2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
27 #include <sys/capability.h>
28 #include <sys/eventfd.h>
30 #include <sys/personality.h>
31 #include <sys/prctl.h>
32 #include <sys/socket.h>
39 #include <security/pam_appl.h>
43 #include <selinux/selinux.h>
51 #include <sys/apparmor.h>
54 #include "sd-messages.h"
57 #include "alloc-util.h"
59 #include "apparmor-util.h"
64 #include "capability-util.h"
67 #include "errno-list.h"
69 #include "exit-status.h"
72 #include "formats-util.h"
74 #include "glob-util.h"
81 #include "namespace.h"
82 #include "parse-util.h"
83 #include "path-util.h"
84 #include "process-util.h"
85 #include "rlimit-util.h"
88 #include "seccomp-util.h"
90 #include "securebits.h"
91 #include "selinux-util.h"
92 #include "signal-util.h"
93 #include "smack-util.h"
95 #include "string-table.h"
96 #include "string-util.h"
98 #include "syslog-util.h"
99 #include "terminal-util.h"
101 #include "user-util.h"
103 #include "utmp-wtmp.h"
105 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
106 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
108 /* This assumes there is a 'tty' group */
109 #define TTY_MODE 0620
111 #define SNDBUF_SIZE (8*1024*1024)
113 static int shift_fds(int fds
[], unsigned n_fds
) {
114 int start
, restart_from
;
119 /* Modifies the fds array! (sorts it) */
129 for (i
= start
; i
< (int) n_fds
; i
++) {
132 /* Already at right index? */
136 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
143 /* Hmm, the fd we wanted isn't free? Then
144 * let's remember that and try again from here */
145 if (nfd
!= i
+3 && restart_from
< 0)
149 if (restart_from
< 0)
152 start
= restart_from
;
158 static int flags_fds(const int fds
[], unsigned n_fds
, bool nonblock
) {
167 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
169 for (i
= 0; i
< n_fds
; i
++) {
171 r
= fd_nonblock(fds
[i
], nonblock
);
175 /* We unconditionally drop FD_CLOEXEC from the fds,
176 * since after all we want to pass these fds to our
179 r
= fd_cloexec(fds
[i
], false);
187 static const char *exec_context_tty_path(const ExecContext
*context
) {
190 if (context
->stdio_as_fds
)
193 if (context
->tty_path
)
194 return context
->tty_path
;
196 return "/dev/console";
199 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
204 path
= exec_context_tty_path(context
);
206 if (context
->tty_vhangup
) {
207 if (p
&& p
->stdin_fd
>= 0)
208 (void) terminal_vhangup_fd(p
->stdin_fd
);
210 (void) terminal_vhangup(path
);
213 if (context
->tty_reset
) {
214 if (p
&& p
->stdin_fd
>= 0)
215 (void) reset_terminal_fd(p
->stdin_fd
, true);
217 (void) reset_terminal(path
);
220 if (context
->tty_vt_disallocate
&& path
)
221 (void) vt_disallocate(path
);
224 static bool is_terminal_input(ExecInput i
) {
227 EXEC_INPUT_TTY_FORCE
,
228 EXEC_INPUT_TTY_FAIL
);
231 static bool is_terminal_output(ExecOutput o
) {
234 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
235 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
236 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
239 static bool exec_context_needs_term(const ExecContext
*c
) {
242 /* Return true if the execution context suggests we should set $TERM to something useful. */
244 if (is_terminal_input(c
->std_input
))
247 if (is_terminal_output(c
->std_output
))
250 if (is_terminal_output(c
->std_error
))
253 return !!c
->tty_path
;
256 static int open_null_as(int flags
, int nfd
) {
261 fd
= open("/dev/null", flags
|O_NOCTTY
);
266 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
274 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
275 union sockaddr_union sa
= {
276 .un
.sun_family
= AF_UNIX
,
277 .un
.sun_path
= "/run/systemd/journal/stdout",
279 uid_t olduid
= UID_INVALID
;
280 gid_t oldgid
= GID_INVALID
;
283 if (gid
!= GID_INVALID
) {
291 if (uid
!= UID_INVALID
) {
301 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
305 /* If we fail to restore the uid or gid, things will likely
306 fail later on. This should only happen if an LSM interferes. */
308 if (uid
!= UID_INVALID
)
309 (void) seteuid(olduid
);
312 if (gid
!= GID_INVALID
)
313 (void) setegid(oldgid
);
318 static int connect_logger_as(
320 const ExecContext
*context
,
330 assert(output
< _EXEC_OUTPUT_MAX
);
334 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
338 r
= connect_journal_socket(fd
, uid
, gid
);
342 if (shutdown(fd
, SHUT_RD
) < 0) {
347 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
357 context
->syslog_identifier
? context
->syslog_identifier
: ident
,
359 context
->syslog_priority
,
360 !!context
->syslog_level_prefix
,
361 output
== EXEC_OUTPUT_SYSLOG
|| output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
362 output
== EXEC_OUTPUT_KMSG
|| output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
,
363 is_terminal_output(output
));
368 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
373 static int open_terminal_as(const char *path
, mode_t mode
, int nfd
) {
379 fd
= open_terminal(path
, mode
| O_NOCTTY
);
384 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
392 static int fixup_input(ExecInput std_input
, int socket_fd
, bool apply_tty_stdin
) {
394 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
395 return EXEC_INPUT_NULL
;
397 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
398 return EXEC_INPUT_NULL
;
403 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
405 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
406 return EXEC_OUTPUT_INHERIT
;
411 static int setup_input(
412 const ExecContext
*context
,
413 const ExecParameters
*params
,
415 int named_iofds
[3]) {
422 if (params
->stdin_fd
>= 0) {
423 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
426 /* Try to make this the controlling tty, if it is a tty, and reset it */
427 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
428 (void) reset_terminal_fd(STDIN_FILENO
, true);
433 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
437 case EXEC_INPUT_NULL
:
438 return open_null_as(O_RDONLY
, STDIN_FILENO
);
441 case EXEC_INPUT_TTY_FORCE
:
442 case EXEC_INPUT_TTY_FAIL
: {
445 fd
= acquire_terminal(exec_context_tty_path(context
),
446 i
== EXEC_INPUT_TTY_FAIL
,
447 i
== EXEC_INPUT_TTY_FORCE
,
453 if (fd
!= STDIN_FILENO
) {
454 r
= dup2(fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
462 case EXEC_INPUT_SOCKET
:
463 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
465 case EXEC_INPUT_NAMED_FD
:
466 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
467 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
470 assert_not_reached("Unknown input type");
474 static int setup_output(
476 const ExecContext
*context
,
477 const ExecParameters
*params
,
484 dev_t
*journal_stream_dev
,
485 ino_t
*journal_stream_ino
) {
495 assert(journal_stream_dev
);
496 assert(journal_stream_ino
);
498 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
500 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
503 return STDOUT_FILENO
;
506 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
507 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
510 return STDERR_FILENO
;
513 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
514 o
= fixup_output(context
->std_output
, socket_fd
);
516 if (fileno
== STDERR_FILENO
) {
518 e
= fixup_output(context
->std_error
, socket_fd
);
520 /* This expects the input and output are already set up */
522 /* Don't change the stderr file descriptor if we inherit all
523 * the way and are not on a tty */
524 if (e
== EXEC_OUTPUT_INHERIT
&&
525 o
== EXEC_OUTPUT_INHERIT
&&
526 i
== EXEC_INPUT_NULL
&&
527 !is_terminal_input(context
->std_input
) &&
531 /* Duplicate from stdout if possible */
532 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
533 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
537 } else if (o
== EXEC_OUTPUT_INHERIT
) {
538 /* If input got downgraded, inherit the original value */
