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
1913 int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
1914         char **i;
1915
1916         assert(c);
1917
1918         if (!runtime_prefix)
1919                 return 0;
1920
1921         STRV_FOREACH(i, c->runtime_directory) {
1922                 _cleanup_free_ char *p;
1923
1924                 p = strjoin(runtime_prefix, "/", *i, NULL);
1925                 if (!p)
1926                         return -ENOMEM;
1927
1928                 /* We execute this synchronously, since we need to be
1929                  * sure this is gone when we start the service
1930                  * next. */
1931                 rm_rf_dangerous(p, false, true, false);
1932         }
1933
1934         return 0;
1935 }
1936
1937 void exec_command_done(ExecCommand *c) {
1938         assert(c);
1939
1940         free(c->path);
1941         c->path = NULL;
1942
1943         strv_free(c->argv);
1944         c->argv = NULL;
1945 }
1946
1947 void exec_command_done_array(ExecCommand *c, unsigned n) {
1948         unsigned i;
1949
1950         for (i = 0; i < n; i++)
1951                 exec_command_done(c+i);
1952 }
1953
1954 void exec_command_free_list(ExecCommand *c) {
1955         ExecCommand *i;
1956
1957         while ((i = c)) {
1958                 LIST_REMOVE(command, c, i);
1959                 exec_command_done(i);
1960                 free(i);
1961         }
1962 }
1963
1964 void exec_command_free_array(ExecCommand **c, unsigned n) {
1965         unsigned i;
1966
1967         for (i = 0; i < n; i++) {
1968                 exec_command_free_list(c[i]);
1969                 c[i] = NULL;
1970         }
1971 }
1972
1973 int exec_context_load_environment(const ExecContext *c, char ***l) {
1974         char **i, **r = NULL;
1975
1976         assert(c);
1977         assert(l);
1978
1979         STRV_FOREACH(i, c->environment_files) {
1980                 char *fn;
1981                 int k;
1982                 bool ignore = false;
1983                 char **p;
1984                 _cleanup_globfree_ glob_t pglob = {};
1985                 int count, n;
1986
1987                 fn = *i;
1988
1989                 if (fn[0] == '-') {
1990                         ignore = true;
1991                         fn ++;
1992                 }
1993
1994                 if (!path_is_absolute(fn)) {
1995                         if (ignore)
1996                                 continue;
1997
1998                         strv_free(r);
1999                         return -EINVAL;
2000                 }
2001
2002                 /* Filename supports globbing, take all matching files */
2003                 errno = 0;
2004                 if (glob(fn, 0, NULL, &pglob) != 0) {
2005                         if (ignore)
2006                                 continue;
2007
2008                         strv_free(r);
2009                         return errno ? -errno : -EINVAL;
2010                 }
2011                 count = pglob.gl_pathc;
2012                 if (count == 0) {
2013                         if (ignore)
2014                                 continue;
2015
2016                         strv_free(r);
2017                         return -EINVAL;
2018                 }
2019                 for (n = 0; n < count; n++) {
2020                         k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p);
2021                         if (k < 0) {
2022                                 if (ignore)
2023                                         continue;
2024
2025                                 strv_free(r);
2026                                 return k;
2027                         }
2028                         /* Log invalid environment variables with filename */
2029                         if (p)
2030                                 p = strv_env_clean_log(p, pglob.gl_pathv[n]);
2031
2032                         if (r == NULL)
2033                                 r = p;
2034                         else {
2035                                 char **m;
2036
2037                                 m = strv_env_merge(2, r, p);
2038                                 strv_free(r);
2039                                 strv_free(p);
2040                                 if (!m)
2041                                         return -ENOMEM;
2042
2043                                 r = m;
2044                         }
2045                 }
2046         }
2047
2048         *l = r;
2049
2050         return 0;
2051 }
2052
2053 static bool tty_may_match_dev_console(const char *tty) {
2054         _cleanup_free_ char *active = NULL;
2055        char *console;
2056
2057         if (startswith(tty, "/dev/"))
2058                 tty += 5;
2059
2060         /* trivial identity? */
2061         if (streq(tty, "console"))
2062                 return true;
2063
2064         console = resolve_dev_console(&active);
