1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <sys/eventfd.h>
10 #include <sys/personality.h>
11 #include <sys/prctl.h>
13 #include <sys/types.h>
19 #include <security/pam_appl.h>
23 #include <selinux/selinux.h>
31 #include <sys/apparmor.h>
34 #include "sd-messages.h"
38 #include "alloc-util.h"
40 #include "apparmor-util.h"
42 #include "argv-util.h"
47 #include "capability-util.h"
48 #include "cgroup-setup.h"
50 #include "chown-recursive.h"
51 #include "constants.h"
52 #include "cpu-set-util.h"
53 #include "creds-util.h"
54 #include "data-fd-util.h"
57 #include "errno-list.h"
60 #include "exit-status.h"
63 #include "format-util.h"
64 #include "glob-util.h"
65 #include "hexdecoct.h"
67 #include "ioprio-util.h"
68 #include "label-util.h"
72 #include "manager-dump.h"
73 #include "memory-util.h"
74 #include "missing_fs.h"
75 #include "missing_ioprio.h"
76 #include "missing_prctl.h"
77 #include "mkdir-label.h"
78 #include "mount-util.h"
79 #include "mountpoint-util.h"
80 #include "namespace.h"
81 #include "parse-util.h"
82 #include "path-util.h"
83 #include "proc-cmdline.h"
84 #include "process-util.h"
86 #include "random-util.h"
87 #include "recurse-dir.h"
88 #include "rlimit-util.h"
91 #include "seccomp-util.h"
93 #include "securebits-util.h"
94 #include "selinux-util.h"
95 #include "signal-util.h"
96 #include "smack-util.h"
97 #include "socket-util.h"
98 #include "sort-util.h"
100 #include "stat-util.h"
101 #include "string-table.h"
102 #include "string-util.h"
104 #include "syslog-util.h"
105 #include "terminal-util.h"
106 #include "tmpfile-util.h"
107 #include "umask-util.h"
108 #include "unit-serialize.h"
109 #include "user-util.h"
110 #include "utmp-wtmp.h"
112 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
113 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
115 #define SNDBUF_SIZE (8*1024*1024)
117 static int shift_fds(int fds
[], size_t n_fds
) {
121 /* Modifies the fds array! (sorts it) */
125 for (int start
= 0;;) {
126 int restart_from
= -1;
128 for (int i
= start
; i
< (int) n_fds
; i
++) {
131 /* Already at right index? */
135 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
142 /* Hmm, the fd we wanted isn't free? Then
143 * let's remember that and try again from here */
144 if (nfd
!= i
+3 && restart_from
< 0)
148 if (restart_from
< 0)
151 start
= restart_from
;
157 static int flags_fds(
170 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
171 * O_NONBLOCK only applies to socket activation though. */
173 for (size_t i
= 0; i
< n_fds
; i
++) {
175 if (i
< n_socket_fds
) {
176 r
= fd_nonblock(fds
[i
], nonblock
);
181 /* We unconditionally drop FD_CLOEXEC from the fds,
182 * since after all we want to pass these fds to our
185 r
= fd_cloexec(fds
[i
], false);
193 static const char *exec_context_tty_path(const ExecContext
*context
) {
196 if (context
->stdio_as_fds
)
199 if (context
->tty_path
)
200 return context
->tty_path
;
202 return "/dev/console";
205 static int exec_context_tty_size(const ExecContext
*context
, unsigned *ret_rows
, unsigned *ret_cols
) {
206 _cleanup_free_
char *rowskey
= NULL
, *rowsvalue
= NULL
, *colskey
= NULL
, *colsvalue
= NULL
;
215 rows
= context
->tty_rows
;
216 cols
= context
->tty_cols
;
218 tty
= exec_context_tty_path(context
);
219 if (!tty
|| (rows
!= UINT_MAX
&& cols
!= UINT_MAX
)) {
225 tty
= skip_dev_prefix(tty
);
226 if (!in_charset(tty
, ALPHANUMERICAL
)) {
227 log_debug("%s contains non-alphanumeric characters, ignoring", tty
);
233 rowskey
= strjoin("systemd.tty.rows.", tty
);
237 colskey
= strjoin("systemd.tty.columns.", tty
);
241 r
= proc_cmdline_get_key_many(/* flags = */ 0,
243 colskey
, &colsvalue
);
245 log_debug_errno(r
, "Failed to read TTY size of %s from kernel cmdline, ignoring: %m", tty
);
247 if (rows
== UINT_MAX
&& rowsvalue
) {
248 r
= safe_atou(rowsvalue
, &rows
);
250 log_debug_errno(r
, "Failed to parse %s=%s, ignoring: %m", rowskey
, rowsvalue
);
253 if (cols
== UINT_MAX
&& colsvalue
) {
254 r
= safe_atou(colsvalue
, &cols
);
256 log_debug_errno(r
, "Failed to parse %s=%s, ignoring: %m", colskey
, colsvalue
);
265 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
270 path
= exec_context_tty_path(context
);
272 if (context
->tty_vhangup
) {
273 if (p
&& p
->stdin_fd
>= 0)
274 (void) terminal_vhangup_fd(p
->stdin_fd
);
276 (void) terminal_vhangup(path
);
279 if (context
->tty_reset
) {
280 if (p
&& p
->stdin_fd
>= 0)
281 (void) reset_terminal_fd(p
->stdin_fd
, true);
283 (void) reset_terminal(path
);
286 if (p
&& p
->stdin_fd
>= 0) {