539 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
540 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
542 /* If the input is connected to anything that's not a /dev/null, inherit that... */
543 if (i
!= EXEC_INPUT_NULL
)
544 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
546 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
550 /* We need to open /dev/null here anew, to get the right access mode. */
551 return open_null_as(O_WRONLY
, fileno
);
556 case EXEC_OUTPUT_NULL
:
557 return open_null_as(O_WRONLY
, fileno
);
559 case EXEC_OUTPUT_TTY
:
560 if (is_terminal_input(i
))
561 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
563 /* We don't reset the terminal if this is just about output */
564 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
566 case EXEC_OUTPUT_SYSLOG
:
567 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
568 case EXEC_OUTPUT_KMSG
:
569 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
570 case EXEC_OUTPUT_JOURNAL
:
571 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
572 r
= connect_logger_as(unit
, context
, o
, ident
, fileno
, uid
, gid
);
574 log_unit_error_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
575 r
= open_null_as(O_WRONLY
, fileno
);
579 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
580 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
581 * services to detect whether they are connected to the journal or not. */
583 if (fstat(fileno
, &st
) >= 0) {
584 *journal_stream_dev
= st
.st_dev
;
585 *journal_stream_ino
= st
.st_ino
;
590 case EXEC_OUTPUT_SOCKET
:
591 assert(socket_fd
>= 0);
592 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
594 case EXEC_OUTPUT_NAMED_FD
:
595 (void) fd_nonblock(named_iofds
[fileno
], false);
596 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
599 assert_not_reached("Unknown error type");
603 static int chown_terminal(int fd
, uid_t uid
) {
608 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
612 /* This might fail. What matters are the results. */
613 (void) fchown(fd
, uid
, -1);
614 (void) fchmod(fd
, TTY_MODE
);
616 if (fstat(fd
, &st
) < 0)
619 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
625 static int setup_confirm_stdio(int *_saved_stdin
, int *_saved_stdout
) {
626 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
629 assert(_saved_stdin
);
630 assert(_saved_stdout
);
632 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
636 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
637 if (saved_stdout
< 0)
640 fd
= acquire_terminal(
645 DEFAULT_CONFIRM_USEC
);
649 r
= chown_terminal(fd
, getuid());
653 r
= reset_terminal_fd(fd
, true);
657 if (dup2(fd
, STDIN_FILENO
) < 0)
660 if (dup2(fd
, STDOUT_FILENO
) < 0)
667 *_saved_stdin
= saved_stdin
;
668 *_saved_stdout
= saved_stdout
;
670 saved_stdin
= saved_stdout
= -1;
675 _printf_(1, 2) static int write_confirm_message(const char *format
, ...) {
676 _cleanup_close_
int fd
= -1;
681 fd
= open_terminal("/dev/console", O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
685 va_start(ap
, format
);
686 vdprintf(fd
, format
, ap
);
692 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
696 assert(saved_stdout
);
700 if (*saved_stdin
>= 0)
701 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
704 if (*saved_stdout
>= 0)
705 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
708 *saved_stdin
= safe_close(*saved_stdin
);
709 *saved_stdout
= safe_close(*saved_stdout
);
714 static int ask_for_confirmation(char *response
, char **argv
) {
715 int saved_stdout
= -1, saved_stdin
= -1, r
;
716 _cleanup_free_
char *line
= NULL
;
718 r
= setup_confirm_stdio(&saved_stdin
, &saved_stdout
);
722 line
= exec_command_line(argv
);
726 r
= ask_char(response
, "yns", "Execute %s? [Yes, No, Skip] ", line
);
728 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
733 static int enforce_groups(const ExecContext
*context
, const char *username
, gid_t gid
) {
734 bool keep_groups
= false;
739 /* Lookup and set GID and supplementary group list. Here too
740 * we avoid NSS lookups for gid=0. */
742 if (context
->group
|| username
) {
743 /* First step, initialize groups from /etc/groups */
744 if (username
&& gid
!= 0) {
745 if (initgroups(username
, gid
) < 0)
751 /* Second step, set our gids */
752 if (setresgid(gid
, gid
, gid
) < 0)
756 if (context
->supplementary_groups
) {
761 /* Final step, initialize any manually set supplementary groups */
762 assert_se((ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
)) > 0);
764 if (!(gids
= new(gid_t
, ngroups_max
)))
768 k
= getgroups(ngroups_max
, gids
);
776 STRV_FOREACH(i
, context
->supplementary_groups
) {
779 if (k
>= ngroups_max
) {
785 r
= get_group_creds(&g
, gids
+k
);
794 r
= maybe_setgroups(k
, gids
);
806 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
809 /* Sets (but doesn't look up) the uid and make sure we keep the
810 * capabilities while doing so. */
812 if (context
->capability_ambient_set
!= 0) {
814 /* First step: If we need to keep capabilities but
815 * drop privileges we need to make sure we keep our
816 * caps, while we drop privileges. */
818 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
820 if (prctl(PR_GET_SECUREBITS
) != sb
)
821 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
826 /* Second step: actually set the uids */
827 if (setresuid(uid
, uid
, uid
) < 0)
830 /* At this point we should have all necessary capabilities but
831 are otherwise a normal user. However, the caps might got
832 corrupted due to the setresuid() so we need clean them up
833 later. This is done outside of this call. */
840 static int null_conv(
842 const struct pam_message
**msg
,
843 struct pam_response
**resp
,
846 /* We don't support conversations */
853 static int setup_pam(
860 int fds
[], unsigned n_fds
) {
864 static const struct pam_conv conv
= {
869 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
870 pam_handle_t
*handle
= NULL
;
872 int pam_code
= PAM_SUCCESS
, r
;
873 char **nv
, **e
= NULL
;
874 bool close_session
= false;
875 pid_t pam_pid
= 0, parent_pid
;
882 /* We set up PAM in the parent process, then fork. The child
883 * will then stay around until killed via PR_GET_PDEATHSIG or
884 * systemd via the cgroup logic. It will then remove the PAM
885 * session again. The parent process will exec() the actual
886 * daemon. We do things this way to ensure that the main PID
887 * of the daemon is the one we initially fork()ed. */
889 r
= barrier_create(&barrier
);
893 if (log_get_max_level() < LOG_DEBUG
)
896 pam_code
= pam_start(name
, user
, &conv
, &handle
);
897 if (pam_code
!= PAM_SUCCESS
) {
903 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
904 if (pam_code
!= PAM_SUCCESS
)
908 STRV_FOREACH(nv
, *env
) {
909 pam_code
= pam_putenv(handle
, *nv
);
910 if (pam_code
!= PAM_SUCCESS
)
914 pam_code
= pam_acct_mgmt(handle
, flags
);
915 if (pam_code
!= PAM_SUCCESS
)
918 pam_code
= pam_open_session(handle
, flags
);
919 if (pam_code
!= PAM_SUCCESS
)
922 close_session
= true;
924 e
= pam_getenvlist(handle
);
926 pam_code
= PAM_BUF_ERR
;
930 /* Block SIGTERM, so that we know that it won't get lost in
933 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
935 parent_pid
= getpid();
944 int sig
, ret
= EXIT_PAM
;
946 /* The child's job is to reset the PAM session on
948 barrier_set_role(&barrier
, BARRIER_CHILD
);
950 /* This string must fit in 10 chars (i.e. the length
951 * of "/sbin/init"), to look pretty in /bin/ps */
952 rename_process("(sd-pam)");
954 /* Make sure we don't keep open the passed fds in this
955 child. We assume that otherwise only those fds are
956 open here that have been opened by PAM. */
957 close_many(fds
, n_fds
);
959 /* Drop privileges - we don't need any to pam_close_session
960 * and this will make PR_SET_PDEATHSIG work in most cases.