2065         /* if we could not resolve, assume it may */
2066         if (!console)
2067                 return true;
2068
2069         /* "tty0" means the active VC, so it may be the same sometimes */
2070         return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
2071 }
2072
2073 bool exec_context_may_touch_console(ExecContext *ec) {
2074         return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
2075                 is_terminal_input(ec->std_input) ||
2076                 is_terminal_output(ec->std_output) ||
2077                 is_terminal_output(ec->std_error)) &&
2078                tty_may_match_dev_console(tty_path(ec));
2079 }
2080
2081 static void strv_fprintf(FILE *f, char **l) {
2082         char **g;
2083
2084         assert(f);
2085
2086         STRV_FOREACH(g, l)
2087                 fprintf(f, " %s", *g);
2088 }
2089
2090 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
2091         char **e;
2092         unsigned i;
2093
2094         assert(c);
2095         assert(f);
2096
2097         prefix = strempty(prefix);
2098
2099         fprintf(f,
2100                 "%sUMask: %04o\n"
2101                 "%sWorkingDirectory: %s\n"
2102                 "%sRootDirectory: %s\n"
2103                 "%sNonBlocking: %s\n"
2104                 "%sPrivateTmp: %s\n"
2105                 "%sPrivateNetwork: %s\n"
2106                 "%sPrivateDevices: %s\n"
2107                 "%sProtectHome: %s\n"
2108                 "%sProtectSystem: %s\n"
2109                 "%sIgnoreSIGPIPE: %s\n",
2110                 prefix, c->umask,
2111                 prefix, c->working_directory ? c->working_directory : "/",
2112                 prefix, c->root_directory ? c->root_directory : "/",
2113                 prefix, yes_no(c->non_blocking),
2114                 prefix, yes_no(c->private_tmp),
2115                 prefix, yes_no(c->private_network),
2116                 prefix, yes_no(c->private_devices),
2117                 prefix, protect_home_to_string(c->protect_home),
2118                 prefix, protect_system_to_string(c->protect_system),
2119                 prefix, yes_no(c->ignore_sigpipe));
2120
2121         STRV_FOREACH(e, c->environment)
2122                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
2123
2124         STRV_FOREACH(e, c->environment_files)
2125                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
2126
2127         if (c->nice_set)
2128                 fprintf(f,
2129                         "%sNice: %i\n",
2130                         prefix, c->nice);
2131
2132         if (c->oom_score_adjust_set)
2133                 fprintf(f,
2134                         "%sOOMScoreAdjust: %i\n",
2135                         prefix, c->oom_score_adjust);
2136
2137         for (i = 0; i < RLIM_NLIMITS; i++)
2138                 if (c->rlimit[i])
2139                         fprintf(f, "%s%s: "RLIM_FMT"\n",
2140                                 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
2141
2142         if (c->ioprio_set) {
2143                 _cleanup_free_ char *class_str = NULL;
2144
2145                 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
2146                 fprintf(f,
2147                         "%sIOSchedulingClass: %s\n"
2148                         "%sIOPriority: %i\n",
2149                         prefix, strna(class_str),
2150                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
2151         }
2152
2153         if (c->cpu_sched_set) {
2154                 _cleanup_free_ char *policy_str = NULL;
2155
2156                 sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
2157                 fprintf(f,
2158                         "%sCPUSchedulingPolicy: %s\n"
2159                         "%sCPUSchedulingPriority: %i\n"
2160                         "%sCPUSchedulingResetOnFork: %s\n",
2161                         prefix, strna(policy_str),
2162                         prefix, c->cpu_sched_priority,
2163                         prefix, yes_no(c->cpu_sched_reset_on_fork));
2164         }
2165
2166         if (c->cpuset) {
2167                 fprintf(f, "%sCPUAffinity:", prefix);
2168                 for (i = 0; i < c->cpuset_ncpus; i++)
2169                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
2170                                 fprintf(f, " %u", i);
2171                 fputs("\n", f);
2172         }
2173
2174         if (c->timer_slack_nsec != NSEC_INFINITY)
2175                 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
2176
2177         fprintf(f,
2178                 "%sStandardInput: %s\n"
2179                 "%sStandardOutput: %s\n"
2180                 "%sStandardError: %s\n",
2181                 prefix, exec_input_to_string(c->std_input),
2182                 prefix, exec_output_to_string(c->std_output),
2183                 prefix, exec_output_to_string(c->std_error));
