_OUTPUT_KMSG ||
2201             c->std_output == EXEC_OUTPUT_JOURNAL ||
2202             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2203             c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2204             c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
2205             c->std_error == EXEC_OUTPUT_SYSLOG ||
2206             c->std_error == EXEC_OUTPUT_KMSG ||
2207             c->std_error == EXEC_OUTPUT_JOURNAL ||
2208             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2209             c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2210             c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
2211
2212                 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
2213
2214                 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2215                 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2216
2217                 fprintf(f,
2218                         "%sSyslogFacility: %s\n"
2219                         "%sSyslogLevel: %s\n",
2220                         prefix, strna(fac_str),
2221                         prefix, strna(lvl_str));
2222         }
2223
2224         if (c->capabilities) {
2225                 _cleanup_cap_free_charp_ char *t;
2226
2227                 t = cap_to_text(c->capabilities, NULL);
2228                 if (t)
2229                         fprintf(f, "%sCapabilities: %s\n", prefix, t);
2230         }
2231
2232         if (c->secure_bits)
2233                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2234                         prefix,
2235                         (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2236                         (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2237                         (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2238                         (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2239                         (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2240                         (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2241
2242         if (c->capability_bounding_set_drop) {
2243                 unsigned long l;
2244                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2245
2246                 for (l = 0; l <= cap_last_cap(); l++)
2247                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2248                                 _cleanup_cap_free_charp_ char *t;
2249
2250                                 t = cap_to_name(l);
2251                                 if (t)
2252                                         fprintf(f, " %s", t);
2253                         }
2254
2255                 fputs("\n", f);
2256         }
2257
2258         if (c->user)
2259                 fprintf(f, "%sUser: %s\n", prefix, c->user);
2260         if (c->group)
2261                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2262
2263         if (strv_length(c->supplementary_groups) > 0) {
2264                 fprintf(f, "%sSupplementaryGroups:", prefix);
2265                 strv_fprintf(f, c->supplementary_groups);
2266                 fputs("\n", f);
2267         }
2268
2269         if (c->pam_name)
2270                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2271
2272         if (strv_length(c->read_write_dirs) > 0) {
2273                 fprintf(f, "%sReadWriteDirs:", prefix);
2274                 strv_fprintf(f, c->read_write_dirs);
2275                 fputs("\n", f);
2276         }
2277
2278         if (strv_length(c->read_only_dirs) > 0) {
2279                 fprintf(f, "%sReadOnlyDirs:", prefix);
2280                 strv_fprintf(f, c->read_only_dirs);
2281                 fputs("\n", f);
2282         }
2283
2284         if (strv_length(c->inaccessible_dirs) > 0) {
2285                 fprintf(f, "%sInaccessibleDirs:", prefix);
2286                 strv_fprintf(f, c->inaccessible_dirs);
2287                 fputs("\n", f);
2288         }
2289
2290         if (c->utmp_id)
2291                 fprintf(f,
2292                         "%sUtmpIdentifier: %s\n",
2293                         prefix, c->utmp_id);
2294
2295         if (c->selinux_context)
2296                 fprintf(f,
2297                         "%sSELinuxContext: %s%s\n",
2298                         prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
2299
2300         if (c->personality != 0xffffffffUL)
2301                 fprintf(f,
2302                         "%sPersonality: %s\n",
2303                         prefix, strna(personality_to_string(c->personality)));
2304
2305         if (c->syscall_filter) {
2306 #ifdef HAVE_SECCOMP
2307                 Iterator j;
2308                 void *id;
2309                 bool first = true;