287 unsigned rows
= context
->tty_rows
, cols
= context
->tty_cols
;
289 (void) exec_context_tty_size(context
, &rows
, &cols
);
290 (void) terminal_set_size_fd(p
->stdin_fd
, path
, rows
, cols
);
293 if (context
->tty_vt_disallocate
&& path
)
294 (void) vt_disallocate(path
);
297 static bool is_terminal_input(ExecInput i
) {
300 EXEC_INPUT_TTY_FORCE
,
301 EXEC_INPUT_TTY_FAIL
);
304 static bool is_terminal_output(ExecOutput o
) {
307 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
308 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
311 static bool is_kmsg_output(ExecOutput o
) {
314 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
317 static bool exec_context_needs_term(const ExecContext
*c
) {
320 /* Return true if the execution context suggests we should set $TERM to something useful. */
322 if (is_terminal_input(c
->std_input
))
325 if (is_terminal_output(c
->std_output
))
328 if (is_terminal_output(c
->std_error
))
331 return !!c
->tty_path
;
334 static int open_null_as(int flags
, int nfd
) {
339 fd
= open("/dev/null", flags
|O_NOCTTY
);
343 return move_fd(fd
, nfd
, false);
346 static int connect_journal_socket(
348 const char *log_namespace
,
352 uid_t olduid
= UID_INVALID
;
353 gid_t oldgid
= GID_INVALID
;
358 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
359 "/run/systemd/journal/stdout";
361 if (gid_is_valid(gid
)) {
364 if (setegid(gid
) < 0)
368 if (uid_is_valid(uid
)) {
371 if (seteuid(uid
) < 0) {
377 r
= connect_unix_path(fd
, AT_FDCWD
, j
);
379 /* If we fail to restore the uid or gid, things will likely fail later on. This should only happen if
380 an LSM interferes. */
382 if (uid_is_valid(uid
))
383 (void) seteuid(olduid
);
386 if (gid_is_valid(gid
))
387 (void) setegid(oldgid
);
392 static int connect_logger_as(
394 const ExecContext
*context
,
395 const ExecParameters
*params
,
402 _cleanup_close_
int fd
= -EBADF
;
407 assert(output
< _EXEC_OUTPUT_MAX
);
411 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
415 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
419 if (shutdown(fd
, SHUT_RD
) < 0)
422 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
432 context
->syslog_identifier
?: ident
,
433 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
434 context
->syslog_priority
,
435 !!context
->syslog_level_prefix
,
437 is_kmsg_output(output
),
438 is_terminal_output(output
)) < 0)
441 return move_fd(TAKE_FD(fd
), nfd
, false);
444 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
450 fd
= open_terminal(path
, flags
| O_NOCTTY
);
454 return move_fd(fd
, nfd
, false);
457 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
458 _cleanup_close_
int fd
= -EBADF
;
463 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
466 fd
= open(path
, flags
|O_NOCTTY
, mode
);
470 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
473 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
475 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
479 r
= connect_unix_path(fd
, AT_FDCWD
, path
);
480 if (IN_SET(r
, -ENOTSOCK
, -EINVAL
))
481 /* Propagate initial error if we get ENOTSOCK or EINVAL, i.e. we have indication that this
482 * wasn't an AF_UNIX socket after all */
487 if ((flags
& O_ACCMODE
) == O_RDONLY
)
488 r
= shutdown(fd
, SHUT_WR
);
489 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
490 r
= shutdown(fd
, SHUT_RD
);
499 static int fixup_input(
500 const ExecContext
*context
,
502 bool apply_tty_stdin
) {
508 std_input
= context
->std_input
;
510 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
511 return EXEC_INPUT_NULL
;
513 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
514 return EXEC_INPUT_NULL
;
516 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
517 return EXEC_INPUT_NULL
;
522 static int fixup_output(ExecOutput output
, int socket_fd
) {
524 if (output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
525 return EXEC_OUTPUT_INHERIT
;
530 static int setup_input(
531 const ExecContext
*context
,
532 const ExecParameters
*params
,
534 const int named_iofds
[static 3]) {
543 if (params
->stdin_fd
>= 0) {
544 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
547 /* Try to make this the controlling tty, if it is a tty, and reset it */
548 if (isatty(STDIN_FILENO
)) {
549 unsigned rows
= context
->tty_rows
, cols
= context
->tty_cols
;
551 (void) exec_context_tty_size(context
, &rows
, &cols
);
552 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
553 (void) reset_terminal_fd(STDIN_FILENO
, true);
554 (void) terminal_set_size_fd(STDIN_FILENO