2184
2185         if (c->tty_path)
2186                 fprintf(f,
2187                         "%sTTYPath: %s\n"
2188                         "%sTTYReset: %s\n"
2189                         "%sTTYVHangup: %s\n"
2190                         "%sTTYVTDisallocate: %s\n",
2191                         prefix, c->tty_path,
2192                         prefix, yes_no(c->tty_reset),
2193                         prefix, yes_no(c->tty_vhangup),
2194                         prefix, yes_no(c->tty_vt_disallocate));
2195
2196         if (c->std_output == EXEC_OUTPUT_SYSLOG ||
2197             c->std_output == EXEC_OUTPUT_KMSG ||
2198             c->std_output == EXEC_OUTPUT_JOURNAL ||
2199             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2200             c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2201             c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
2202             c->std_error == EXEC_OUTPUT_SYSLOG ||
2203             c->std_error == EXEC_OUTPUT_KMSG ||
2204             c->std_error == EXEC_OUTPUT_JOURNAL ||
2205             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2206             c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2207             c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
2208
2209                 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
2210
2211                 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2212                 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2213
2214                 fprintf(f,
2215                         "%sSyslogFacility: %s\n"
2216                         "%sSyslogLevel: %s\n",
2217                         prefix, strna(fac_str),
2218                         prefix, strna(lvl_str));
2219         }
2220
2221         if (c->capabilities) {
2222                 _cleanup_cap_free_charp_ char *t;
2223
2224                 t = cap_to_text(c->capabilities, NULL);
2225                 if (t)
2226                         fprintf(f, "%sCapabilities: %s\n", prefix, t);
2227         }
2228
2229         if (c->secure_bits)
2230                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2231                         prefix,
2232                         (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2233                         (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2234                         (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2235                         (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2236                         (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2237                         (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2238
2239         if (c->capability_bounding_set_drop) {
2240                 unsigned long l;
2241                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2242
2243                 for (l = 0; l <= cap_last_cap(); l++)
2244                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2245                                 _cleanup_cap_free_charp_ char *t;
2246
2247                                 t = cap_to_name(l);
2248                                 if (t)
2249                                         fprintf(f, " %s", t);
2250                         }
2251
2252                 fputs("\n", f);
2253         }
2254
2255         if (c->user)
2256                 fprintf(f, "%sUser: %s\n", prefix, c->user);
2257         if (c->group)
2258                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2259
2260         if (strv_length(c->supplementary_groups) > 0) {
2261                 fprintf(f, "%sSupplementaryGroups:", prefix);
2262                 strv_fprintf(f, c->supplementary_groups);
2263                 fputs("\n", f);
2264         }
2265
2266         if (c->pam_name)
2267                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2268
2269         if (strv_length(c->read_write_dirs) > 0) {
2270                 fprintf(f, "%sReadWriteDirs:", prefix);
2271                 strv_fprintf(f, c->read_write_dirs);
2272                 fputs("\n", f);
2273         }
2274
2275         if (strv_length(c->read_only_dirs) > 0) {
2276                 fprintf(f, "%sReadOnlyDirs:", prefix);
2277                 strv_fprintf(f, c->read_only_dirs);
2278                 fputs("\n", f);
2279         }
2280
2281         if (strv_length(c->inaccessible_dirs) > 0) {
2282                 fprintf(f, "%sInaccessibleDirs:", prefix);
2283                 strv_fprintf(f, c->inaccessible_dirs);
2284                 fputs("\n", f);
2285         }
2286
2287         if (c->utmp_id)
2288                 fprintf(f,
2289                         "%sUtmpIdentifier: %s\n",
2290                         prefix, c->utmp_id);
2291
2292         if (c->selinux_context)
2293                 fprintf(f,
2294                         "%sSELinuxContext: %s%s\n",