961 * If this fails, ignore the error - but expect sd-pam threads
962 * to fail to exit normally */
964 r
= maybe_setgroups(0, NULL
);
966 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
967 if (setresgid(gid
, gid
, gid
) < 0)
968 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
969 if (setresuid(uid
, uid
, uid
) < 0)
970 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
972 (void) ignore_signals(SIGPIPE
, -1);
974 /* Wait until our parent died. This will only work if
975 * the above setresuid() succeeds, otherwise the kernel
976 * will not allow unprivileged parents kill their privileged
977 * children this way. We rely on the control groups kill logic
978 * to do the rest for us. */
979 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
982 /* Tell the parent that our setup is done. This is especially
983 * important regarding dropping privileges. Otherwise, unit
984 * setup might race against our setresuid(2) call. */
985 barrier_place(&barrier
);
987 /* Check if our parent process might already have
989 if (getppid() == parent_pid
) {
992 assert_se(sigemptyset(&ss
) >= 0);
993 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
996 if (sigwait(&ss
, &sig
) < 0) {
1003 assert(sig
== SIGTERM
);
1008 /* If our parent died we'll end the session */
1009 if (getppid() != parent_pid
) {
1010 pam_code
= pam_close_session(handle
, flags
);
1011 if (pam_code
!= PAM_SUCCESS
)
1018 pam_end(handle
, pam_code
| flags
);
1022 barrier_set_role(&barrier
, BARRIER_PARENT
);
1024 /* If the child was forked off successfully it will do all the
1025 * cleanups, so forget about the handle here. */
1028 /* Unblock SIGTERM again in the parent */
1029 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1031 /* We close the log explicitly here, since the PAM modules
1032 * might have opened it, but we don't want this fd around. */
1035 /* Synchronously wait for the child to initialize. We don't care for
1036 * errors as we cannot recover. However, warn loudly if it happens. */
1037 if (!barrier_place_and_sync(&barrier
))
1038 log_error("PAM initialization failed");
1046 if (pam_code
!= PAM_SUCCESS
) {
1047 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1048 r
= -EPERM
; /* PAM errors do not map to errno */
1050 log_error_errno(r
, "PAM failed: %m");
1054 pam_code
= pam_close_session(handle
, flags
);
1056 pam_end(handle
, pam_code
| flags
);
1068 static void rename_process_from_path(const char *path
) {
1069 char process_name
[11];
1073 /* This resulting string must fit in 10 chars (i.e. the length
1074 * of "/sbin/init") to look pretty in /bin/ps */
1078 rename_process("(...)");
1084 /* The end of the process name is usually more
1085 * interesting, since the first bit might just be
1091 process_name
[0] = '(';
1092 memcpy(process_name
+1, p
, l
);
1093 process_name
[1+l
] = ')';
1094 process_name
[1+l
+1] = 0;
1096 rename_process(process_name
);
1101 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1102 if (!is_seccomp_available()) {
1104 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1111 static int apply_seccomp(const Unit
* u
, const ExecContext
*c
) {
1112 uint32_t negative_action
, action
;
1113 scmp_filter_ctx
*seccomp
;
1120 if (skip_seccomp_unavailable(u
, "syscall filtering"))
1123 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1125 seccomp
= seccomp_init(c
->syscall_whitelist
? negative_action
: SCMP_ACT_ALLOW
);
1129 if (c
->syscall_archs
) {
1131 SET_FOREACH(id
, c
->syscall_archs
, i
) {
1132 r
= seccomp_arch_add(seccomp
, PTR_TO_UINT32(id
) - 1);
1140 r
= seccomp_add_secondary_archs(seccomp
);
1145 action
= c
->syscall_whitelist
? SCMP_ACT_ALLOW
: negative_action
;
1146 SET_FOREACH(id
, c
->syscall_filter
, i
) {
1147 r
= seccomp_rule_add(seccomp
, action
, PTR_TO_INT(id
) - 1, 0);
1152 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1156 r
= seccomp_load(seccomp
);
1159 seccomp_release(seccomp
);
1163 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1164 scmp_filter_ctx
*seccomp
;
1170 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1173 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1177 r
= seccomp_add_secondary_archs(seccomp
);
1181 if (c
->address_families_whitelist
) {
1182 int af
, first
= 0, last
= 0;
1185 /* If this is a whitelist, we first block the address
1186 * families that are out of range and then everything
1187 * that is not in the set. First, we find the lowest
1188 * and highest address family in the set. */
1190 SET_FOREACH(afp
, c
->address_families
, i
) {
1191 af
= PTR_TO_INT(afp
);
1193 if (af
<= 0 || af
>= af_max())
1196 if (first
== 0 || af
< first
)
1199 if (last
== 0 || af
> last
)
1203 assert((first
== 0) == (last
== 0));
1207 /* No entries in the valid range, block everything */
1208 r
= seccomp_rule_add(
1210 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1218 /* Block everything below the first entry */
1219 r
= seccomp_rule_add(
1221 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1224 SCMP_A0(SCMP_CMP_LT
, first
));
1228 /* Block everything above the last entry */
1229 r
= seccomp_rule_add(
1231 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1234 SCMP_A0(SCMP_CMP_GT
, last
));
1238 /* Block everything between the first and last
1240 for (af
= 1; af
< af_max(); af
++) {
1242 if (set_contains(c
->address_families
, INT_TO_PTR(af
)))
1245 r
= seccomp_rule_add(
1247 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1250 SCMP_A0(SCMP_CMP_EQ
, af
));
1259 /* If this is a blacklist, then generate one rule for
1260 * each address family that are then combined in OR
1263 SET_FOREACH(af
, c
->address_families
, i
) {
1265 r
= seccomp_rule_add(
1267 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1270 SCMP_A0(SCMP_CMP_EQ
, PTR_TO_INT(af
)));
1276 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1280 r
= seccomp_load(seccomp
);
1283 seccomp_release(seccomp
);
1287 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1288 scmp_filter_ctx
*seccomp
;
1293 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1296 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1300 r
= seccomp_add_secondary_archs(seccomp
);
1304 r
= seccomp_rule_add(
1306 SCMP_ACT_ERRNO(EPERM
),
1309 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
|PROT_WRITE
, PROT_EXEC
|PROT_WRITE
));
1313 r
= seccomp_rule_add(
1315 SCMP_ACT_ERRNO(EPERM
),
1318 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
, PROT_EXEC
));
1322 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1326 r
= seccomp_load(seccomp
);
1329 seccomp_release(seccomp
);
1333 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1334 static const int permitted_policies
[] = {
1340 scmp_filter_ctx
*seccomp
;
1342 int r
, p
, max_policy
= 0;
1346 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1349 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1353 r
= seccomp_add_secondary_archs(seccomp
);
1357 /* Determine the highest policy constant we want to allow */
1358 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1359 if (permitted_policies
[i
] > max_policy
)
1360 max_policy
= permitted_policies
[i
];
1362 /* Go through all policies with lower values than that, and block them -- unless they appear in the
1364 for (p
= 0; p
< max_policy
; p
++) {
1367 /* Check if this is in the whitelist. */
1368 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1369 if (permitted_policies
[i
] == p
) {
1377 /* Deny this policy */
1378 r
= seccomp_rule_add(
1380 SCMP_ACT_ERRNO(EPERM
),
1381 SCMP_SYS(sched_setscheduler
),
1383 SCMP_A1(SCMP_CMP_EQ
, p
));
1388 /* Blacklist all other policies, i.e. the ones with higher values. Note that all comparisons are unsigned here,
1389 * hence no need no check for < 0 values. */
1390 r
= seccomp_rule_add(
1392 SCMP_ACT_ERRNO(EPERM
),
1393 SCMP_SYS(sched_setscheduler
),
1395 SCMP_A1(SCMP_CMP_GT
, max_policy
));
1399 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1403 r
= seccomp_load(seccomp
);
1406 seccomp_release(seccomp
);
1410 static int apply_protect_sysctl(Unit
*u
, const ExecContext
*c
) {
1411 scmp_filter_ctx
*seccomp
;
1416 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1417 * let's protect even those systems where this is left on in the kernel. */
1419 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1422 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1426 r
= seccomp_add_secondary_archs(seccomp
);
1430 r
= seccomp_rule_add(
1432 SCMP_ACT_ERRNO(EPERM
),
1438 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1442 r
= seccomp_load(seccomp
);
1445 seccomp_release(seccomp
);
1449 static int apply_protect_kernel_modules(Unit
*u
, const ExecContext
*c
) {
1450 static const int module_syscalls
[] = {
1451 SCMP_SYS(delete_module
),
1452 SCMP_SYS(finit_module
),
1453 SCMP_SYS(init_module
),
1456 scmp_filter_ctx
*seccomp
;
1462 /* Turn of module syscalls on ProtectKernelModules=yes */
1464 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1467 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1471 r
= seccomp_add_secondary_archs(seccomp
);
1475 for (i
= 0; i
< ELEMENTSOF(module_syscalls
); i
++) {
1476 r
= seccomp_rule_add(seccomp
, SCMP_ACT_ERRNO(EPERM
),
1477 module_syscalls
[i
], 0);
1482 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1486 r
= seccomp_load(seccomp
);
1489 seccomp_release(seccomp
);
1493 static int apply_private_devices(Unit
*u
, const ExecContext
*c
) {
1494 const SystemCallFilterSet
*set
;
1495 scmp_filter_ctx
*seccomp
;
1497 bool syscalls_found
= false;
1502 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1504 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1507 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1511 r