, NULL
, rows
, cols
);
560 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
564 case EXEC_INPUT_NULL
:
565 return open_null_as(O_RDONLY
, STDIN_FILENO
);
568 case EXEC_INPUT_TTY_FORCE
:
569 case EXEC_INPUT_TTY_FAIL
: {
573 fd
= acquire_terminal(exec_context_tty_path(context
),
574 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
575 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
576 ACQUIRE_TERMINAL_WAIT
,
581 r
= exec_context_tty_size(context
, &rows
, &cols
);
585 r
= terminal_set_size_fd(fd
, exec_context_tty_path(context
), rows
, cols
);
589 return move_fd(fd
, STDIN_FILENO
, false);
592 case EXEC_INPUT_SOCKET
:
593 assert(socket_fd
>= 0);
595 return RET_NERRNO(dup2(socket_fd
, STDIN_FILENO
));
597 case EXEC_INPUT_NAMED_FD
:
598 assert(named_iofds
[STDIN_FILENO
] >= 0);
600 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
601 return RET_NERRNO(dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
));
603 case EXEC_INPUT_DATA
: {
606 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
610 return move_fd(fd
, STDIN_FILENO
, false);
613 case EXEC_INPUT_FILE
: {
617 assert(context
->stdio_file
[STDIN_FILENO
]);
619 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
620 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
622 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
626 return move_fd(fd
, STDIN_FILENO
, false);
630 assert_not_reached();
634 static bool can_inherit_stderr_from_stdout(
635 const ExecContext
*context
,
641 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
644 if (e
== EXEC_OUTPUT_INHERIT
)
649 if (e
== EXEC_OUTPUT_NAMED_FD
)
650 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
652 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
, EXEC_OUTPUT_FILE_TRUNCATE
))
653 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
658 static int setup_output(
660 const ExecContext
*context
,
661 const ExecParameters
*params
,
664 const int named_iofds
[static 3],
668 dev_t
*journal_stream_dev
,
669 ino_t
*journal_stream_ino
) {
679 assert(journal_stream_dev
);
680 assert(journal_stream_ino
);
682 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
684 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
687 return STDOUT_FILENO
;
690 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
691 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
694 return STDERR_FILENO
;
697 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
698 o
= fixup_output(context
->std_output
, socket_fd
);
700 if (fileno
== STDERR_FILENO
) {
702 e
= fixup_output(context
->std_error
, socket_fd
);
704 /* This expects the input and output are already set up */
706 /* Don't change the stderr file descriptor if we inherit all
707 * the way and are not on a tty */
708 if (e
== EXEC_OUTPUT_INHERIT
&&
709 o
== EXEC_OUTPUT_INHERIT
&&
710 i
== EXEC_INPUT_NULL
&&
711 !is_terminal_input(context
->std_input
) &&
715 /* Duplicate from stdout if possible */
716 if (can_inherit_stderr_from_stdout(context
, o
, e
))
717 return RET_NERRNO(dup2(STDOUT_FILENO
, fileno
));
721 } else if (o
== EXEC_OUTPUT_INHERIT
) {
722 /* If input got downgraded, inherit the original value */
723 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
724 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
726 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
727 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
728 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
730 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
734 /* We need to open /dev/null here anew, to get the right access mode. */
735 return open_null_as(O_WRONLY
, fileno
);
740 case EXEC_OUTPUT_NULL
:
741 return open_null_as(O_WRONLY
, fileno
);
743 case EXEC_OUTPUT_TTY
:
744 if (is_terminal_input(i
))
745 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
747 /* We don't reset the terminal if this is just about output */
748 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
750 case EXEC_OUTPUT_KMSG
:
751 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
752 case EXEC_OUTPUT_JOURNAL
:
753 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
754 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
756 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m",
757 fileno
== STDOUT_FILENO
? "stdout" : "stderr");
758 r
= open_null_as(O_WRONLY
, fileno
);
762 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
763 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
764 * services to detect whether they are connected to the journal or not.