2295                         prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
2296
2297         if (c->personality != 0xffffffffUL)
2298                 fprintf(f,
2299                         "%sPersonality: %s\n",
2300                         prefix, strna(personality_to_string(c->personality)));
2301
2302         if (c->syscall_filter) {
2303 #ifdef HAVE_SECCOMP
2304                 Iterator j;
2305                 void *id;
2306                 bool first = true;
2307 #endif
2308
2309                 fprintf(f,
2310                         "%sSystemCallFilter: ",
2311                         prefix);
2312
2313                 if (!c->syscall_whitelist)
2314                         fputc('~', f);
2315
2316 #ifdef HAVE_SECCOMP
2317                 SET_FOREACH(id, c->syscall_filter, j) {
2318                         _cleanup_free_ char *name = NULL;
2319
2320                         if (first)
2321                                 first = false;
2322                         else
2323                                 fputc(' ', f);
2324
2325                         name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
2326                         fputs(strna(name), f);
2327                 }
2328 #endif
2329
2330                 fputc('\n', f);
2331         }
2332
2333         if (c->syscall_archs) {
2334 #ifdef HAVE_SECCOMP
2335                 Iterator j;
2336                 void *id;
2337 #endif
2338
2339                 fprintf(f,
2340                         "%sSystemCallArchitectures:",
2341                         prefix);
2342
2343 #ifdef HAVE_SECCOMP
2344                 SET_FOREACH(id, c->syscall_archs, j)
2345                         fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
2346 #endif
2347                 fputc('\n', f);
2348         }
2349
2350         if (c->syscall_errno != 0)
2351                 fprintf(f,
2352                         "%sSystemCallErrorNumber: %s\n",
2353                         prefix, strna(errno_to_name(c->syscall_errno)));
2354
2355         if (c->apparmor_profile)
2356                 fprintf(f,
2357                         "%sAppArmorProfile: %s%s\n",
2358                         prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
2359 }
2360
2361 void exec_status_start(ExecStatus *s, pid_t pid) {
2362         assert(s);
2363
2364         zero(*s);
2365         s->pid = pid;
2366         dual_timestamp_get(&s->start_timestamp);
2367 }
2368
2369 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2370         assert(s);
2371
2372         if (s->pid && s->pid != pid)
2373                 zero(*s);
2374
2375         s->pid = pid;
2376         dual_timestamp_get(&s->exit_timestamp);
2377
2378         s->code = code;
2379         s->status = status;
2380
2381         if (context) {
2382                 if (context->utmp_id)
2383                         utmp_put_dead_process(context->utmp_id, pid, code, status);
2384
2385                 exec_context_tty_reset(context);
2386         }
2387 }
2388
2389 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2390         char buf[FORMAT_TIMESTAMP_MAX];
2391
2392         assert(s);
2393         assert(f);
2394
2395         if (s->pid <= 0)
2396                 return;
2397
2398         prefix = strempty(prefix);
2399
2400         fprintf(f,
2401                 "%sPID: "PID_FMT"\n",
2402                 prefix, s->pid);
2403
2404         if (s->start_timestamp.realtime > 0)
2405                 fprintf(f,
2406                         "%sStart Timestamp: %s\n",
2407                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2408
2409         if (s->exit_timestamp.realtime > 0)
2410                 fprintf(f,
2411                         "%sExit Timestamp: %s\n"
2412                         "%sExit Code: %s\n"
2413                         "%sExit Status: %i\n",
2414                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2415                         prefix, sigchld_code_to_string(s->code),
2416                         prefix, s->status);
2417 }
2418
2419 char *exec_command_line(char **argv) {
2420         size_t k;
2421         char *n, *p, **a;
2422         bool first = true;
2423
2424         assert(argv);
2425
2426         k = 1;
2427         STRV_FOREACH(a, argv)
2428                 k += strlen(*a)+3;
2429
2430         if (!(n = new(char, k)))
2431                 return NULL;
2432
2433         p = n;
2434         STRV_FOREACH(a, argv) {
2435
2436                 if (!first)
2437                         *(p++) = ' ';
2438                 else
2439                         first = false;
2440
2441                 if (strpbrk(*a, WHITESPACE)) {
2442                         *(p++) = '\'';
2443                         p = stpcpy(p, *a);
2444                         *(p++) = '\'';
2445                 } else
2446                         p = stpcpy(p, *a);
2447
2448         }
2449
2450         *p = 0;