= seccomp_add_secondary_archs(seccomp
);
1515 for (set
= syscall_filter_sets
; set
->set_name
; set
++)
1516 if (streq(set
->set_name
, "@raw-io")) {
1517 syscalls_found
= true;
1521 /* We should never fail here */
1522 if (!syscalls_found
) {
1527 NULSTR_FOREACH(sys
, set
->value
) {
1531 #ifndef __NR_s390_pci_mmio_read
1532 if (streq(sys
, "s390_pci_mmio_read"))
1535 #ifndef __NR_s390_pci_mmio_write
1536 if (streq(sys
, "s390_pci_mmio_write"))
1543 id
= seccomp_syscall_resolve_name(sys
);
1545 r
= seccomp_rule_add(
1547 SCMP_ACT_ERRNO(EPERM
),
1553 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1557 r
= seccomp_load(seccomp
);
1560 seccomp_release(seccomp
);
1566 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1569 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1570 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1572 if (idle_pipe
[0] >= 0) {
1575 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1577 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1580 /* Signal systemd that we are bored and want to continue. */
1581 n
= write(idle_pipe
[3], "x", 1);
1583 /* Wait for systemd to react to the signal above. */
1584 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1587 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1591 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1594 static int build_environment(
1596 const ExecContext
*c
,
1597 const ExecParameters
*p
,
1600 const char *username
,
1602 dev_t journal_stream_dev
,
1603 ino_t journal_stream_ino
,
1606 _cleanup_strv_free_
char **our_env
= NULL
;
1614 our_env
= new0(char*, 14);
1619 _cleanup_free_
char *joined
= NULL
;
1621 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid()) < 0)
1623 our_env
[n_env
++] = x
;
1625 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1627 our_env
[n_env
++] = x
;
1629 joined
= strv_join(p
->fd_names
, ":");
1633 x
= strjoin("LISTEN_FDNAMES=", joined
, NULL
);
1636 our_env
[n_env
++] = x
;
1639 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1640 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid()) < 0)
1642 our_env
[n_env
++] = x
;
1644 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1646 our_env
[n_env
++] = x
;
1649 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1650 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1651 * check the database directly. */
1652 if (unit_has_name(u
, SPECIAL_DBUS_SERVICE
)) {
1653 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1656 our_env
[n_env
++] = x
;
1660 x
= strappend("HOME=", home
);
1663 our_env
[n_env
++] = x
;
1667 x
= strappend("LOGNAME=", username
);
1670 our_env
[n_env
++] = x
;
1672 x
= strappend("USER=", username
);
1675 our_env
[n_env
++] = x
;
1679 x
= strappend("SHELL=", shell
);
1682 our_env
[n_env
++] = x
;
1685 if (!sd_id128_is_null(u
->invocation_id
)) {
1686 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1689 our_env
[n_env
++] = x
;
1692 if (exec_context_needs_term(c
)) {
1693 const char *tty_path
, *term
= NULL
;
1695 tty_path
= exec_context_tty_path(c
);
1697 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1698 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1699 * passes to PID 1 ends up all the way in the console login shown. */
1701 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1702 term
= getenv("TERM");
1704 term
= default_term_for_tty(tty_path
);
1706 x
= strappend("TERM=", term
);
1709 our_env
[n_env
++] = x
;
1712 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1713 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1716 our_env
[n_env
++] = x
;
1719 our_env
[n_env
++] = NULL
;
1720 assert(n_env
<= 12);
1728 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1729 _cleanup_strv_free_
char **pass_env
= NULL
;
1730 size_t n_env
= 0, n_bufsize
= 0;
1733 STRV_FOREACH(i
, c
->pass_environment
) {
1734 _cleanup_free_
char *x
= NULL
;
1740 x
= strjoin(*i
, "=", v
, NULL
);
1743 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1745 pass_env
[n_env
++] = x
;
1746 pass_env
[n_env
] = NULL
;
1756 static bool exec_needs_mount_namespace(
1757 const ExecContext
*context
,
1758 const ExecParameters
*params
,
1759 ExecRuntime
*runtime
) {
1764 if (!strv_isempty(context
->read_write_paths
) ||
1765 !strv_isempty(context
->read_only_paths
) ||
1766 !strv_isempty(context
->inaccessible_paths
))
1769 if (context
->mount_flags
!= 0)
1772 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1775 if (context
->private_devices
||
1776 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1777 context
->protect_home
!= PROTECT_HOME_NO
||
1778 context
->protect_kernel_tunables
||
1779 context
->protect_kernel_modules
||
1780 context
->protect_control_groups
)
1786 static int setup_private_users(uid_t uid
, gid_t gid
) {
1787 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1788 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1789 _cleanup_close_
int unshare_ready_fd
= -1;
1790 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1796 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1797 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1798 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1799 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1800 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1801 * continues execution normally. */
1803 if (uid
!= 0 && uid_is_valid(uid
))
1805 "0 0 1\n" /* Map root → root */
1806 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1807 uid
, uid
); /* The case where the above is the same */
1809 uid_map
= strdup("0 0 1\n");
1813 if (gid
!= 0 && gid_is_valid(gid
))
1815 "0 0 1\n" /* Map root → root */
1816 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1819 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1823 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1825 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1826 if (unshare_ready_fd
< 0)
1829 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1831 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1839 _cleanup_close_
int fd
= -1;
1843 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1844 * here, after the parent opened its own user namespace. */
1847 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1849 /* Wait until the parent unshared the user namespace */
1850 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1855 /* Disable the setgroups() system call in the child user namespace, for good. */
1856 a
= procfs_file_alloca(ppid
, "setgroups");
1857 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1859 if (errno
!= ENOENT
) {
1864 /* If the file is missing the kernel is too old, let's continue anyway. */
1866 if (write(fd
, "deny\n", 5) < 0) {
1871 fd
= safe_close(fd
);
1874 /* First write the GID map */
1875 a
= procfs_file_alloca(ppid
, "gid_map");
1876 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1881 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1885 fd
= safe_close(fd
);
1887 /* The write the UID map */
1888 a
= procfs_file_alloca(ppid
, "uid_map");
1889 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1894 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1899 _exit(EXIT_SUCCESS
);
1902 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1903 _exit(EXIT_FAILURE
);
1906 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1908 if (unshare(CLONE_NEWUSER
) < 0)
1911 /* Let the child know that the namespace is ready now */
1912 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1915 /* Try to read an error code from the child */
1916 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1919 if (n
== sizeof(r
)) { /* an error code was sent to us */
1924 if (n
!= 0) /* on success we should have read 0 bytes */
1927 r
= wait_for_terminate(pid
, &si
);
1932 /* If something strange happened with the child, let's consider this fatal, too */
1933 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1939 static int setup_runtime_directory(
1940 const ExecContext
*context
,
1941 const ExecParameters
*params
,
1951 STRV_FOREACH(rt
, context
->runtime_directory
) {
1952 _cleanup_free_
char *p
;
1954 p
= strjoin(params
->runtime_prefix
, "/", *rt
, NULL
);
1958 r
= mkdir_p_label(p
, context
->runtime_directory_mode
);
1962 r
= chmod_and_chown(p
, context
->runtime_directory_mode
, uid
, gid
);
1970 static int setup_smack(
1971 const ExecContext
*context
,
1972 const ExecCommand
*command
) {
1980 if (!mac_smack_use())
1983 if (context
->smack_process_label
) {
1984 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1988 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1990 _cleanup_free_
char *exec_label
= NULL
;
1992 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1993 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1996 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2006 static int compile_read_write_paths(
2007 const ExecContext
*context
,
2008 const ExecParameters
*params
,
2011 _cleanup_strv_free_
char **l
= NULL
;
2014 /* Compile the list of writable paths. This is the combination of the explicitly configured paths, plus all
2015 * runtime directories. */
2017 if (strv_isempty(context
->read_write_paths
) &&
2018 strv_isempty(context
->runtime_directory
)) {
2019 *ret
= NULL
; /* NOP if neither is set */
2023 l
= strv_copy(context
->read_write_paths
);
2027 STRV_FOREACH(rt
, context
->runtime_directory
) {
2030 s
= strjoin(params
->runtime_prefix
, "/", *rt
, NULL
);
2034 if (strv_consume(&l
, s
) < 0)
2044 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
2052 array
[(*n
)++] = pair
[0];
2054 array
[(*n
)++] = pair
[1];
2057 static int close_remaining_fds(