766 * If both stdout and stderr are connected to a stream then let's make sure to store the data
767 * about STDERR as that's usually the best way to do logging. */
769 if (fstat(fileno
, &st
) >= 0 &&
770 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
771 *journal_stream_dev
= st
.st_dev
;
772 *journal_stream_ino
= st
.st_ino
;
777 case EXEC_OUTPUT_SOCKET
:
778 assert(socket_fd
>= 0);
780 return RET_NERRNO(dup2(socket_fd
, fileno
));
782 case EXEC_OUTPUT_NAMED_FD
:
783 assert(named_iofds
[fileno
] >= 0);
785 (void) fd_nonblock(named_iofds
[fileno
], false);
786 return RET_NERRNO(dup2(named_iofds
[fileno
], fileno
));
788 case EXEC_OUTPUT_FILE
:
789 case EXEC_OUTPUT_FILE_APPEND
:
790 case EXEC_OUTPUT_FILE_TRUNCATE
: {
794 assert(context
->stdio_file
[fileno
]);
796 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
797 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
800 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
803 if (o
== EXEC_OUTPUT_FILE_APPEND
)
805 else if (o
== EXEC_OUTPUT_FILE_TRUNCATE
)
808 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
812 return move_fd(fd
, fileno
, 0);
816 assert_not_reached();
820 static int chown_terminal(int fd
, uid_t uid
) {
825 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
826 if (isatty(fd
) < 1) {
827 if (IN_SET(errno
, EINVAL
, ENOTTY
))
828 return 0; /* not a tty */
833 /* This might fail. What matters are the results. */
834 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, GID_INVALID
);
841 static int setup_confirm_stdio(
842 const ExecContext
*context
,
844 int *ret_saved_stdin
,
845 int *ret_saved_stdout
) {
847 _cleanup_close_
int fd
= -EBADF
, saved_stdin
= -EBADF
, saved_stdout
= -EBADF
;
851 assert(ret_saved_stdin
);
852 assert(ret_saved_stdout
);
854 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
858 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
859 if (saved_stdout
< 0)
862 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
866 r
= chown_terminal(fd
, getuid());
870 r
= reset_terminal_fd(fd
, true);
874 r
= exec_context_tty_size(context
, &rows
, &cols
);
878 r
= terminal_set_size_fd(fd
, vc
, rows
, cols
);
882 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
); /* Invalidates 'fd' also on failure */
887 *ret_saved_stdin
= TAKE_FD(saved_stdin
);
888 *ret_saved_stdout
= TAKE_FD(saved_stdout
);
892 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
895 if (err
== -ETIMEDOUT
)
896 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
899 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
903 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
904 _cleanup_close_
int fd
= -EBADF
;
908 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
912 write_confirm_error_fd(err
, fd
, u
);
915 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
919 assert(saved_stdout
);
923 if (*saved_stdin
>= 0)
924 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
927 if (*saved_stdout
>= 0)
928 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
931 *saved_stdin
= safe_close(*saved_stdin
);
932 *saved_stdout
= safe_close(*saved_stdout
);
938 CONFIRM_PRETEND_FAILURE
= -1,
939 CONFIRM_PRETEND_SUCCESS
= 0,
943 static int ask_for_confirmation(const ExecContext
*context
, const char *vc
, Unit
*u
, const char *cmdline
) {
944 int saved_stdout
= -1, saved_stdin
= -1, r
;
945 _cleanup_free_
char *e
= NULL
;
948 /* For any internal errors, assume a positive response. */
949 r
= setup_confirm_stdio(context
, vc
, &saved_stdin
, &saved_stdout
);
951 write_confirm_error(r
, vc
, u
);
952 return CONFIRM_EXECUTE
;
955 /* confirm_spawn might have been disabled while we were sleeping. */
956 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
961 e
= ellipsize(cmdline
, 60, 100);
969 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
971 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
978 printf("Resuming normal execution.\n");
979 manager_disable_confirm_spawn();
983 unit_dump(u
, stdout
, " ");
984 continue; /* ask again */
986 printf("Failing execution.\n");
987 r
= CONFIRM_PRETEND_FAILURE
;
990 printf(" c - continue, proceed without asking anymore\n"
991 " D - dump, show the state of the unit\n"
992 " f - fail, don't execute the command and pretend it failed\n"
994 " i - info, show a short summary of the unit\n"
995 " j - jobs, show jobs that are in progress\n"
996 " s - skip, don't execute the command and pretend it succeeded\n"
997 " y - yes, execute the command\n");
998 continue; /* ask again */
1000 printf(" Description: %s\n"
1003 u
->id
, u
->description
, cmdline
);
1004 continue; /* ask again */
1006 manager_dump_jobs(u
->manager
, stdout
, /* patterns= */ NULL
, " ");
1007 continue; /* ask again */
1009 /* 'n' was removed in favor of 'f'. */
1010 printf("Didn't understand 'n', did you mean 'f'?\n");
1011 continue; /* ask again */
1013 printf("Skipping execution.\n");
1014 r
= CONFIRM_PRETEND_SUCCESS
;
1017 r
= CONFIRM_EXECUTE
;
1020 assert_not_reached();
1026 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
1030 static int get_fixed_user(const ExecContext
*c
, const char **user
,
1031 uid_t
*uid
, gid_t
*gid
,
1032 const char **home
, const char **shell
) {
1041 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
1042 * (i.e. are "/" or "/bin/nologin"). */
1045 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
1053 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
1063 r
= get_group_creds(&name
, gid
, 0);
1071 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
1072 const char *group
, gid_t gid
,
1073 gid_t
**supplementary_gids
, int *ngids
) {
1076 bool keep_groups
= false;
1077 gid_t
*groups
= NULL
;
1078 _cleanup_free_ gid_t
*l_gids
= NULL
;
1083 * If user is given, then lookup GID and supplementary groups list.