2310 #endif
2311
2312                 fprintf(f,
2313                         "%sSystemCallFilter: ",
2314                         prefix);
2315
2316                 if (!c->syscall_whitelist)
2317                         fputc('~', f);
2318
2319 #ifdef HAVE_SECCOMP
2320                 SET_FOREACH(id, c->syscall_filter, j) {
2321                         _cleanup_free_ char *name = NULL;
2322
2323                         if (first)
2324                                 first = false;
2325                         else
2326                                 fputc(' ', f);
2327
2328                         name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
2329                         fputs(strna(name), f);
2330                 }
2331 #endif
2332
2333                 fputc('\n', f);
2334         }
2335
2336         if (c->syscall_archs) {
2337 #ifdef HAVE_SECCOMP
2338                 Iterator j;
2339                 void *id;
2340 #endif
2341
2342                 fprintf(f,
2343                         "%sSystemCallArchitectures:",
2344                         prefix);
2345
2346 #ifdef HAVE_SECCOMP
2347                 SET_FOREACH(id, c->syscall_archs, j)
2348                         fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
2349 #endif
2350                 fputc('\n', f);
2351         }
2352
2353         if (c->syscall_errno != 0)
2354                 fprintf(f,
2355                         "%sSystemCallErrorNumber: %s\n",
2356                         prefix, strna(errno_to_name(c->syscall_errno)));
2357
2358         if (c->apparmor_profile)
2359                 fprintf(f,
2360                         "%sAppArmorProfile: %s%s\n",
2361                         prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
2362 }
2363
2364 void exec_status_start(ExecStatus *s, pid_t pid) {
2365         assert(s);
2366
2367         zero(*s);
2368         s->pid = pid;
2369         dual_timestamp_get(&s->start_timestamp);
2370 }
2371
2372 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2373         assert(s);
2374
2375         if (s->pid && s->pid != pid)
2376                 zero(*s);
2377
2378         s->pid = pid;
2379         dual_timestamp_get(&s->exit_timestamp);
2380
2381         s->code = code;
2382         s->status = status;
2383
2384         if (context) {
2385                 if (context->utmp_id)
2386                         utmp_put_dead_process(context->utmp_id, pid, code, status);
2387
2388                 exec_context_tty_reset(context);
2389         }
2390 }
2391
2392 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2393         char buf[FORMAT_TIMESTAMP_MAX];
2394
2395         assert(s);
2396         assert(f);
2397
2398         if (s->pid <= 0)
2399                 return;
2400
2401         prefix = strempty(prefix);
2402
2403         fprintf(f,
2404                 "%sPID: "PID_FMT"\n",
2405                 prefix, s->pid);
2406
2407         if (s->start_timestamp.realtime > 0)
2408                 fprintf(f,
2409                         "%sStart Timestamp: %s\n",
2410                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2411
2412         if (s->exit_timestamp.realtime > 0)
2413                 fprintf(f,
2414                         "%sExit Timestamp: %s\n"
2415                         "%sExit Code: %s\n"
2416                         "%sExit Status: %i\n",
2417                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2418                         prefix, sigchld_code_to_string(s->code),
2419                         prefix, s->status);
2420 }
2421
2422 char *exec_command_line(char **argv) {
2423         size_t k;
2424         char *n, *p, **a;
2425         bool first = true;
2426
2427         assert(argv);
2428
2429         k = 1;
2430         STRV_FOREACH(a, argv)
2431                 k += strlen(*a)+3;
2432
2433         if (!(n = new(char, k)))
2434                 return NULL;
2435
2436         p = n;
2437         STRV_FOREACH(a, argv) {
2438
2439                 if (!first)
2440                         *(p++) = ' ';
2441                 else
2442                         first = false;
2443
2444                 if (strpbrk(*a, WHITESPACE)) {
2445                         *(p++) = '\'';
2446                         p = stpcpy(p, *a);
2447                         *(p++) = '\'';
2448                 } else
2449                         p = stpcpy(p, *a);
2450
2451         }
2452
2453         *p = 0;
2454
2455         /* FIXME: this doesn't really handle arguments that have
2456          * spaces and ticks in them */
2457
2458         return n;
2459 }
2460
2461 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2462         _cleanup_free_ char *cmd = NULL;
2463         const char *prefix2;
2464
