chiark / gitweb /
execute: make flags_fds() parameters const
[elogind.git] / execute.c
1 /*-*- Mode: C; c-basic-offset: 8 -*-*/
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 General Public License as published by
10   the Free Software Foundation; either version 2 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   General Public License for more details.
17
18   You should have received a copy of the GNU 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
38 #include "execute.h"
39 #include "strv.h"
40 #include "macro.h"
41 #include "util.h"
42 #include "log.h"
43 #include "ioprio.h"
44 #include "securebits.h"
45 #include "cgroup.h"
46
47 /* This assumes there is a 'tty' group */
48 #define TTY_MODE 0620
49
50 static int shift_fds(int fds[], unsigned n_fds) {
51         int start, restart_from;
52
53         if (n_fds <= 0)
54                 return 0;
55
56         /* Modifies the fds array! (sorts it) */
57
58         assert(fds);
59
60         start = 0;
61         for (;;) {
62                 int i;
63
64                 restart_from = -1;
65
66                 for (i = start; i < (int) n_fds; i++) {
67                         int nfd;
68
69                         /* Already at right index? */
70                         if (fds[i] == i+3)
71                                 continue;
72
73                         if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
74                                 return -errno;
75
76                         close_nointr_nofail(fds[i]);
77                         fds[i] = nfd;
78
79                         /* Hmm, the fd we wanted isn't free? Then
80                          * let's remember that and try again from here*/
81                         if (nfd != i+3 && restart_from < 0)
82                                 restart_from = i;
83                 }
84
85                 if (restart_from < 0)
86                         break;
87
88                 start = restart_from;
89         }
90
91         return 0;
92 }
93
94 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
95         unsigned i;
96         int r;
97
98         if (n_fds <= 0)
99                 return 0;
100
101         assert(fds);
102
103         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
104
105         for (i = 0; i < n_fds; i++) {
106
107                 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
108                         return r;
109
110                 /* We unconditionally drop FD_CLOEXEC from the fds,
111                  * since after all we want to pass these fds to our
112                  * children */
113
114                 if ((r = fd_cloexec(fds[i], false)) < 0)
115                         return r;
116         }
117
118         return 0;
119 }
120
121 static const char *tty_path(const ExecContext *context) {
122         assert(context);
123
124         if (context->tty_path)
125                 return context->tty_path;
126
127         return "/dev/console";
128 }
129
130 static int open_null_as(int flags, int nfd) {
131         int fd, r;
132
133         assert(nfd >= 0);
134
135         if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
136                 return -errno;
137
138         if (fd != nfd) {
139                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
140                 close_nointr_nofail(fd);
141         } else
142                 r = nfd;
143
144         return r;
145 }
146
147 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
148         int fd, r;
149         union {
150                 struct sockaddr sa;
151                 struct sockaddr_un un;
152         } sa;
153
154         assert(context);
155         assert(output < _EXEC_OUTPUT_MAX);
156         assert(ident);
157         assert(nfd >= 0);
158
159         if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
160                 return -errno;
161
162         zero(sa);
163         sa.sa.sa_family = AF_UNIX;
164         strncpy(sa.un.sun_path+1, LOGGER_SOCKET, sizeof(sa.un.sun_path)-1);
165
166         if (connect(fd, &sa.sa, sizeof(sa)) < 0) {
167                 close_nointr_nofail(fd);
168                 return -errno;
169         }
170
171         if (shutdown(fd, SHUT_RD) < 0) {
172                 close_nointr_nofail(fd);
173                 return -errno;
174         }
175
176         /* We speak a very simple protocol between log server
177          * and client: one line for the log destination (kmsg
178          * or syslog), followed by the priority field,
179          * followed by the process name. Since we replaced
180          * stdin/stderr we simple use stdio to write to
181          * it. Note that we use stderr, to minimize buffer
182          * flushing issues. */
183
184         dprintf(fd,
185                 "%s\n"
186                 "%i\n"
187                 "%s\n",
188                 output == EXEC_OUTPUT_KERNEL ? "kmsg" : "syslog",
189                 context->syslog_priority,
190                 context->syslog_identifier ? context->syslog_identifier : ident);
191
192         if (fd != nfd) {
193                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
194                 close_nointr_nofail(fd);
195         } else
196                 r = nfd;
197
198         return r;
199 }
200 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
201         int fd, r;
202
203         assert(path);
204         assert(nfd >= 0);
205
206         if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
207                 return fd;
208
209         if (fd != nfd) {
210                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
211                 close_nointr_nofail(fd);
212         } else
213                 r = nfd;
214
215         return r;
216 }
217
218 static bool is_terminal_input(ExecInput i) {
219         return
220                 i == EXEC_INPUT_TTY ||
221                 i == EXEC_INPUT_TTY_FORCE ||
222                 i == EXEC_INPUT_TTY_FAIL;
223 }
224
225 static int setup_input(const ExecContext *context) {
226         assert(context);
227
228         switch (context->std_input) {
229
230         case EXEC_INPUT_NULL:
231                 return open_null_as(O_RDONLY, STDIN_FILENO);
232
233         case EXEC_INPUT_TTY:
234         case EXEC_INPUT_TTY_FORCE:
235         case EXEC_INPUT_TTY_FAIL: {
236                 int fd, r;