1084 * We avoid NSS lookups for gid=0. Also we have to initialize groups
1085 * here and as early as possible so we keep the list of supplementary
1086 * groups of the caller.
1088 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
1089 /* First step, initialize groups from /etc/groups */
1090 if (initgroups(user
, gid
) < 0)
1096 if (strv_isempty(c
->supplementary_groups
))
1100 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1101 * be positive, otherwise fail.
1104 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1105 if (ngroups_max
<= 0)
1106 return errno_or_else(EOPNOTSUPP
);
1108 l_gids
= new(gid_t
, ngroups_max
);
1114 * Lookup the list of groups that the user belongs to, we
1115 * avoid NSS lookups here too for gid=0.
1118 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1123 STRV_FOREACH(i
, c
->supplementary_groups
) {
1126 if (k
>= ngroups_max
)
1130 r
= get_group_creds(&g
, l_gids
+k
, 0);
1138 * Sets ngids to zero to drop all supplementary groups, happens
1139 * when we are under root and SupplementaryGroups= is empty.
1146 /* Otherwise get the final list of supplementary groups */
1147 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1151 *supplementary_gids
= groups
;
1159 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1162 /* Handle SupplementaryGroups= if it is not empty */
1164 r
= maybe_setgroups(ngids
, supplementary_gids
);
1169 if (gid_is_valid(gid
)) {
1170 /* Then set our gids */
1171 if (setresgid(gid
, gid
, gid
) < 0)
1178 static int set_securebits(unsigned bits
, unsigned mask
) {
1182 current
= prctl(PR_GET_SECUREBITS
);
1186 /* Clear all securebits defined in mask and set bits */
1187 applied
= ((unsigned) current
& ~mask
) | bits
;
1188 if ((unsigned) current
== applied
)
1191 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1197 static int enforce_user(
1198 const ExecContext
*context
,
1200 uint64_t capability_ambient_set
) {
1204 if (!uid_is_valid(uid
))
1207 /* Sets (but doesn't look up) the UIS and makes sure we keep the capabilities while doing so. For
1208 * setting secure bits the capability CAP_SETPCAP is required, so we also need keep-caps in this
1211 if ((capability_ambient_set
!= 0 || context
->secure_bits
!= 0) && uid
!= 0) {
1213 /* First step: If we need to keep capabilities but drop privileges we need to make sure we
1214 * keep our caps, while we drop privileges. Add KEEP_CAPS to the securebits */
1215 r
= set_securebits(1U << SECURE_KEEP_CAPS
, 0);
1220 /* Second step: actually set the uids */
1221 if (setresuid(uid
, uid
, uid
) < 0)
1224 /* At this point we should have all necessary capabilities but are otherwise a normal user. However,
1225 * the caps might got corrupted due to the setresuid() so we need clean them up later. This is done
1226 * outside of this call. */
1232 static int null_conv(
1234 const struct pam_message
**msg
,
1235 struct pam_response
**resp
,
1236 void *appdata_ptr
) {
1238 /* We don't support conversations */
1240 return PAM_CONV_ERR
;
1245 static int setup_pam(
1251 char ***env
, /* updated on success */
1252 const int fds
[], size_t n_fds
) {
1256 static const struct pam_conv conv
= {
1261 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1262 _cleanup_strv_free_
char **e
= NULL
;
1263 pam_handle_t
*handle
= NULL
;
1265 int pam_code
= PAM_SUCCESS
, r
;
1266 bool close_session
= false;
1267 pid_t pam_pid
= 0, parent_pid
;
1274 /* We set up PAM in the parent process, then fork. The child
1275 * will then stay around until killed via PR_GET_PDEATHSIG or
1276 * systemd via the cgroup logic. It will then remove the PAM
1277 * session again. The parent process will exec() the actual
1278 * daemon. We do things this way to ensure that the main PID
1279 * of the daemon is the one we initially fork()ed. */
1281 r
= barrier_create(&barrier
);
1285 if (log_get_max_level() < LOG_DEBUG
)
1286 flags
|= PAM_SILENT
;
1288 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1289 if (pam_code
!= PAM_SUCCESS
) {
1295 _cleanup_free_
char *q
= NULL
;
1297 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1298 * out if that's the case, and read the TTY off it. */
1300 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1301 tty
= strjoina("/dev/", q
);
1305 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1306 if (pam_code
!= PAM_SUCCESS
)
1310 STRV_FOREACH(nv
, *env
) {
1311 pam_code
= pam_putenv(handle
, *nv
);
1312 if (pam_code
!= PAM_SUCCESS
)
1316 pam_code
= pam_acct_mgmt(handle
, flags
);
1317 if (pam_code
!= PAM_SUCCESS
)
1320 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1321 if (pam_code
!= PAM_SUCCESS
)
1322 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1324 pam_code
= pam_open_session(handle
, flags
);
1325 if (pam_code