2451
2452         /* FIXME: this doesn't really handle arguments that have
2453          * spaces and ticks in them */
2454
2455         return n;
2456 }
2457
2458 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2459         _cleanup_free_ char *cmd = NULL;
2460         const char *prefix2;
2461
2462         assert(c);
2463         assert(f);
2464
2465         prefix = strempty(prefix);
2466         prefix2 = strappenda(prefix, "\t");
2467
2468         cmd = exec_command_line(c->argv);
2469         fprintf(f,
2470                 "%sCommand Line: %s\n",
2471                 prefix, cmd ? cmd : strerror(ENOMEM));
2472
2473         exec_status_dump(&c->exec_status, f, prefix2);
2474 }
2475
2476 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2477         assert(f);
2478
2479         prefix = strempty(prefix);
2480
2481         LIST_FOREACH(command, c, c)
2482                 exec_command_dump(c, f, prefix);
2483 }
2484
2485 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2486         ExecCommand *end;
2487
2488         assert(l);
2489         assert(e);
2490
2491         if (*l) {
2492                 /* It's kind of important, that we keep the order here */
2493                 LIST_FIND_TAIL(command, *l, end);
2494                 LIST_INSERT_AFTER(command, *l, end, e);
2495         } else
2496               *l = e;
2497 }
2498
2499 int exec_command_set(ExecCommand *c, const char *path, ...) {
2500         va_list ap;
2501         char **l, *p;
2502
2503         assert(c);
2504         assert(path);
2505
2506         va_start(ap, path);
2507         l = strv_new_ap(path, ap);
2508         va_end(ap);
2509
2510         if (!l)
2511                 return -ENOMEM;
2512
2513         p = strdup(path);
2514         if (!p) {
2515                 strv_free(l);
2516                 return -ENOMEM;
2517         }
2518
2519         free(c->path);
2520         c->path = p;
2521
2522         strv_free(c->argv);
2523         c->argv = l;
2524
2525         return 0;
2526 }
2527
2528 static int exec_runtime_allocate(ExecRuntime **rt) {
2529
2530         if (*rt)
2531                 return 0;
2532
2533         *rt = new0(ExecRuntime, 1);
2534         if (!*rt)
2535                 return -ENOMEM;
2536
2537         (*rt)->n_ref = 1;
2538         (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2539
2540         return 0;
2541 }
2542
2543 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2544         int r;
2545
2546         assert(rt);
2547         assert(c);
2548         assert(id);
2549
2550         if (*rt)
2551                 return 1;
2552
2553         if (!c->private_network && !c->private_tmp)
2554                 return 0;
2555
2556         r = exec_runtime_allocate(rt);
2557         if (r < 0)
2558                 return r;
2559
2560         if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2561                 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2562                         return -errno;
2563         }
2564
2565         if (c->private_tmp && !(*rt)->tmp_dir) {
2566                 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2567                 if (r < 0)
2568                         return r;
2569         }
2570
2571         return 1;
2572 }
2573
2574 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2575         assert(r);
2576         assert(r->n_ref > 0);
2577
2578         r->n_ref++;
2579         return r;
2580 }
2581
2582 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2583
2584         if (!r)
2585                 return NULL;
2586
2587         assert(r->n_ref > 0);
2588
2589         r->n_ref--;
2590         if (r->n_ref <= 0) {
2591                 free(r->tmp_dir);
2592                 free(r->var_tmp_dir);
2593                 safe_close_pair(r->netns_storage_socket);
2594                 free(r);
2595         }
2596
2597         return NULL;
2598 }
2599
2600 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2601         assert(u);
2602         assert(f);
2603         assert(fds);
2604
2605         if (!rt)
2606                 return 0;
2607
2608         if (rt->tmp_dir)
2609                 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2610
2611         if (rt->var_tmp_dir)
2612                 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2613
2614         if (rt->netns_storage_socket[0] >= 0) {
2615                 int copy;
2616
2617                 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2618                 if (copy < 0)
2619                         return copy;
2620
2621                 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2622         }
2623
2624         if (rt->netns_storage_socket[1] >= 0) {
2625                 int copy;
2626
2627                 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2628                 if (copy < 0)
2629                         return copy;
2630