237
238                 if ((fd = acquire_terminal(
239                                      tty_path(context),
240                                      context->std_input == EXEC_INPUT_TTY_FAIL,
241                                      context->std_input == EXEC_INPUT_TTY_FORCE)) < 0)
242                         return fd;
243
244                 if (fd != STDIN_FILENO) {
245                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
246                         close_nointr_nofail(fd);
247                 } else
248                         r = STDIN_FILENO;
249
250                 return r;
251         }
252
253         default:
254                 assert_not_reached("Unknown input type");
255         }
256 }
257
258 static int setup_output(const ExecContext *context, const char *ident) {
259         assert(context);
260         assert(ident);
261
262         /* This expects the input is already set up */
263
264         switch (context->std_output) {
265
266         case EXEC_OUTPUT_INHERIT:
267
268                 /* If the input is connected to a terminal, inherit that... */
269                 if (is_terminal_input(context->std_input))
270                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
271
272                 return 0;
273
274         case EXEC_OUTPUT_NULL:
275                 return open_null_as(O_WRONLY, STDOUT_FILENO);
276
277         case EXEC_OUTPUT_TTY: {
278                 if (is_terminal_input(context->std_input))
279                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
280
281                 /* We don't reset the terminal if this is just about output */
282                 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
283         }
284
285         case EXEC_OUTPUT_SYSLOG:
286         case EXEC_OUTPUT_KERNEL:
287                 return connect_logger_as(context, context->std_output, ident, STDOUT_FILENO);
288
289         default:
290                 assert_not_reached("Unknown output type");
291         }
292 }
293
294 static int setup_error(const ExecContext *context, const char *ident) {
295         assert(context);
296         assert(ident);
297
298         /* This expects the input and output are already set up */
299
300         /* Don't change the stderr file descriptor if we inherit all
301          * the way and are not on a tty */
302         if (context->std_error == EXEC_OUTPUT_INHERIT &&
303             context->std_output == EXEC_OUTPUT_INHERIT &&
304             !is_terminal_input(context->std_input))
305                 return STDERR_FILENO;
306
307         /* Duplicate form stdout if possible */
308         if (context->std_error == context->std_output ||
309             context->std_error == EXEC_OUTPUT_INHERIT)
310                 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
311
312         switch (context->std_error) {
313
314         case EXEC_OUTPUT_NULL:
315                 return open_null_as(O_WRONLY, STDERR_FILENO);
316
317         case EXEC_OUTPUT_TTY:
318                 if (is_terminal_input(context->std_input))
319                         return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
320
321                 /* We don't reset the terminal if this is just about output */
322                 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
323
324         case EXEC_OUTPUT_SYSLOG:
325         case EXEC_OUTPUT_KERNEL:
326                 return connect_logger_as(context, context->std_error, ident, STDERR_FILENO);
327
328         default:
329                 assert_not_reached("Unknown error type");
330         }
331 }
332
333 static int chown_terminal(int fd, uid_t uid) {
334         struct stat st;
335
336         assert(fd >= 0);
337
338         /* This might fail. What matters are the results. */
339         fchown(fd, uid, -1);
340         fchmod(fd, TTY_MODE);
341
342         if (fstat(fd, &st) < 0)
343                 return -errno;
344
345         if (st.st_uid != uid ||
346             st.st_mode != TTY_MODE)
347                 return -EPERM;
348
349         return 0;
350 }
351
352 static int setup_confirm_stdio(const ExecContext *context,
353                                int *_saved_stdin,
354                                int *_saved_stdout) {
355         int fd = -1, saved_stdin, saved_stdout = -1, r;
356
357         assert(context);
358         assert(_saved_stdin);
359         assert(_saved_stdout);
360
361         /* This returns positive EXIT_xxx return values instead of
362          * negative errno style values! */
363
364         if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
365                 return EXIT_STDIN;
366
367         if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
368                 r = EXIT_STDOUT;
369                 goto fail;
370         }
371
372         if ((fd = acquire_terminal(
373                              tty_path(context),
374                              context->std_input == EXEC_INPUT_TTY_FAIL,
375                              context->std_input == EXEC_INPUT_TTY_FORCE)) < 0) {
376                 r = EXIT_STDIN;
377                 goto fail;
378         }
379
380         if (chown_terminal(fd, getuid()) < 0) {
381                 r = EXIT_STDIN;
382                 goto fail;
383         }
384
385         if (dup2(fd, STDIN_FILENO) < 0) {
386                 r = EXIT_STDIN;
387                 goto fail;
388         }
389
390         if (dup2(fd, STDOUT_FILENO) < 0) {
391                 r = EXIT_STDOUT;
392                 goto fail;
393         }
394
395         if (fd >= 2)
396                 close_nointr_nofail(fd);
397
398         *_saved_stdin = saved_stdin;
399         *_saved_stdout = saved_stdout;
400
401         return 0;
402
403 fail:
404         if (saved_stdout >= 0)
405                 close_nointr_nofail(saved_stdout);
406
407         if (saved_stdin >= 0)
408                 close_nointr_nofail(saved_stdin);
409
410         if (fd >= 0)
411                 close_nointr_nofail(fd);
412
413         return r;
414 }
415
416 static int restore_conform_stdio(const ExecContext *context,
417                                  int *saved_stdin,
418                                  int *saved_stdout,
419                                  bool *keep_stdin,
420                                  bool *keep_stdout) {
421
422         assert(context);