2465         assert(c);
2466         assert(f);
2467
2468         prefix = strempty(prefix);
2469         prefix2 = strappenda(prefix, "\t");
2470
2471         cmd = exec_command_line(c->argv);
2472         fprintf(f,
2473                 "%sCommand Line: %s\n",
2474                 prefix, cmd ? cmd : strerror(ENOMEM));
2475
2476         exec_status_dump(&c->exec_status, f, prefix2);
2477 }
2478
2479 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2480         assert(f);
2481
2482         prefix = strempty(prefix);
2483
2484         LIST_FOREACH(command, c, c)
2485                 exec_command_dump(c, f, prefix);
2486 }
2487
2488 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2489         ExecCommand *end;
2490
2491         assert(l);
2492         assert(e);
2493
2494         if (*l) {
2495                 /* It's kind of important, that we keep the order here */
2496                 LIST_FIND_TAIL(command, *l, end);
2497                 LIST_INSERT_AFTER(command, *l, end, e);
2498         } else
2499               *l = e;
2500 }
2501
2502 int exec_command_set(ExecCommand *c, const char *path, ...) {
2503         va_list ap;
2504         char **l, *p;
2505
2506         assert(c);
2507         assert(path);
2508
2509         va_start(ap, path);
2510         l = strv_new_ap(path, ap);
2511         va_end(ap);
2512
2513         if (!l)
2514                 return -ENOMEM;
2515
2516         p = strdup(path);
2517         if (!p) {
2518                 strv_free(l);
2519                 return -ENOMEM;
2520         }
2521
2522         free(c->path);
2523         c->path = p;
2524
2525         strv_free(c->argv);
2526         c->argv = l;
2527
2528         return 0;
2529 }
2530
2531 static int exec_runtime_allocate(ExecRuntime **rt) {
2532
2533         if (*rt)
2534                 return 0;
2535
2536         *rt = new0(ExecRuntime, 1);
2537         if (!*rt)
2538                 return -ENOMEM;
2539
2540         (*rt)->n_ref = 1;
2541         (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2542
2543         return 0;
2544 }
2545
2546 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2547         int r;
2548
2549         assert(rt);
2550         assert(c);
2551         assert(id);
2552
2553         if (*rt)
2554                 return 1;
2555
2556         if (!c->private_network && !c->private_tmp)
2557                 return 0;
2558
2559         r = exec_runtime_allocate(rt);
2560         if (r < 0)
2561                 return r;
2562
2563         if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2564                 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2565                         return -errno;
2566         }
2567
2568         if (c->private_tmp && !(*rt)->tmp_dir) {
2569                 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2570                 if (r < 0)
2571                         return r;
2572         }
2573
2574         return 1;
2575 }
2576
2577 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2578         assert(r);
2579         assert(r->n_ref > 0);
2580
2581         r->n_ref++;
2582         return r;
2583 }
2584
2585 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2586
2587         if (!r)
2588                 return NULL;
2589
2590         assert(r->n_ref > 0);
2591
2592         r->n_ref--;
2593         if (r->n_ref <= 0) {
2594                 free(r->tmp_dir);
2595                 free(r->var_tmp_dir);
2596                 safe_close_pair(r->netns_storage_socket);
2597                 free(r);
2598         }
2599
2600         return NULL;
2601 }
2602
2603 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2604         assert(u);
2605         assert(f);
2606         assert(fds);
2607
2608         if (!rt)
2609                 return 0;
2610
2611         if (rt->tmp_dir)
2612                 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2613
2614         if (rt->var_tmp_dir)
2615                 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2616
2617         if (rt->netns_storage_socket[0] >= 0) {
2618                 int copy;
2619
2620                 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2621                 if (copy < 0)
2622                         return copy;
2623
2624                 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2625         }
2626
2627         if (rt->netns_storage_socket[1] >= 0) {
2628                 int copy;
2629
2630                 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2631                 if (copy < 0)
2632                         return copy;
2633
2634                 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2635         }
2636
2637         return 0;
2638 }
2639