!= PAM_SUCCESS
)
1328 close_session
= true;
1330 e
= pam_getenvlist(handle
);
1332 pam_code
= PAM_BUF_ERR
;
1336 /* Block SIGTERM, so that we know that it won't get lost in the child */
1338 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1340 parent_pid
= getpid_cached();
1342 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1346 int sig
, ret
= EXIT_PAM
;
1348 /* The child's job is to reset the PAM session on termination */
1349 barrier_set_role(&barrier
, BARRIER_CHILD
);
1351 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1352 * those fds are open here that have been opened by PAM. */
1353 (void) close_many(fds
, n_fds
);
1355 /* Drop privileges - we don't need any to pam_close_session and this will make
1356 * PR_SET_PDEATHSIG work in most cases. If this fails, ignore the error - but expect sd-pam
1357 * threads to fail to exit normally */
1359 r
= maybe_setgroups(0, NULL
);
1361 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1362 if (setresgid(gid
, gid
, gid
) < 0)
1363 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1364 if (setresuid(uid
, uid
, uid
) < 0)
1365 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1367 (void) ignore_signals(SIGPIPE
);
1369 /* Wait until our parent died. This will only work if the above setresuid() succeeds,
1370 * otherwise the kernel will not allow unprivileged parents kill their privileged children
1371 * this way. We rely on the control groups kill logic to do the rest for us. */
1372 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1375 /* Tell the parent that our setup is done. This is especially important regarding dropping
1376 * privileges. Otherwise, unit setup might race against our setresuid(2) call.
1378 * If the parent aborted, we'll detect this below, hence ignore return failure here. */
1379 (void) barrier_place(&barrier
);
1381 /* Check if our parent process might already have died? */
1382 if (getppid() == parent_pid
) {
1385 assert_se(sigemptyset(&ss
) >= 0);
1386 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1389 if (sigwait(&ss
, &sig
) < 0) {
1396 assert(sig
== SIGTERM
);
1401 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1402 if (pam_code
!= PAM_SUCCESS
)
1405 /* If our parent died we'll end the session */
1406 if (getppid() != parent_pid
) {
1407 pam_code
= pam_close_session(handle
, flags
);
1408 if (pam_code
!= PAM_SUCCESS
)
1415 /* NB: pam_end() when called in child processes should set PAM_DATA_SILENT to let the module
1416 * know about this. See pam_end(3) */
1417 (void) pam_end(handle
, pam_code
| flags
| PAM_DATA_SILENT
);
1421 barrier_set_role(&barrier
, BARRIER_PARENT
);
1423 /* If the child was forked off successfully it will do all the cleanups, so forget about the handle
1427 /* Unblock SIGTERM again in the parent */
1428 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1430 /* We close the log explicitly here, since the PAM modules might have opened it, but we don't want
1431 * this fd around. */
1434 /* Synchronously wait for the child to initialize. We don't care for errors as we cannot
1435 * recover. However, warn loudly if it happens. */
1436 if (!barrier_place_and_sync(&barrier
))
1437 log_error("PAM initialization failed");
1439 return strv_free_and_replace(*env
, e
);
1442 if (pam_code
!= PAM_SUCCESS
) {
1443 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1444 r
= -EPERM
; /* PAM errors do not map to errno */
1446 log_error_errno(r
, "PAM failed: %m");
1450 pam_code
= pam_close_session(handle
, flags
);
1452 (void) pam_end(handle
, pam_code
| flags
);
1462 static void rename_process_from_path(const char *path
) {
1463 _cleanup_free_
char *buf
= NULL
;
1468 /* This resulting string must fit in 10 chars (i.e. the length of "/sbin/init") to look pretty in
1471 if (path_extract_filename(path
, &buf
) < 0) {
1472 rename_process("(...)");
1476 size_t l
= strlen(buf
);
1478 /* The end of the process name is usually more interesting, since the first bit might just be
1485 char process_name
[11];
1486 process_name
[0] = '(';
1487 memcpy(process_name
+1, p
, l
);
1488 process_name
[1+l
] = ')';
1489 process_name
[1+l
+1] = 0;
1491 rename_process(process_name
);
1494 static bool context_has_address_families(const ExecContext
*c
) {
1497 return c
->address_families_allow_list
||
1498 !set_isempty(c
->address_families
);
1501 static bool context_has_syscall_filters(const ExecContext
*c
) {
1504 return c
->syscall_allow_list
||
1505 !hashmap_isempty(c
->syscall_filter
);
1508 static bool context_has_syscall_logs(const ExecContext
*c
) {
1511 return c
->syscall_log_allow_list
||
1512 !hashmap_isempty(c
->syscall_log
);
1515 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1518 if (c
->no_new_privileges
)
1521 if (have_effective_cap(CAP_SYS_ADMIN