423         assert(saved_stdin);
424         assert(*saved_stdin >= 0);
425         assert(saved_stdout);
426         assert(*saved_stdout >= 0);
427
428         /* This returns positive EXIT_xxx return values instead of
429          * negative errno style values! */
430
431         if (is_terminal_input(context->std_input)) {
432
433                 /* The service wants terminal input. */
434
435                 *keep_stdin = true;
436                 *keep_stdout =
437                         context->std_output == EXEC_OUTPUT_INHERIT ||
438                         context->std_output == EXEC_OUTPUT_TTY;
439
440         } else {
441                 /* If the service doesn't want a controlling terminal,
442                  * then we need to get rid entirely of what we have
443                  * already. */
444
445                 if (release_terminal() < 0)
446                         return EXIT_STDIN;
447
448                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
449                         return EXIT_STDIN;
450
451                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
452                         return EXIT_STDOUT;
453
454                 *keep_stdout = *keep_stdin = false;
455         }
456
457         return 0;
458 }
459
460 static int get_group_creds(const char *groupname, gid_t *gid) {
461         struct group *g;
462         unsigned long lu;
463
464         assert(groupname);
465         assert(gid);
466
467         /* We enforce some special rules for gid=0: in order to avoid
468          * NSS lookups for root we hardcode its data. */
469
470         if (streq(groupname, "root") || streq(groupname, "0")) {
471                 *gid = 0;
472                 return 0;
473         }
474
475         if (safe_atolu(groupname, &lu) >= 0) {
476                 errno = 0;
477                 g = getgrgid((gid_t) lu);
478         } else {
479                 errno = 0;
480                 g = getgrnam(groupname);
481         }
482
483         if (!g)
484                 return errno != 0 ? -errno : -ESRCH;
485
486         *gid = g->gr_gid;
487         return 0;
488 }
489
490 static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
491         struct passwd *p;
492         unsigned long lu;
493
494         assert(username);
495         assert(*username);
496         assert(uid);
497         assert(gid);
498         assert(home);
499
500         /* We enforce some special rules for uid=0: in order to avoid
501          * NSS lookups for root we hardcode its data. */
502
503         if (streq(*username, "root") || streq(*username, "0")) {
504                 *username = "root";
505                 *uid = 0;
506                 *gid = 0;
507                 *home = "/root";
508                 return 0;
509         }
510
511         if (safe_atolu(*username, &lu) >= 0) {
512                 errno = 0;
513                 p = getpwuid((uid_t) lu);
514
515                 /* If there are multiple users with the same id, make
516                  * sure to leave $USER to the configured value instead
517                  * of the first occurence in the database. However if
518                  * the uid was configured by a numeric uid, then let's
519                  * pick the real username from /etc/passwd. */
520                 if (*username && p)
521                         *username = p->pw_name;
522         } else {
523                 errno = 0;
524                 p = getpwnam(*username);
525         }
526
527         if (!p)
528                 return errno != 0 ? -errno : -ESRCH;
529
530         *uid = p->pw_uid;
531         *gid = p->pw_gid;
532         *home = p->pw_dir;
533         return 0;
534 }
535
536 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
537         bool keep_groups = false;
538         int r;
539
540         assert(context);
541
542         /* Lookup and ser GID and supplementary group list. Here too
543          * we avoid NSS lookups for gid=0. */
544
545         if (context->group || username) {
546
547                 if (context->group)
548                         if ((r = get_group_creds(context->group, &gid)) < 0)
549                                 return r;
550
551                 /* First step, initialize groups from /etc/groups */
552                 if (username && gid != 0) {
553                         if (initgroups(username, gid) < 0)
554                                 return -errno;
555
556                         keep_groups = true;
557                 }
558
559                 /* Second step, set our gids */
560                 if (setresgid(gid, gid, gid) < 0)
561                         return -errno;
562         }
563
564         if (context->supplementary_groups) {
565                 int ngroups_max, k;
566                 gid_t *gids;
567                 char **i;
568
569                 /* Final step, initialize any manually set supplementary groups */
570                 ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
571
572                 if (!(gids = new(gid_t, ngroups_max)))
573                         return -ENOMEM;
574
575                 if (keep_groups) {
576                         if ((k = getgroups(ngroups_max, gids)) < 0) {
577                                 free(gids);
578                                 return -errno;
579                         }
580                 } else
581                         k = 0;
582
583                 STRV_FOREACH(i, context->supplementary_groups) {
584
585                         if (k >= ngroups_max) {
586                                 free(gids);
587                                 return -E2BIG;
588                         }
589
590                         if ((r = get_group_creds(*i, gids+k)) < 0) {
591                                 free(gids);
592                                 return r;
593                         }
594
595                         k++;
596                 }
597
598                 if (setgroups(k, gids) < 0) {
599                         free(gids);
600                         return -errno;
601                 }
602
603                 free(gids);
604         }
605
606         return 0;
607 }
608
609 static int enforce_user(const ExecContext *context, uid_t uid) {
610         int r;
611         assert(context);
612
613         /* Sets (but doesn't lookup) the uid and make sure we keep the