2640 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2641         int r;
2642
2643         assert(rt);
2644         assert(key);
2645         assert(value);
2646
2647         if (streq(key, "tmp-dir")) {
2648                 char *copy;
2649
2650                 r = exec_runtime_allocate(rt);
2651                 if (r < 0)
2652                         return r;
2653
2654                 copy = strdup(value);
2655                 if (!copy)
2656                         return log_oom();
2657
2658                 free((*rt)->tmp_dir);
2659                 (*rt)->tmp_dir = copy;
2660
2661         } else if (streq(key, "var-tmp-dir")) {
2662                 char *copy;
2663
2664                 r = exec_runtime_allocate(rt);
2665                 if (r < 0)
2666                         return r;
2667
2668                 copy = strdup(value);
2669                 if (!copy)
2670                         return log_oom();
2671
2672                 free((*rt)->var_tmp_dir);
2673                 (*rt)->var_tmp_dir = copy;
2674
2675         } else if (streq(key, "netns-socket-0")) {
2676                 int fd;
2677
2678                 r = exec_runtime_allocate(rt);
2679                 if (r < 0)
2680                         return r;
2681
2682                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2683                         log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2684                 else {
2685                         safe_close((*rt)->netns_storage_socket[0]);
2686                         (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2687                 }
2688         } else if (streq(key, "netns-socket-1")) {
2689                 int fd;
2690
2691                 r = exec_runtime_allocate(rt);
2692                 if (r < 0)
2693                         return r;
2694
2695                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2696                         log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2697                 else {
2698                         safe_close((*rt)->netns_storage_socket[1]);
2699                         (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2700                 }
2701         } else
2702                 return 0;
2703
2704         return 1;
2705 }
2706
2707 static void *remove_tmpdir_thread(void *p) {
2708         _cleanup_free_ char *path = p;
2709
2710         rm_rf_dangerous(path, false, true, false);
2711         return NULL;
2712 }
2713
2714 void exec_runtime_destroy(ExecRuntime *rt) {
2715         int r;
2716
2717         if (!rt)
2718                 return;
2719
2720         /* If there are multiple users of this, let's leave the stuff around */
2721         if (rt->n_ref > 1)
2722                 return;
2723
2724         if (rt->tmp_dir) {
2725                 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2726
2727                 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2728                 if (r < 0) {
2729                         log_warning("Failed to nuke %s: %s", rt->tmp_dir, strerror(-r));
2730                         free(rt->tmp_dir);
2731                 }
2732
2733                 rt->tmp_dir = NULL;
2734         }
2735
2736         if (rt->var_tmp_dir) {
2737                 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2738
2739                 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2740                 if (r < 0) {
2741                         log_warning("Failed to nuke %s: %s", rt->var_tmp_dir, strerror(-r));
2742                         free(rt->var_tmp_dir);
2743                 }
2744
2745                 rt->var_tmp_dir = NULL;
2746         }
2747
2748         safe_close_pair(rt->netns_storage_socket);
2749 }
2750
2751 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2752         [EXEC_INPUT_NULL] = "null",
2753         [EXEC_INPUT_TTY] = "tty",
2754         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2755         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2756         [EXEC_INPUT_SOCKET] = "socket"
2757 };
2758
2759 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2760
2761 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2762         [EXEC_OUTPUT_INHERIT] = "inherit",
2763         [EXEC_OUTPUT_NULL] = "null",
2764         [EXEC_OUTPUT_TTY] = "tty",
2765         [EXEC_OUTPUT_SYSLOG] = "syslog",
2766         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2767         [EXEC_OUTPUT_KMSG] = "kmsg",
2768         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2769         [EXEC_OUTPUT_JOURNAL] = "journal",
2770         [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2771         [EXEC_OUTPUT_SOCKET] = "socket"
2772 };
2773
2774 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);