2058 const ExecParameters
*params
,
2059 ExecRuntime
*runtime
,
2060 DynamicCreds
*dcreds
,
2063 int *fds
, unsigned n_fds
) {
2065 unsigned n_dont_close
= 0;
2066 int dont_close
[n_fds
+ 12];
2070 if (params
->stdin_fd
>= 0)
2071 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2072 if (params
->stdout_fd
>= 0)
2073 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2074 if (params
->stderr_fd
>= 0)
2075 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2078 dont_close
[n_dont_close
++] = socket_fd
;
2080 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2081 n_dont_close
+= n_fds
;
2085 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2089 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2091 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2094 if (user_lookup_fd
>= 0)
2095 dont_close
[n_dont_close
++] = user_lookup_fd
;
2097 return close_all_fds(dont_close
, n_dont_close
);
2100 static bool context_has_address_families(const ExecContext
*c
) {
2103 return c
->address_families_whitelist
||
2104 !set_isempty(c
->address_families
);
2107 static bool context_has_syscall_filters(const ExecContext
*c
) {
2110 return c
->syscall_whitelist
||
2111 !set_isempty(c
->syscall_filter
) ||
2112 !set_isempty(c
->syscall_archs
);
2115 static bool context_has_no_new_privileges(const ExecContext
*c
) {
2118 if (c
->no_new_privileges
)
2121 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
2124 return context_has_address_families(c
) || /* we need NNP if we have any form of seccomp and are unprivileged */
2125 c
->memory_deny_write_execute
||
2126 c
->restrict_realtime
||
2127 c
->protect_kernel_tunables
||
2128 c
->protect_kernel_modules
||
2129 c
->private_devices
||
2130 context_has_syscall_filters(c
);
2133 static int send_user_lookup(
2141 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2142 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2145 if (user_lookup_fd
< 0)
2148 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2151 if (writev(user_lookup_fd
,
2153 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2154 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2155 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2161 static int exec_child(
2163 ExecCommand
*command
,
2164 const ExecContext
*context
,
2165 const ExecParameters
*params
,
2166 ExecRuntime
*runtime
,
2167 DynamicCreds
*dcreds
,
2171 int *fds
, unsigned n_fds
,
2176 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2177 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2178 const char *username
= NULL
, *home
= NULL
, *shell
= NULL
, *wd
;
2179 dev_t journal_stream_dev
= 0;
2180 ino_t journal_stream_ino
= 0;
2181 bool needs_mount_namespace
;
2182 uid_t uid
= UID_INVALID
;
2183 gid_t gid
= GID_INVALID
;
2190 assert(exit_status
);
2192 rename_process_from_path(command
->path
);
2194 /* We reset exactly these signals, since they are the
2195 * only ones we set to SIG_IGN in the main daemon. All
2196 * others we leave untouched because we set them to
2197 * SIG_DFL or a valid handler initially, both of which
2198 * will be demoted to SIG_DFL. */
2199 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2200 SIGNALS_IGNORE
, -1);
2202 if (context
->ignore_sigpipe
)
2203 (void) ignore_signals(SIGPIPE
, -1);
2205 r
= reset_signal_mask();
2207 *exit_status
= EXIT_SIGNAL_MASK
;
2211 if (params
->idle_pipe
)
2212 do_idle_pipe_dance(params
->idle_pipe
);
2214 /* Close sockets very early to make sure we don't
2215 * block init reexecution because it cannot bind its
2220 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2222 *exit_status
= EXIT_FDS
;
2226 if (!context
->same_pgrp
)
2228 *exit_status
= EXIT_SETSID
;
2232 exec_context_tty_reset(context
, params
);
2234 if (params
->flags
& EXEC_CONFIRM_SPAWN
) {
2237 r
= ask_for_confirmation(&response
, argv
);
2238 if (r
== -ETIMEDOUT
)
2239 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
2241 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-r
));
2242 else if (response
== 's') {
2243 write_confirm_message("Skipping execution.\n");
2244 *exit_status
= EXIT_CONFIRM
;
2246 } else if (response
== 'n') {
2247 write_confirm_message("Failing execution.\n");
2253 if (context
->dynamic_user
&& dcreds
) {
2255 /* Make sure we bypass our own NSS module for any NSS checks */
2256 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2257 *exit_status
= EXIT_USER
;
2261 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2263 *exit_status
= EXIT_USER
;
2267 if (!uid_is_valid(uid
) || !gid_is_valid(gid
)) {
2268 *exit_status
= EXIT_USER
;
2273 username
= dcreds
->user
->name
;
2276 if (context
->user
) {
2277 username
= context
->user
;
2278 r
= get_user_creds_clean(&username
, &uid
, &gid
, &home
, &shell
);
2280 *exit_status
= EXIT_USER
;
2284 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
2285 * (i.e. are "/" or "/bin/nologin"). */
2288 if (context
->group
) {
2289 const char *g
= context
->group
;
2291 r
= get_group_creds(&g
, &gid
);
2293 *exit_status
= EXIT_GROUP
;
2299 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2301 *exit_status
= EXIT_USER
;
2305 user_lookup_fd
= safe_close(user_lookup_fd
);
2307 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2308 * must sure to drop O_NONBLOCK */
2310 (void) fd_nonblock(socket_fd
, false);
2312 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2314 *exit_status
= EXIT_STDIN
;
2318 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2320 *exit_status
= EXIT_STDOUT
;
2324 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2326 *exit_status
= EXIT_STDERR
;
2330 if (params
->cgroup_path
) {
2331 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2333 *exit_status
= EXIT_CGROUP
;
2338 if (context
->oom_score_adjust_set
) {
2339 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2341 /* When we can't make this change due to EPERM, then
2342 * let's silently skip over it. User namespaces
2343 * prohibit write access to this file, and we
2344 * shouldn't trip up over that. */
2346 sprintf(t
, "%i", context
->oom_score_adjust
);
2347 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2348 if (r
== -EPERM
|| r
== -EACCES
) {
2350 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2353 *exit_status
= EXIT_OOM_ADJUST
;
2358 if (context
->nice_set
)
2359 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2360 *exit_status
= EXIT_NICE
;
2364 if (context
->cpu_sched_set
) {
2365 struct sched_param param
= {
2366 .sched_priority
= context
->cpu_sched_priority
,
2369 r
= sched_setscheduler(0,
2370 context
->cpu_sched_policy
|
2371 (context
->cpu_sched_reset_on_fork
?
2372 SCHED_RESET_ON_FORK
: 0),
2375 *exit_status
= EXIT_SETSCHEDULER
;
2380 if (context
->cpuset
)
2381 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2382 *exit_status
= EXIT_CPUAFFINITY
;
2386 if (context
->ioprio_set
)
2387 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2388 *exit_status
= EXIT_IOPRIO
;
2392 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2393 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2394 *exit_status
= EXIT_TIMERSLACK
;
2398 if (context
->personality
!= PERSONALITY_INVALID
)
2399 if (personality(context
->personality
) < 0) {
2400 *exit_status
= EXIT_PERSONALITY
;
2404 if (context
->utmp_id
)
2405 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0), context
->tty_path
,
2406 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2407 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2409 username
? "root" : context
->user
);
2411 if (context
->user
) {
2412 r
= chown_terminal(STDIN_FILENO
, uid
);
2414 *exit_status
= EXIT_STDIN
;
2419 /* If delegation is enabled we'll pass ownership of the cgroup
2420 * (but only in systemd's own controller hierarchy!) to the
2421 * user of the new process. */
2422 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2423 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2425 *exit_status
= EXIT_CGROUP
;
2430 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2432 *exit_status
= EXIT_CGROUP
;
2437 if (!strv_isempty(context
->runtime_directory
) && params
->runtime_prefix
) {
2438 r
= setup_runtime_directory(context
, params
, uid
, gid
);
2440 *exit_status
= EXIT_RUNTIME_DIRECTORY
;
2445 r
= build_environment(
2457 *exit_status
= EXIT_MEMORY
;
2461 r
= build_pass_environment(context
, &pass_env
);
2463 *exit_status
= EXIT_MEMORY
;
2467 accum_env
= strv_env_merge(5,
2468 params
->environment
,
2471 context
->environment
,
2475 *exit_status
= EXIT_MEMORY
;
2478 accum_env
= strv_env_clean(accum_env
);
2480 (void) umask(context
->umask
);
2482 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2483 r
= setup_smack(context
, command
);
2485 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2489 if (context
->pam_name
&& username
) {
2490 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2492 *exit_status
= EXIT_PAM
;
2498 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2499 r
= setup_netns(runtime
->netns_storage_socket
);
2501 *exit_status
= EXIT_NETWORK
;
2506 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2507 if (needs_mount_namespace
) {
2508 _cleanup_free_
char **rw
= NULL
;
2509 char *tmp
= NULL
, *var
= NULL
;
2510 NameSpaceInfo ns_info
= {
2511 .private_dev
= context
->private_devices
,
2512 .protect_control_groups
= context
->protect_control_groups
,
2513 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2514 .protect_kernel_modules
= context
->protect_kernel_modules
,
2517 /* The runtime struct only contains the parent
2518 * of the private /tmp, which is
2519 * non-accessible to world users. Inside of it
2520 * there's a /tmp that is sticky, and that's