614          * capabilities while doing so. */
615
616         if (context->capabilities) {
617                 cap_t d;
618                 static const cap_value_t bits[] = {
619                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
620                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
621                 };
622
623                 /* First step: If we need to keep capabilities but
624                  * drop privileges we need to make sure we keep our
625                  * caps, whiel we drop priviliges. */
626                 if (uid != 0) {
627                         int sb = context->secure_bits|SECURE_KEEP_CAPS;
628
629                         if (prctl(PR_GET_SECUREBITS) != sb)
630                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
631                                         return -errno;
632                 }
633
634                 /* Second step: set the capabilites. This will reduce
635                  * the capabilities to the minimum we need. */
636
637                 if (!(d = cap_dup(context->capabilities)))
638                         return -errno;
639
640                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
641                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
642                         r = -errno;
643                         cap_free(d);
644                         return r;
645                 }
646
647                 if (cap_set_proc(d) < 0) {
648                         r = -errno;
649                         cap_free(d);
650                         return r;
651                 }
652
653                 cap_free(d);
654         }
655
656         /* Third step: actually set the uids */
657         if (setresuid(uid, uid, uid) < 0)
658                 return -errno;
659
660         /* At this point we should have all necessary capabilities but
661            are otherwise a normal user. However, the caps might got
662            corrupted due to the setresuid() so we need clean them up
663            later. This is done outside of this call. */
664
665         return 0;
666 }
667
668 int exec_spawn(ExecCommand *command,
669                const ExecContext *context,
670                int fds[], unsigned n_fds,
671                bool apply_permissions,
672                bool apply_chroot,
673                bool confirm_spawn,
674                CGroupBonding *cgroup_bondings,
675                pid_t *ret) {
676
677         pid_t pid;
678         int r;
679         char *line;
680
681         assert(command);
682         assert(context);
683         assert(ret);
684         assert(fds || n_fds <= 0);
685
686         if (!(line = exec_command_line(command)))
687                 return -ENOMEM;
688
689         log_debug("About to execute: %s", line);
690         free(line);
691
692         if (cgroup_bondings)
693                 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
694                         return r;
695
696         if ((pid = fork()) < 0)
697                 return -errno;
698
699         if (pid == 0) {
700                 int i;
701                 sigset_t ss;
702                 const char *username = NULL, *home = NULL;
703                 uid_t uid = (uid_t) -1;
704                 gid_t gid = (gid_t) -1;
705                 char **our_env = NULL, **final_env = NULL;
706                 unsigned n_env = 0;
707                 int saved_stdout = -1, saved_stdin = -1;
708                 bool keep_stdout = false, keep_stdin = false;
709
710                 /* child */
711
712                 reset_all_signal_handlers();
713
714                 if (sigemptyset(&ss) < 0 ||
715                     sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
716                         r = EXIT_SIGNAL_MASK;
717                         goto fail;
718                 }
719
720                 if (setsid() < 0) {
721                         r = EXIT_SETSID;
722                         goto fail;
723                 }
724
725                 umask(context->umask);
726
727                 if (confirm_spawn) {
728                         char response;
729
730                         /* Set up terminal for the question */
731                         if ((r = setup_confirm_stdio(context,
732                                                      &saved_stdin, &saved_stdout)))
733                                 goto fail;
734
735                         /* Now ask the question. */
736                         if (!(line = exec_command_line(command))) {
737                                 r = EXIT_MEMORY;
738                                 goto fail;
739                         }
740
741                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
742                         free(line);
743
744                         if (r < 0 || response == 'n') {
745                                 r = EXIT_CONFIRM;
746                                 goto fail;
747                         } else if (response == 's') {
748                                 r = 0;
749                                 goto fail;
750                         }
751
752                         /* Release terminal for the question */
753                         if ((r = restore_conform_stdio(context,
754                                                        &saved_stdin, &saved_stdout,
755                                                        &keep_stdin, &keep_stdout)))
756                                 goto fail;
757                 }
758
759                 if (!keep_stdin)
760                         if (setup_input(context) < 0) {
761                                 r = EXIT_STDIN;
762                                 goto fail;
763                         }
764
765                 if (!keep_stdout)
766                         if (setup_output(context, file_name_from_path(command->path)) < 0) {
767                                 r = EXIT_STDOUT;
768                                 goto fail;
769                         }
770
771                 if (setup_error(context, file_name_from_path(command->path)) < 0) {
772                         r = EXIT_STDERR;
773                         goto fail;
774                 }
775
776                 if (cgroup_bondings)
777                         if ((r = cgroup_bonding_install_list(cgroup_bondings, 0)) < 0) {