) > 0) /* if we are privileged, we don't need NNP */
1524 /* We need NNP if we have any form of seccomp and are unprivileged */
1525 return c
->lock_personality
||
1526 c
->memory_deny_write_execute
||
1527 c
->private_devices
||
1529 c
->protect_hostname
||
1530 c
->protect_kernel_tunables
||
1531 c
->protect_kernel_modules
||
1532 c
->protect_kernel_logs
||
1533 context_has_address_families(c
) ||
1534 exec_context_restrict_namespaces_set(c
) ||
1535 c
->restrict_realtime
||
1536 c
->restrict_suid_sgid
||
1537 !set_isempty(c
->syscall_archs
) ||
1538 context_has_syscall_filters(c
) ||
1539 context_has_syscall_logs(c
);
1542 bool exec_context_has_credentials(const ExecContext
*context
) {
1546 return !hashmap_isempty(context
->set_credentials
) ||
1547 !hashmap_isempty(context
->load_credentials
);
1552 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1554 if (is_seccomp_available())
1557 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1561 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1562 uint32_t negative_action
, default_action
, action
;
1568 if (!context_has_syscall_filters(c
))
1571 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1574 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1576 if (c
->syscall_allow_list
) {
1577 default_action
= negative_action
;
1578 action
= SCMP_ACT_ALLOW
;
1580 default_action
= SCMP_ACT_ALLOW
;
1581 action
= negative_action
;
1584 if (needs_ambient_hack
) {
1585 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1590 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1593 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1595 uint32_t default_action
, action
;
1601 if (!context_has_syscall_logs(c
))
1605 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1608 if (c
->syscall_log_allow_list
) {
1609 /* Log nothing but the ones listed */
1610 default_action
= SCMP_ACT_ALLOW
;
1611 action
= SCMP_ACT_LOG
;
1613 /* Log everything but the ones listed */
1614 default_action
= SCMP_ACT_LOG
;
1615 action
= SCMP_ACT_ALLOW
;
1618 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1620 /* old libseccomp */
1621 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1626 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1630 if (set_isempty(c
->syscall_archs
))
1633 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1636 return seccomp_restrict_archs(c
->syscall_archs
);
1639 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1643 if (!context_has_address_families(c
))
1646 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1649 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1652 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1658 if (!c
->memory_deny_write_execute
)
1661 /* use prctl() if kernel supports it (6.3) */
1662 r
= prctl(PR_SET_MDWE
, PR_MDWE_REFUSE_EXEC_GAIN
, 0, 0, 0);
1664 log_unit_debug(u
, "Enabled MemoryDenyWriteExecute= with PR_SET_MDWE");
1667 if (r
< 0 && errno
!= EINVAL
)
1668 return log_unit_debug_errno(u
, errno
, "Failed to enable MemoryDenyWriteExecute= with PR_SET_MDWE: %m");
1669 /* else use seccomp */
1670 log_unit_debug(u
, "Kernel doesn't support PR_SET_MDWE: falling back to seccomp");
1672 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1675 return seccomp_memory_deny_write_execute();
1678 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1682 if (!c
->restrict_realtime
)
1685 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1688 return seccomp_restrict_realtime();
1691 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1695 if (!c
->restrict_suid_sgid
)
1698 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1701 return seccomp_restrict_suid_sgid();
1704 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1708 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1709 * let's protect even those systems where this is left on in the kernel. */
1711 if (!c
->protect_kernel_tunables
)
1714 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1717 return seccomp_protect_sysctl();
1720 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1724 /* Turn off module syscalls on ProtectKernelModules=yes */
1726 if (!c
->protect_kernel_modules
)
1729 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1732 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1735 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1739 if (!c
->protect_kernel_logs
)
1742 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1745 return seccomp_protect_syslog();
1748 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1752 if (!c
->protect_clock
)
1755 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1758 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1761 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1765 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1767 if (!c