2521 * the one we want to use here. */
2523 if (context
->private_tmp
&& runtime
) {
2524 if (runtime
->tmp_dir
)
2525 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2526 if (runtime
->var_tmp_dir
)
2527 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2530 r
= compile_read_write_paths(context
, params
, &rw
);
2532 *exit_status
= EXIT_NAMESPACE
;
2536 r
= setup_namespace(
2537 (params
->flags
& EXEC_APPLY_CHROOT
) ? context
->root_directory
: NULL
,
2540 context
->read_only_paths
,
2541 context
->inaccessible_paths
,
2544 context
->protect_home
,
2545 context
->protect_system
,
2546 context
->mount_flags
);
2548 /* If we couldn't set up the namespace this is
2549 * probably due to a missing capability. In this case,
2550 * silently proceeed. */
2551 if (r
== -EPERM
|| r
== -EACCES
) {
2553 log_unit_debug_errno(unit
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2556 *exit_status
= EXIT_NAMESPACE
;
2561 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2562 r
= enforce_groups(context
, username
, gid
);
2564 *exit_status
= EXIT_GROUP
;
2569 if (context
->working_directory_home
)
2571 else if (context
->working_directory
)
2572 wd
= context
->working_directory
;
2576 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2577 if (!needs_mount_namespace
&& context
->root_directory
)
2578 if (chroot(context
->root_directory
) < 0) {
2579 *exit_status
= EXIT_CHROOT
;
2583 if (chdir(wd
) < 0 &&
2584 !context
->working_directory_missing_ok
) {
2585 *exit_status
= EXIT_CHDIR
;
2591 d
= strjoina(strempty(context
->root_directory
), "/", strempty(wd
));
2593 !context
->working_directory_missing_ok
) {
2594 *exit_status
= EXIT_CHDIR
;
2600 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) &&
2601 mac_selinux_use() &&
2602 params
->selinux_context_net
&&
2604 !command
->privileged
) {
2606 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2608 *exit_status
= EXIT_SELINUX_CONTEXT
;
2614 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2615 r
= setup_private_users(uid
, gid
);
2617 *exit_status
= EXIT_USER
;
2622 /* We repeat the fd closing here, to make sure that
2623 * nothing is leaked from the PAM modules. Note that
2624 * we are more aggressive this time since socket_fd
2625 * and the netns fds we don't need anymore. The custom
2626 * endpoint fd was needed to upload the policy and can
2627 * now be closed as well. */
2628 r
= close_all_fds(fds
, n_fds
);
2630 r
= shift_fds(fds
, n_fds
);
2632 r
= flags_fds(fds
, n_fds
, context
->non_blocking
);
2634 *exit_status
= EXIT_FDS
;
2638 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2640 int secure_bits
= context
->secure_bits
;
2642 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2644 if (!context
->rlimit
[i
])
2647 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2649 *exit_status
= EXIT_LIMITS
;
2654 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2655 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2656 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2657 *exit_status
= EXIT_LIMITS
;
2662 if (!cap_test_all(context
->capability_bounding_set
)) {
2663 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2665 *exit_status
= EXIT_CAPABILITIES
;
2670 /* This is done before enforce_user, but ambient set
2671 * does not survive over setresuid() if keep_caps is not set. */
2672 if (context
->capability_ambient_set
!= 0) {
2673 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2675 *exit_status
= EXIT_CAPABILITIES
;
2680 if (context
->user
) {
2681 r
= enforce_user(context
, uid
);
2683 *exit_status
= EXIT_USER
;
2686 if (context
->capability_ambient_set
!= 0) {
2688 /* Fix the ambient capabilities after user change. */
2689 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2691 *exit_status
= EXIT_CAPABILITIES
;
2695 /* If we were asked to change user and ambient capabilities
2696 * were requested, we had to add keep-caps to the securebits
2697 * so that we would maintain the inherited capability set
2698 * through the setresuid(). Make sure that the bit is added
2699 * also to the context secure_bits so that we don't try to
2700 * drop the bit away next. */
2702 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2706 /* PR_GET_SECUREBITS is not privileged, while
2707 * PR_SET_SECUREBITS is. So to suppress
2708 * potential EPERMs we'll try not to call
2709 * PR_SET_SECUREBITS unless necessary. */
2710 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2711 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2712 *exit_status
= EXIT_SECUREBITS
;
2716 if (context_has_no_new_privileges(context
))
2717 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2718 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2723 if (context_has_address_families(context
)) {
2724 r
= apply_address_families(unit
, context
);
2726 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2731 if (context
->memory_deny_write_execute
) {
2732 r
= apply_memory_deny_write_execute(unit
, context
);
2734 *exit_status
= EXIT_SECCOMP
;
2739 if (context
->restrict_realtime
) {
2740 r
= apply_restrict_realtime(unit
, context
);
2742 *exit_status
= EXIT_SECCOMP
;
2747 if (context
->protect_kernel_tunables
) {
2748 r
= apply_protect_sysctl(unit
, context
);
2750 *exit_status
= EXIT_SECCOMP
;
2755 if (context
->protect_kernel_modules
) {
2756 r
= apply_protect_kernel_modules(unit
, context
);
2758 *exit_status
= EXIT_SECCOMP
;
2763 if (context
->private_devices
) {
2764 r
= apply_private_devices(unit
, context
);
2766 *exit_status
= EXIT_SECCOMP
;
2771 if (context_has_syscall_filters(context
)) {
2772 r
= apply_seccomp(unit
, context
);
2774 *exit_status
= EXIT_SECCOMP
;
2781 if (mac_selinux_use()) {
2782 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2785 r
= setexeccon(exec_context
);
2787 *exit_status
= EXIT_SELINUX_CONTEXT
;
2794 #ifdef HAVE_APPARMOR
2795 if (context
->apparmor_profile
&& mac_apparmor_use()) {
2796 r
= aa_change_onexec(context
->apparmor_profile
);
2797 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2798 *exit_status
= EXIT_APPARMOR_PROFILE
;
2805 final_argv
= replace_env_argv(argv
, accum_env
);
2807 *exit_status
= EXIT_MEMORY
;
2811 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2812 _cleanup_free_
char *line
;
2814 line
= exec_command_line(final_argv
);
2817 log_struct(LOG_DEBUG
,
2819 "EXECUTABLE=%s", command
->path
,
2820 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2826 execve(command
->path
, final_argv
, accum_env
);
2827 *exit_status
= EXIT_EXEC
;
2831 int exec_spawn(Unit
*unit
,
2832 ExecCommand
*command
,
2833 const ExecContext
*context
,
2834 const ExecParameters
*params
,
2835 ExecRuntime
*runtime
,
2836 DynamicCreds
*dcreds
,
2839 _cleanup_strv_free_
char **files_env
= NULL
;
2840 int *fds
= NULL
; unsigned n_fds
= 0;
2841 _cleanup_free_
char *line
= NULL
;
2843 int named_iofds
[3] = { -1, -1, -1 };
2852 assert(params
->fds
|| params
->n_fds
<= 0);
2854 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2855 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2856 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
2858 if (params
->n_fds
!= 1) {
2859 log_unit_error(unit
, "Got more than one socket.");
2863 socket_fd
= params
->fds
[0];
2867 n_fds
= params
->n_fds
;
2870 r
= exec_context_named_iofds(unit
, context
, params
, named_iofds
);
2872 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
2874 r
= exec_context_load_environment(unit
, context
, &files_env
);
2876 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
2878 argv
= params
->argv
?: command
->argv
;
2879 line
= exec_command_line(argv
);
2883 log_struct(LOG_DEBUG
,
2885 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
2886 "EXECUTABLE=%s", command
->path
,
2890 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
2895 r
= exec_child(unit
,
2906 unit
->manager
->user_lookup_fds
[1],
2910 log_struct_errno(LOG_ERR
, r
,
2911 LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED
),
2913 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
2914 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
2916 "EXECUTABLE=%s", command
->path
,
2923 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
2925 /* We add the new process to the cgroup both in the child (so
2926 * that we can be sure that no user code is ever executed
2927 * outside of the cgroup) and in the parent (so that we can be
2928 * sure that when we kill the cgroup the process will be
2930 if (params
->cgroup_path
)
2931 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
2933 exec_status_start(&command
->exec_status
, pid
);
2939 void exec_context_init(ExecContext
*c
) {
2943 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
2944 c
->cpu_sched_policy
= SCHED_OTHER
;
2945 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
2946 c
->syslog_level_prefix
= true;
2947 c
->ignore_sigpipe
= true;
2948 c
->timer_slack_nsec
= NSEC_INFINITY
;
2949 c
->personality
= PERSONALITY_INVALID
;
2950 c
->runtime_directory_mode
= 0755;
2951 c
->capability_bounding_set
= CAP_ALL
;
2954 void exec_context_done(ExecContext
*c
) {
2959 c
->environment
= strv_free(c
->environment
);
2960 c
->environment_files
= strv_free(c
->environment_files
);
2961 c
->pass_environment
= strv_free(c
->pass_environment
);
2963 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
2964 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
2966 for (l
= 0; l
< 3; l
++)
2967 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
2969 c
->working_directory
= mfree(c
->working_directory
);
2970 c
->root_directory
= mfree(c
->root_directory
);
2971 c
->tty_path
= mfree(c
->tty_path
);