778                                 r = EXIT_CGROUP;
779                                 goto fail;
780                         }
781
782                 if (context->oom_adjust_set) {
783                         char t[16];
784
785                         snprintf(t, sizeof(t), "%i", context->oom_adjust);
786                         char_array_0(t);
787
788                         if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
789                                 r = EXIT_OOM_ADJUST;
790                                 goto fail;
791                         }
792                 }
793
794                 if (context->nice_set)
795                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
796                                 r = EXIT_NICE;
797                                 goto fail;
798                         }
799
800                 if (context->cpu_sched_set) {
801                         struct sched_param param;
802
803                         zero(param);
804                         param.sched_priority = context->cpu_sched_priority;
805
806                         if (sched_setscheduler(0, context->cpu_sched_policy |
807                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
808                                 r = EXIT_SETSCHEDULER;
809                                 goto fail;
810                         }
811                 }
812
813                 if (context->cpu_affinity_set)
814                         if (sched_setaffinity(0, sizeof(context->cpu_affinity), &context->cpu_affinity) < 0) {
815                                 r = EXIT_CPUAFFINITY;
816                                 goto fail;
817                         }
818
819                 if (context->ioprio_set)
820                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
821                                 r = EXIT_IOPRIO;
822                                 goto fail;
823                         }
824
825                 if (context->timer_slack_ns_set)
826                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_ns_set) < 0) {
827                                 r = EXIT_TIMERSLACK;
828                                 goto fail;
829                         }
830
831                 if (context->user) {
832                         username = context->user;
833                         if (get_user_creds(&username, &uid, &gid, &home) < 0) {
834                                 r = EXIT_USER;
835                                 goto fail;
836                         }
837                 }
838
839                 if (apply_permissions)
840                         if (enforce_groups(context, username, uid) < 0) {
841                                 r = EXIT_GROUP;
842                                 goto fail;
843                         }
844
845                 if (is_terminal_input(context->std_input))
846                         if (chown_terminal(STDIN_FILENO, uid) < 0) {
847                                 r = EXIT_STDIN;
848                                 goto fail;
849                         }
850
851                 if (apply_chroot) {
852                         if (context->root_directory)
853                                 if (chroot(context->root_directory) < 0) {
854                                         r = EXIT_CHROOT;
855                                         goto fail;
856                                 }
857
858                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
859                                 r = EXIT_CHDIR;
860                                 goto fail;
861                         }
862                 } else {
863
864                         char *d;
865
866                         if (asprintf(&d, "%s/%s",
867                                      context->root_directory ? context->root_directory : "",
868                                      context->working_directory ? context->working_directory : "") < 0) {
869                                 r = EXIT_MEMORY;
870                                 goto fail;
871                         }
872
873                         if (chdir(d) < 0) {
874                                 free(d);
875                                 r = EXIT_CHDIR;
876                                 goto fail;
877                         }
878
879                         free(d);
880                 }
881
882                 if (close_all_fds(fds, n_fds) < 0 ||
883                     shift_fds(fds, n_fds) < 0 ||
884                     flags_fds(fds, n_fds, context->non_blocking) < 0) {
885                         r = EXIT_FDS;
886                         goto fail;
887                 }
888
889                 if (apply_permissions) {
890
891                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
892                                 if (!context->rlimit[i])
893                                         continue;
894
895                                 if (setrlimit(i, context->rlimit[i]) < 0) {
896                                         r = EXIT_LIMITS;
897                                         goto fail;
898                                 }
899                         }
900
901                         if (context->user)
902                                 if (enforce_user(context, uid) < 0) {
903                                         r = EXIT_USER;
904                                         goto fail;
905                                 }
906
907                         /* PR_GET_SECUREBITS is not priviliged, while
908                          * PR_SET_SECUREBITS is. So to suppress
909                          * potential EPERMs we'll try not to call
910                          * PR_SET_SECUREBITS unless necessary. */
911                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
912                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
913                                         r = EXIT_SECUREBITS;
914                                         goto fail;
915                                 }
916
917                         if (context->capabilities)
918                                 if (cap_set_proc(context->capabilities) < 0) {
919                                         r = EXIT_CAPABILITIES;
920                                         goto fail;
921                                 }