2631                 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2632         }
2633
2634         return 0;
2635 }
2636
2637 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2638         int r;
2639
2640         assert(rt);
2641         assert(key);
2642         assert(value);
2643
2644         if (streq(key, "tmp-dir")) {
2645                 char *copy;
2646
2647                 r = exec_runtime_allocate(rt);
2648                 if (r < 0)
2649                         return r;
2650
2651                 copy = strdup(value);
2652                 if (!copy)
2653                         return log_oom();
2654
2655                 free((*rt)->tmp_dir);
2656                 (*rt)->tmp_dir = copy;
2657
2658         } else if (streq(key, "var-tmp-dir")) {
2659                 char *copy;
2660
2661                 r = exec_runtime_allocate(rt);
2662                 if (r < 0)
2663                         return r;
2664
2665                 copy = strdup(value);
2666                 if (!copy)
2667                         return log_oom();
2668
2669                 free((*rt)->var_tmp_dir);
2670                 (*rt)->var_tmp_dir = copy;
2671
2672         } else if (streq(key, "netns-socket-0")) {
2673                 int fd;
2674
2675                 r = exec_runtime_allocate(rt);
2676                 if (r < 0)
2677                         return r;
2678
2679                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2680                         log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2681                 else {
2682                         safe_close((*rt)->netns_storage_socket[0]);
2683                         (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2684                 }
2685         } else if (streq(key, "netns-socket-1")) {
2686                 int fd;
2687
2688                 r = exec_runtime_allocate(rt);
2689                 if (r < 0)
2690                         return r;
2691
2692                 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2693                         log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2694                 else {
2695                         safe_close((*rt)->netns_storage_socket[1]);
2696                         (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2697                 }
2698         } else
2699                 return 0;
2700
2701         return 1;
2702 }
2703
2704 static void *remove_tmpdir_thread(void *p) {
2705         _cleanup_free_ char *path = p;
2706
2707         rm_rf_dangerous(path, false, true, false);
2708         return NULL;
2709 }
2710
2711 void exec_runtime_destroy(ExecRuntime *rt) {
2712         int r;
2713
2714         if (!rt)
2715                 return;
2716
2717         /* If there are multiple users of this, let's leave the stuff around */
2718         if (rt->n_ref > 1)
2719                 return;
2720
2721         if (rt->tmp_dir) {
2722                 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2723
2724                 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2725                 if (r < 0) {
2726                         log_warning("Failed to nuke %s: %s", rt->tmp_dir, strerror(-r));
2727                         free(rt->tmp_dir);
2728                 }
2729
2730                 rt->tmp_dir = NULL;
2731         }
2732
2733         if (rt->var_tmp_dir) {
2734                 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2735
2736                 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2737                 if (r < 0) {
2738                         log_warning("Failed to nuke %s: %s", rt->var_tmp_dir, strerror(-r));
2739                         free(rt->var_tmp_dir);
2740                 }
2741
2742                 rt->var_tmp_dir = NULL;
2743         }
2744
2745         safe_close_pair(rt->netns_storage_socket);
2746 }
2747
2748 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2749         [EXEC_INPUT_NULL] = "null",
2750         [EXEC_INPUT_TTY] = "tty",
2751         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2752         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2753         [EXEC_INPUT_SOCKET] = "socket"
2754 };
2755
2756 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2757
2758 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2759         [EXEC_OUTPUT_INHERIT] = "inherit",
2760         [EXEC_OUTPUT_NULL] = "null",
2761         [EXEC_OUTPUT_TTY] = "tty",
2762         [EXEC_OUTPUT_SYSLOG] = "syslog",
2763         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2764         [EXEC_OUTPUT_KMSG] = "kmsg",
2765         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2766         [EXEC_OUTPUT_JOURNAL] = "journal",
2767         [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2768         [EXEC_OUTPUT_SOCKET] = "socket"
2769 };
2770
2771 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);