2972 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
2973 c
->user
= mfree(c
->user
);
2974 c
->group
= mfree(c
->group
);
2976 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
2978 c
->pam_name
= mfree(c
->pam_name
);
2980 c
->read_only_paths
= strv_free(c
->read_only_paths
);
2981 c
->read_write_paths
= strv_free(c
->read_write_paths
);
2982 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
2985 CPU_FREE(c
->cpuset
);
2987 c
->utmp_id
= mfree(c
->utmp_id
);
2988 c
->selinux_context
= mfree(c
->selinux_context
);
2989 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
2991 c
->syscall_filter
= set_free(c
->syscall_filter
);
2992 c
->syscall_archs
= set_free(c
->syscall_archs
);
2993 c
->address_families
= set_free(c
->address_families
);
2995 c
->runtime_directory
= strv_free(c
->runtime_directory
);
2998 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
3003 if (!runtime_prefix
)
3006 STRV_FOREACH(i
, c
->runtime_directory
) {
3007 _cleanup_free_
char *p
;
3009 p
= strjoin(runtime_prefix
, "/", *i
, NULL
);
3013 /* We execute this synchronously, since we need to be
3014 * sure this is gone when we start the service
3016 (void) rm_rf(p
, REMOVE_ROOT
);
3022 void exec_command_done(ExecCommand
*c
) {
3025 c
->path
= mfree(c
->path
);
3027 c
->argv
= strv_free(c
->argv
);
3030 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3033 for (i
= 0; i
< n
; i
++)
3034 exec_command_done(c
+i
);
3037 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3041 LIST_REMOVE(command
, c
, i
);
3042 exec_command_done(i
);
3049 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3052 for (i
= 0; i
< n
; i
++)
3053 c
[i
] = exec_command_free_list(c
[i
]);
3056 typedef struct InvalidEnvInfo
{
3061 static void invalid_env(const char *p
, void *userdata
) {
3062 InvalidEnvInfo
*info
= userdata
;
3064 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3067 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3072 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3074 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3076 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3078 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3080 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3082 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3088 int exec_context_named_iofds(Unit
*unit
, const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3089 unsigned i
, targets
;
3090 const char *stdio_fdname
[3];
3095 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3096 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3097 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3099 for (i
= 0; i
< 3; i
++)
3100 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3102 for (i
= 0; i
< p
->n_fds
&& targets
> 0; i
++)
3103 if (named_iofds
[STDIN_FILENO
] < 0 && c
->std_input
== EXEC_INPUT_NAMED_FD
&& stdio_fdname
[STDIN_FILENO
] && streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3104 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3106 } else if (named_iofds
[STDOUT_FILENO
] < 0 && c
->std_output
== EXEC_OUTPUT_NAMED_FD
&& stdio_fdname
[STDOUT_FILENO
] && streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3107 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3109 } else if (named_iofds
[STDERR_FILENO
] < 0 && c
->std_error
== EXEC_OUTPUT_NAMED_FD
&& stdio_fdname
[STDERR_FILENO
] && streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3110 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3114 return (targets
== 0 ? 0 : -ENOENT
);
3117 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
3118 char **i
, **r
= NULL
;
3123 STRV_FOREACH(i
, c
->environment_files
) {
3126 bool ignore
= false;
3128 _cleanup_globfree_ glob_t pglob
= {};
3138 if (!path_is_absolute(fn
)) {
3146 /* Filename supports globbing, take all matching files */
3148 if (glob(fn
, 0, NULL
, &pglob
) != 0) {
3153 return errno
> 0 ? -errno
: -EINVAL
;
3155 count
= pglob
.gl_pathc
;
3163 for (n
= 0; n
< count
; n
++) {
3164 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3172 /* Log invalid environment variables with filename */
3174 InvalidEnvInfo info
= {
3176 .path
= pglob
.gl_pathv
[n
]
3179 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3187 m
= strv_env_merge(2, r
, p
);
3203 static bool tty_may_match_dev_console(const char *tty
) {
3204 _cleanup_free_
char *active
= NULL
;
3210 if (startswith(tty
, "/dev/"))
3213 /* trivial identity? */
3214 if (streq(tty
, "console"))
3217 console
= resolve_dev_console(&active
);
3218 /* if we could not resolve, assume it may */
3222 /* "tty0" means the active VC, so it may be the same sometimes */
3223 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3226 bool exec_context_may_touch_console(ExecContext
*ec
) {
3228 return (ec
->tty_reset
||
3230 ec
->tty_vt_disallocate
||
3231 is_terminal_input(ec
->std_input
) ||
3232 is_terminal_output(ec
->std_output
) ||
3233 is_terminal_output(ec
->std_error
)) &&
3234 tty_may_match_dev_console(exec_context_tty_path(ec
));
3237 static void strv_fprintf(FILE *f
, char **l
) {
3243 fprintf(f
, " %s", *g
);
3246 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3253 prefix
= strempty(prefix
);
3257 "%sWorkingDirectory: %s\n"
3258 "%sRootDirectory: %s\n"
3259 "%sNonBlocking: %s\n"
3260 "%sPrivateTmp: %s\n"
3261 "%sPrivateDevices: %s\n"
3262 "%sProtectKernelTunables: %s\n"
3263 "%sProtectKernelModules: %s\n"
3264 "%sProtectControlGroups: %s\n"
3265 "%sPrivateNetwork: %s\n"
3266 "%sPrivateUsers: %s\n"
3267 "%sProtectHome: %s\n"
3268 "%sProtectSystem: %s\n"
3269 "%sIgnoreSIGPIPE: %s\n"
3270 "%sMemoryDenyWriteExecute: %s\n"
3271 "%sRestrictRealtime: %s\n",
3273 prefix
, c
->working_directory
? c
->working_directory
: "/",
3274 prefix
, c
->root_directory
? c
->root_directory
: "/",
3275 prefix
, yes_no(c
->non_blocking
),
3276 prefix
, yes_no(c
->private_tmp
),
3277 prefix
, yes_no(c
->private_devices
),
3278 prefix
, yes_no(c
->protect_kernel_tunables
),
3279 prefix
, yes_no(c
->protect_kernel_modules
),
3280 prefix
, yes_no(c
->protect_control_groups
),
3281 prefix
, yes_no(c
->private_network
),
3282 prefix
, yes_no(c
->private_users
),
3283 prefix
, protect_home_to_string(c
->protect_home
),
3284 prefix
, protect_system_to_string(c
->protect_system
),
3285 prefix
, yes_no(c
->ignore_sigpipe
),
3286 prefix
, yes_no(c
->memory_deny_write_execute
),
3287 prefix
, yes_no(c
->restrict_realtime
));
3289 STRV_FOREACH(e
, c
->environment
)
3290 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3292 STRV_FOREACH(e
, c
->environment_files
)
3293 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3295 STRV_FOREACH(e
, c
->pass_environment
)
3296 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3298 fprintf(f
, "%sRuntimeDirectoryMode: %04o\n", prefix
, c
->runtime_directory_mode
);
3300 STRV_FOREACH(d
, c
->runtime_directory
)
3301 fprintf(f
, "%sRuntimeDirectory: %s\n", prefix
, *d
);
3308 if (c
->oom_score_adjust_set
)
3310 "%sOOMScoreAdjust: %i\n",
3311 prefix
, c
->oom_score_adjust
);
3313 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3315 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3316 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3317 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3318 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3321 if (c
->ioprio_set
) {
3322 _cleanup_free_
char *class_str
= NULL
;
3324 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3326 "%sIOSchedulingClass: %s\n"
3327 "%sIOPriority: %i\n",
3328 prefix
, strna(class_str
),
3329 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
3332 if (c
->cpu_sched_set
) {
3333 _cleanup_free_
char *policy_str
= NULL
;
3335 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3337 "%sCPUSchedulingPolicy: %s\n"
3338 "%sCPUSchedulingPriority: %i\n"
3339 "%sCPUSchedulingResetOnFork: %s\n",
3340 prefix
, strna(policy_str
),
3341 prefix
, c
->cpu_sched_priority
,
3342 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3346 fprintf(f
, "%sCPUAffinity:", prefix
);
3347 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3348 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3349 fprintf(f
, " %u", i
);
3353 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3354 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3357 "%sStandardInput: %s\n"
3358 "%sStandardOutput: %s\n"
3359 "%sStandardError: %s\n",
3360 prefix
, exec_input_to_string(c
->std_input
),
3361 prefix
, exec_output_to_string(c
->std_output
),
3362 prefix
, exec_output_to_string(c
->std_error
));
3368 "%sTTYVHangup: %s\n"
3369 "%sTTYVTDisallocate: %s\n",
3370 prefix
, c
->tty_path
,
3371 prefix
, yes_no(c
->tty_reset
),
3372 prefix
, yes_no(c
->tty_vhangup
),
3373 prefix
, yes_no(c
->tty_vt_disallocate
));
3375 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3376 c
->std_output
== EXEC_OUTPUT_KMSG
||
3377 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3378 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3379 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3380 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3381 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3382 c
->std_error
== EXEC_OUTPUT_KMSG
||
3383 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3384 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3385 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3386 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3388 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3390 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3391 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3394 "%sSyslogFacility: %s\n"
3395 "%sSyslogLevel: %s\n",
3396 prefix