922                 }
923
924                 if (!(our_env = new0(char*, 6))) {
925                         r = EXIT_MEMORY;
926                         goto fail;
927                 }
928
929                 if (n_fds > 0)
930                         if (asprintf(our_env + n_env++, "LISTEN_PID=%llu", (unsigned long long) getpid()) < 0 ||
931                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
932                                 r = EXIT_MEMORY;
933                                 goto fail;
934                         }
935
936                 if (home)
937                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
938                                 r = EXIT_MEMORY;
939                                 goto fail;
940                         }
941
942                 if (username)
943                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
944                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
945                                 r = EXIT_MEMORY;
946                                 goto fail;
947                         }
948
949                 if (!(final_env = strv_env_merge(environ, our_env, context->environment, NULL))) {
950                         r = EXIT_MEMORY;
951                         goto fail;
952                 }
953
954                 execve(command->path, command->argv, final_env);
955                 r = EXIT_EXEC;
956
957         fail:
958                 strv_free(our_env);
959                 strv_free(final_env);
960
961                 if (saved_stdin >= 0)
962                         close_nointr_nofail(saved_stdin);
963
964                 if (saved_stdout >= 0)
965                         close_nointr_nofail(saved_stdout);
966
967                 _exit(r);
968         }
969
970         /* We add the new process to the cgroup both in the child (so
971          * that we can be sure that no user code is ever executed
972          * outside of the cgroup) and in the parent (so that we can be
973          * sure that when we kill the cgroup the process will be
974          * killed too). */
975         if (cgroup_bondings)
976                 if ((r = cgroup_bonding_install_list(cgroup_bondings, pid)) < 0) {
977                         r = EXIT_CGROUP;
978                         goto fail;
979                 }
980
981         log_debug("Forked %s as %llu", command->path, (unsigned long long) pid);
982
983         command->exec_status.pid = pid;
984         command->exec_status.start_timestamp = now(CLOCK_REALTIME);
985
986         *ret = pid;
987         return 0;
988 }
989
990 void exec_context_init(ExecContext *c) {
991         assert(c);
992
993         c->umask = 0002;
994         c->oom_adjust = 0;
995         c->oom_adjust_set = false;
996         c->nice = 0;
997         c->nice_set = false;
998         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
999         c->ioprio_set = false;
1000         c->cpu_sched_policy = SCHED_OTHER;
1001         c->cpu_sched_priority = 0;
1002         c->cpu_sched_set = false;
1003         CPU_ZERO(&c->cpu_affinity);
1004         c->cpu_affinity_set = false;
1005         c->timer_slack_ns = 0;
1006         c->timer_slack_ns_set = false;
1007
1008         c->cpu_sched_reset_on_fork = false;
1009         c->non_blocking = false;
1010
1011         c->std_input = 0;
1012         c->std_output = 0;
1013         c->std_error = 0;
1014         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1015
1016         c->secure_bits = 0;
1017         c->capability_bounding_set_drop = 0;
1018 }
1019
1020 void exec_context_done(ExecContext *c) {
1021         unsigned l;
1022
1023         assert(c);
1024
1025         strv_free(c->environment);
1026         c->environment = NULL;
1027
1028         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1029                 free(c->rlimit[l]);
1030                 c->rlimit[l] = NULL;
1031         }
1032
1033         free(c->working_directory);
1034         c->working_directory = NULL;
1035         free(c->root_directory);
1036         c->root_directory = NULL;
1037
1038         free(c->tty_path);
1039         c->tty_path = NULL;
1040
1041         free(c->syslog_identifier);
1042         c->syslog_identifier = NULL;
1043
1044         free(c->user);
1045         c->user = NULL;
1046
1047         free(c->group);
1048         c->group = NULL;
1049
1050         strv_free(c->supplementary_groups);
1051         c->supplementary_groups = NULL;
1052
1053         if (c->capabilities) {
1054                 cap_free(c->capabilities);
1055                 c->capabilities = NULL;
1056         }
1057 }
1058
1059 void exec_command_done(ExecCommand *c) {
1060         assert(c);
1061
1062         free(c->path);
1063         c->path = NULL;
1064
1065         strv_free(c->argv);
1066         c->argv = NULL;
1067 }
1068
1069 void exec_command_done_array(ExecCommand *c, unsigned n) {
1070         unsigned i;
1071
1072         for (i = 0; i < n; i++)
1073                 exec_command_done(c+i);
1074 }
1075
1076 void exec_command_free_list(ExecCommand *c) {
1077         ExecCommand *i;
1078
1079         while ((i = c)) {
1080                 LIST_REMOVE(ExecCommand, command, c, i);
1081                 exec_command_done(i);
1082                 free(i);
1083         }
1084 }
1085
1086 void exec_command_free_array(ExecCommand **c, unsigned n) {
1087         unsigned i;
1088
1089         for (i = 0; i < n; i++) {
1090                 exec_command_free_list(c[i]);
1091                 c[i] = NULL;
1092         }
1093 }
1094
1095 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1096         char ** e;
1097         unsigned i;
1098
1099         assert(c);
1100         assert(f);
1101
1102         if (!prefix)
1103                 prefix = "";
1104
1105         fprintf(f,
1106                 "%sUMask: %04o\n"
1107                 "%sWorkingDirectory: %s\n"
1108                 "%sRootDirectory: %s\n"
1109                 "%sNonBlocking: %s\n",
1110                 prefix, c->umask,
1111                 prefix, c->working_directory ? c->working_directory : "/",
1112                 prefix, c->root_directory ? c->root_directory : "/",
1113                 prefix, yes_no(c->non_blocking));
1114
1115         if (c->environment)
1116                 for (e = c->environment; *e; e++)
1117                         fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1118
1119         if (c->nice_set)
1120                 fprintf(f,
1121                         "%sNice: %i\n",
1122                         prefix, c->nice);
1123
1124         if (c->oom_adjust_set)
1125                 fprintf(f,
1126                         "%sOOMAdjust: %i\n",
1127                         prefix, c->oom_adjust);
1128
1129         for (i = 0; i < RLIM_NLIMITS; i++)
1130                 if (c->rlimit[i])
1131                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1132
1133         if (c->ioprio_set)
1134                 fprintf(f,
1135                         "%sIOSchedulingClass: %s\n"
1136                         "%sIOPriority: %i\n",
1137                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1138                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1139
1140         if (c->cpu_sched_set)
1141                 fprintf(f,
1142                         "%sCPUSchedulingPolicy: %s\n"
1143                         "%sCPUSchedulingPriority: %i\n"
1144                         "%sCPUSchedulingResetOnFork: %s\n",
1145                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1146                         prefix, c->cpu_sched_priority,
1147                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1148
1149         if (c->cpu_affinity_set) {
1150                 fprintf(f, "%sCPUAffinity:", prefix);
1151                 for (i = 0; i < CPU_SETSIZE; i++)
1152                         if (CPU_ISSET(i, &c->cpu_affinity))
1153                                 fprintf(f, " %i", i);
1154                 fputs("\n", f);
1155         }
1156
1157         if (c->timer_slack_ns_set)
1158                 fprintf(f, "%sTimerSlackNS: %lu\n", prefix, c->timer_slack_ns);
1159
1160         fprintf(f,
1161                 "%sStandardInput: %s\n"
1162                 "%sStandardOutput: %s\n"
1163                 "%sStandardError: %s\n",
1164                 prefix, exec_input_to_string(c->std_input),
1165                 prefix, exec_output_to_string(c->std_output),
1166                 prefix, exec_output_to_string(c->std_error));
1167
1168         if (c->tty_path)
1169                 fprintf(f,
1170                         "%sTTYPath: %s\n",
1171                         prefix, c->tty_path);
1172
1173         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KERNEL ||
1174             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KERNEL)
1175                 fprintf(f,
1176                         "%sSyslogFacility: %s\n"
1177                         "%sSyslogLevel: %s\n",
1178                         prefix, log_facility_to_string(LOG_FAC(c->syslog_priority)),
1179                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1180
1181         if (c->capabilities) {
1182                 char *t;
1183                 if ((t = cap_to_text(c->capabilities, NULL))) {
1184                         fprintf(f, "%sCapabilities: %s\n",
1185                                 prefix, t);
1186                         cap_free(t);
1187                 }
1188         }
1189
1190         if (c->secure_bits)
1191                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1192                         prefix,
1193                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1194                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1195                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1196                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1197                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1198                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1199
1200         if (c->capability_bounding_set_drop) {
1201                 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
1202
1203                 for (i = 0; i <= CAP_LAST_CAP; i++)
1204                         if (c->capability_bounding_set_drop & (1 << i)) {
1205                                 char *t;
1206
1207                                 if ((t = cap_to_name(i))) {
1208                                         fprintf(f, " %s", t);
1209                                         free(t);
1210                                 }
1211                         }
1212
1213                 fputs("\n", f);
1214         }
1215
1216         if (c->user)
1217                 fprintf(f, "%sUser: %s", prefix, c->user);
1218         if (c->group)
1219                 fprintf(f, "%sGroup: %s", prefix, c->group);
1220
1221         if (c->supplementary_groups) {
1222                 char **g;
1223
1224                 fprintf(f, "%sSupplementaryGroups:", prefix);
1225
1226                 STRV_FOREACH(g, c->supplementary_groups)
1227                         fprintf(f, " %s", *g);
1228
1229                 fputs("\n", f);
1230         }
1231 }
1232
1233 void exec_status_fill(ExecStatus *s, pid_t pid, int code, int status) {
1234         assert(s);
1235
1236         s->pid = pid;
1237         s->exit_timestamp = now(CLOCK_REALTIME);
1238
1239         s->code = code;
1240         s->status = status;
1241 }
1242
1243 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1244         char buf[FORMAT_TIMESTAMP_MAX];
1245
1246         assert(s);
1247         assert(f);
1248
1249         if (!prefix)
1250                 prefix = "";
1251
1252         if (s->pid <= 0)
1253                 return;
1254
1255         fprintf(f,
1256                 "%sPID: %llu\n",
1257                 prefix, (unsigned long long) s->pid);
1258
1259         if (s->start_timestamp > 0)
1260                 fprintf(f,
1261                         "%sStart Timestamp: %s\n",
1262                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp));
1263
1264         if (s->exit_timestamp > 0)
1265                 fprintf(f,
1266                         "%sExit Timestamp: %s\n"
1267                         "%sExit Code: %s\n"
1268                         "%sExit Status: %i\n",
1269                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp),
1270                         prefix, sigchld_code_to_string(s->code),
1271                         prefix, s->status);
1272 }
1273
1274 char *exec_command_line(ExecCommand *c) {
1275         size_t k;
1276         char *n, *p, **a;
1277         bool first = true;