, strna(fac_str
),
3397 prefix
, strna(lvl_str
));
3401 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
3403 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
3404 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
3405 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
3406 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
3407 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
3408 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
3410 if (c
->capability_bounding_set
!= CAP_ALL
) {
3412 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
3414 for (l
= 0; l
<= cap_last_cap(); l
++)
3415 if (c
->capability_bounding_set
& (UINT64_C(1) << l
))
3416 fprintf(f
, " %s", strna(capability_to_name(l
)));
3421 if (c
->capability_ambient_set
!= 0) {
3423 fprintf(f
, "%sAmbientCapabilities:", prefix
);
3425 for (l
= 0; l
<= cap_last_cap(); l
++)
3426 if (c
->capability_ambient_set
& (UINT64_C(1) << l
))
3427 fprintf(f
, " %s", strna(capability_to_name(l
)));
3433 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3435 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3437 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3439 if (strv_length(c
->supplementary_groups
) > 0) {
3440 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3441 strv_fprintf(f
, c
->supplementary_groups
);
3446 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3448 if (strv_length(c
->read_write_paths
) > 0) {
3449 fprintf(f
, "%sReadWritePaths:", prefix
);
3450 strv_fprintf(f
, c
->read_write_paths
);
3454 if (strv_length(c
->read_only_paths
) > 0) {
3455 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3456 strv_fprintf(f
, c
->read_only_paths
);
3460 if (strv_length(c
->inaccessible_paths
) > 0) {
3461 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3462 strv_fprintf(f
, c
->inaccessible_paths
);
3468 "%sUtmpIdentifier: %s\n",
3469 prefix
, c
->utmp_id
);
3471 if (c
->selinux_context
)
3473 "%sSELinuxContext: %s%s\n",
3474 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3476 if (c
->personality
!= PERSONALITY_INVALID
)
3478 "%sPersonality: %s\n",
3479 prefix
, strna(personality_to_string(c
->personality
)));
3481 if (c
->syscall_filter
) {
3489 "%sSystemCallFilter: ",
3492 if (!c
->syscall_whitelist
)
3496 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3497 _cleanup_free_
char *name
= NULL
;
3504 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3505 fputs(strna(name
), f
);
3512 if (c
->syscall_archs
) {
3519 "%sSystemCallArchitectures:",
3523 SET_FOREACH(id
, c
->syscall_archs
, j
)
3524 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3529 if (c
->syscall_errno
> 0)
3531 "%sSystemCallErrorNumber: %s\n",
3532 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3534 if (c
->apparmor_profile
)
3536 "%sAppArmorProfile: %s%s\n",
3537 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3540 bool exec_context_maintains_privileges(ExecContext
*c
) {
3543 /* Returns true if the process forked off would run under
3544 * an unchanged UID or as root. */
3549 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3555 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3560 dual_timestamp_get(&s
->start_timestamp
);
3563 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3566 if (s
->pid
&& s
->pid
!= pid
)
3570 dual_timestamp_get(&s
->exit_timestamp
);
3576 if (context
->utmp_id
)
3577 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3579 exec_context_tty_reset(context
, NULL
);
3583 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3584 char buf
[FORMAT_TIMESTAMP_MAX
];
3592 prefix
= strempty(prefix
);
3595 "%sPID: "PID_FMT
"\n",
3598 if (dual_timestamp_is_set(&s
->start_timestamp
))
3600 "%sStart Timestamp: %s\n",
3601 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3603 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3605 "%sExit Timestamp: %s\n"
3607 "%sExit Status: %i\n",
3608 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3609 prefix
, sigchld_code_to_string(s
->code
),
3613 char *exec_command_line(char **argv
) {
3621 STRV_FOREACH(a
, argv
)
3624 if (!(n
= new(char, k
)))
3628 STRV_FOREACH(a
, argv
) {
3635 if (strpbrk(*a
, WHITESPACE
)) {
3646 /* FIXME: this doesn't really handle arguments that have
3647 * spaces and ticks in them */
3652 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3653 _cleanup_free_
char *cmd
= NULL
;
3654 const char *prefix2
;
3659 prefix
= strempty(prefix
);
3660 prefix2
= strjoina(prefix
, "\t");
3662 cmd
= exec_command_line(c
->argv
);
3664 "%sCommand Line: %s\n",
3665 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3667 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3670 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3673 prefix
= strempty(prefix
);
3675 LIST_FOREACH(command
, c
, c
)
3676 exec_command_dump(c
, f
, prefix
);
3679 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3686 /* It's kind of important, that we keep the order here */
3687 LIST_FIND_TAIL(command
, *l
, end
);
3688 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3693 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3701 l
= strv_new_ap(path
, ap
);
3722 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3723 _cleanup_strv_free_
char **l
= NULL
;
3731 l
= strv_new_ap(path
, ap
);
3737 r
= strv_extend_strv(&c
->argv
, l
, false);
3745 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3750 *rt
= new0(ExecRuntime
, 1);
3755 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3760 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
3770 if (!c
->private_network
&& !c
->private_tmp
)
3773 r
= exec_runtime_allocate(rt
);
3777 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
3778 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
3782 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
3783 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
3791 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
3793 assert(r
->n_ref
> 0);
3799 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
3804 assert(r
->n_ref
> 0);
3811 free(r
->var_tmp_dir
);
3812 safe_close_pair(r
->netns_storage_socket
);
3816 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
3825 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
3827 if (rt
->var_tmp_dir
)
3828 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
3830 if (rt
->netns_storage_socket
[0] >= 0) {
3833 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
3837 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
3840 if (rt
->netns_storage_socket
[1] >= 0) {
3843 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
3847 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
3853 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
3860 if (streq(key
, "tmp-dir")) {
3863 r
= exec_runtime_allocate(rt
);
3867 copy
= strdup(value
);
3871 free((*rt
)->tmp_dir
);
3872 (*rt
)->tmp_dir
= copy
;
3874 } else if (streq(key
, "var-tmp-dir")) {
3877 r
= exec_runtime_allocate(rt
);
3881 copy
= strdup(value
);
3885 free((*rt
)->var_tmp_dir
);
3886 (*rt
)->var_tmp_dir
= copy
;
3888 } else if (streq(key
, "netns-socket-0")) {
3891 r
= exec_runtime_allocate(rt
);
3895 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
3896 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
3898 safe_close((*rt
)->netns_storage_socket
[0]);
3899 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
3901 } else if (streq(key
, "netns-socket-1")) {
3904 r
= exec_runtime_allocate(rt
);
3908 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
3909 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
3911 safe_close((*rt
)->netns_storage_socket
[1]);
3912 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
3920 static void *remove_tmpdir_thread(void *p
) {
3921 _cleanup_free_
char *path
= p
;
3923 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
3927 void exec_runtime_destroy(ExecRuntime
*rt
) {
3933 /* If there are multiple users of this, let's leave the stuff around */
3938 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
3940 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
3942 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
3949 if (rt
->var_tmp_dir
) {
3950 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
3952 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
3954 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
3955 free(rt
->var_tmp_dir
);
3958 rt
->var_tmp_dir
= NULL
;
3961 safe_close_pair(rt
->netns_storage_socket
);
3964 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
3965 [EXEC_INPUT_NULL
] = "null",
3966 [EXEC_INPUT_TTY
] = "tty",
3967 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
3968 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
3969 [EXEC_INPUT_SOCKET
] = "socket",
3970 [EXEC_INPUT_NAMED_FD
] = "fd",
3973 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
3975 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
3976 [EXEC_OUTPUT_INHERIT
] = "inherit",
3977 [EXEC_OUTPUT_NULL
] = "null",
3978 [EXEC_OUTPUT_TTY
] = "tty",
3979 [EXEC_OUTPUT_SYSLOG
] = "syslog",
3980 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
3981 [EXEC_OUTPUT_KMSG
] = "kmsg",
3982 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
3983 [EXEC_OUTPUT_JOURNAL
] = "journal",
3984 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
3985 [EXEC_OUTPUT_SOCKET
] = "socket",
3986 [EXEC_OUTPUT_NAMED_FD
] = "fd",
3989 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
3991 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
3992 [EXEC_UTMP_INIT
] = "init",
3993 [EXEC_UTMP_LOGIN
] = "login",
3994 [EXEC_UTMP_USER
] = "user",
3997 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);