1278
1279         assert(c);
1280         assert(c->argv);
1281
1282         k = 1;
1283         STRV_FOREACH(a, c->argv)
1284                 k += strlen(*a)+3;
1285
1286         if (!(n = new(char, k)))
1287                 return NULL;
1288
1289         p = n;
1290         STRV_FOREACH(a, c->argv) {
1291
1292                 if (!first)
1293                         *(p++) = ' ';
1294                 else
1295                         first = false;
1296
1297                 if (strpbrk(*a, WHITESPACE)) {
1298                         *(p++) = '\'';
1299                         p = stpcpy(p, *a);
1300                         *(p++) = '\'';
1301                 } else
1302                         p = stpcpy(p, *a);
1303
1304         }
1305
1306         *p = 0;
1307
1308         /* FIXME: this doesn't really handle arguments that have
1309          * spaces and ticks in them */
1310
1311         return n;
1312 }
1313
1314 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1315         char *p2;
1316         const char *prefix2;
1317
1318         char *cmd;
1319
1320         assert(c);
1321         assert(f);
1322
1323         if (!prefix)
1324                 prefix = "";
1325         p2 = strappend(prefix, "\t");
1326         prefix2 = p2 ? p2 : prefix;
1327
1328         cmd = exec_command_line(c);
1329
1330         fprintf(f,
1331                 "%sCommand Line: %s\n",
1332                 prefix, cmd ? cmd : strerror(ENOMEM));
1333
1334         free(cmd);
1335
1336         exec_status_dump(&c->exec_status, f, prefix2);
1337
1338         free(p2);
1339 }
1340
1341 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1342         assert(f);
1343
1344         if (!prefix)
1345                 prefix = "";
1346
1347         LIST_FOREACH(command, c, c)
1348                 exec_command_dump(c, f, prefix);
1349 }
1350
1351 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1352         ExecCommand *end;
1353
1354         assert(l);
1355         assert(e);
1356
1357         if (*l) {
1358                 /* It's kinda important that we keep the order here */
1359                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1360                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1361         } else
1362               *l = e;
1363 }
1364
1365 int exec_command_set(ExecCommand *c, const char *path, ...) {
1366         va_list ap;
1367         char **l, *p;
1368
1369         assert(c);
1370         assert(path);
1371
1372         va_start(ap, path);
1373         l = strv_new_ap(path, ap);
1374         va_end(ap);
1375
1376         if (!l)
1377                 return -ENOMEM;
1378
1379         if (!(p = strdup(path))) {
1380                 strv_free(l);
1381                 return -ENOMEM;
1382         }
1383
1384         free(c->path);
1385         c->path = p;
1386
1387         strv_free(c->argv);
1388         c->argv = l;
1389
1390         return 0;
1391 }
1392
1393 const char* exit_status_to_string(ExitStatus status) {
1394
1395         /* We cast to int here, so that -Wenum doesn't complain that
1396          * EXIT_SUCCESS/EXIT_FAILURE aren't in the enum */
1397
1398         switch ((int) status) {
1399
1400         case EXIT_SUCCESS:
1401                 return "SUCCESS";
1402
1403         case EXIT_FAILURE:
1404                 return "FAILURE";
1405
1406         case EXIT_INVALIDARGUMENT:
1407                 return "INVALIDARGUMENT";
1408
1409         case EXIT_NOTIMPLEMENTED:
1410                 return "NOTIMPLEMENTED";
1411
1412         case EXIT_NOPERMISSION:
1413                 return "NOPERMISSION";
1414
1415         case EXIT_NOTINSTALLED:
1416                 return "NOTINSSTALLED";
1417
1418         case EXIT_NOTCONFIGURED:
1419                 return "NOTCONFIGURED";
1420
1421         case EXIT_NOTRUNNING:
1422                 return "NOTRUNNING";
1423
1424         case EXIT_CHDIR:
1425                 return "CHDIR";
1426
1427         case EXIT_NICE:
1428                 return "NICE";
1429
1430         case EXIT_FDS:
1431                 return "FDS";
1432
1433         case EXIT_EXEC:
1434                 return "EXEC";
1435
1436         case EXIT_MEMORY:
1437                 return "MEMORY";
1438
1439         case EXIT_LIMITS:
1440                 return "LIMITS";
1441
1442         case EXIT_OOM_ADJUST:
1443                 return "OOM_ADJUST";
1444
1445         case EXIT_SIGNAL_MASK:
1446                 return "SIGNAL_MASK";
1447
1448         case EXIT_STDIN:
1449                 return "STDIN";
1450
1451         case EXIT_STDOUT:
1452                 return "STDOUT";
1453
1454         case EXIT_CHROOT:
1455                 return "CHROOT";
1456
1457         case EXIT_IOPRIO:
1458                 return "IOPRIO";
1459
1460         case EXIT_TIMERSLACK:
1461                 return "TIMERSLACK";
1462
1463         case EXIT_SECUREBITS:
1464                 return "SECUREBITS";
1465
1466         case EXIT_SETSCHEDULER:
1467                 return "SETSCHEDULER";
1468
1469         case EXIT_CPUAFFINITY:
1470                 return "CPUAFFINITY";
1471
1472         case EXIT_GROUP:
1473                 return "GROUP";
1474
1475         case EXIT_USER:
1476                 return "USER";
1477
1478         case EXIT_CAPABILITIES:
1479                 return "CAPABILITIES";
1480
1481         case EXIT_CGROUP:
1482                 return "CGROUP";
1483
1484         case EXIT_SETSID:
1485                 return "SETSID";
1486
1487         case EXIT_CONFIRM:
1488                 return "CONFIRM";
1489
1490         case EXIT_STDERR:
1491                 return "STDERR";
1492
1493         default:
1494                 return NULL;
1495         }
1496 }
1497
1498 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1499         [EXEC_INPUT_NULL] = "null",
1500         [EXEC_INPUT_TTY] = "tty",
1501         [EXEC_INPUT_TTY_FORCE] = "tty-force",
1502         [EXEC_INPUT_TTY_FAIL] = "tty-fail"
1503 };
1504
1505 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
1506         [EXEC_OUTPUT_INHERIT] = "inherit",
1507         [EXEC_OUTPUT_NULL] = "null",
1508         [EXEC_OUTPUT_TTY] = "tty",
1509         [EXEC_OUTPUT_SYSLOG] = "syslog",
1510         [EXEC_OUTPUT_KERNEL] = "kernel"
1511 };
1512
1513 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
1514
1515 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);