->private_devices
)
1770 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1773 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1776 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1780 if (!exec_context_restrict_namespaces_set(c
))
1783 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1786 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1789 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1790 unsigned long personality
;
1796 if (!c
->lock_personality
)
1799 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1802 personality
= c
->personality
;
1804 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1805 if (personality
== PERSONALITY_INVALID
) {
1807 r
= opinionated_personality(&personality
);
1812 return seccomp_lock_personality(personality
);
1818 static int apply_restrict_filesystems(Unit
*u
, const ExecContext
*c
) {
1822 if (!exec_context_restrict_filesystems_set(c
))
1825 if (!u
->manager
->restrict_fs
) {
1826 /* LSM BPF is unsupported or lsm_bpf_setup failed */
1827 log_unit_debug(u
, "LSM BPF not supported, skipping RestrictFileSystems=");
1831 return lsm_bpf_unit_restrict_filesystems(u
, c
->restrict_filesystems
, c
->restrict_filesystems_allow_list
);
1835 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1839 if (!c
->protect_hostname
)
1842 if (ns_type_supported(NAMESPACE_UTS
)) {
1843 if (unshare(CLONE_NEWUTS
) < 0) {
1844 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1845 *ret_exit_status
= EXIT_NAMESPACE
;
1846 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1849 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1852 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1857 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1860 r
= seccomp_protect_hostname();
1862 *ret_exit_status
= EXIT_SECCOMP
;
1863 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1870 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1873 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1874 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1876 if (idle_pipe
[0] >= 0) {
1879 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1881 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1884 /* Signal systemd that we are bored and want to continue. */
1885 n
= write(idle_pipe
[3], "x", 1);
1887 /* Wait for systemd to react to the signal above. */
1888 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1891 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1895 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1898 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1900 static int build_environment(
1902 const ExecContext
*c
,
1903 const ExecParameters
*p
,
1904 const CGroupContext
*cgroup_context
,
1908 const char *username
,
1910 dev_t journal_stream_dev
,
1911 ino_t journal_stream_ino
,
1912 const char *memory_pressure_path
,
1915 _cleanup_strv_free_
char **our_env
= NULL
;
1925 #define N_ENV_VARS 19
1926 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1931 _cleanup_free_
char *joined
= NULL
;
1933 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1935 our_env
[n_env
++] = x
;
1937 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1939 our_env
[n_env
++] = x
;
1941 joined
= strv_join(fdnames
, ":");
1945 x
= strjoin("LISTEN_FDNAMES=", joined
);
1948 our_env
[n_env
++] = x
;
1951 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1952 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1954 our_env
[n_env
++] = x
;
1956 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1958 our_env
[n_env
++] = x
;
1961 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use blocking
1962 * Varlink calls back to us for look up dynamic users in PID 1. Break the deadlock between D-Bus and
1963 * PID 1 by disabling use of PID1' NSS interface for looking up dynamic users. */
1964 if (p
->flags
& EXEC_NSS_DYNAMIC_BYPASS
) {
1965 x
= strdup("SYSTEMD_NSS_DYNAMIC_BYPASS=1");
1968 our_env
[n_env
++] = x
;
1972 x
= strjoin("HOME=", home
);
1976 path_simplify(x
+ 5);
1977 our_env
[n_env
++] = x
;
1981 x
= strjoin("LOGNAME=", username
);
1984 our_env
[n_env
++] = x
;
1986 x
= strjoin("USER=", username
);
1989 our_env
[n_env
++] = x
;
1993 x
= strjoin("SHELL=", shell
);
1997 path_simplify(x
+ 6);
1998 our_env
[n_env
++] = x
;
2001 if (!sd_id128_is_null(u
->invocation_id
)) {
2002 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
2005 our_env
[n_env
++] = x
;
2008 if (exec_context_needs_term(c
)) {
2009 _cleanup_free_
char *cmdline
= NULL
;
2010 const char *tty_path
, *term
= NULL
;
2012 tty_path
= exec_context_tty_path(c
);
2014 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
2015 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
2016 * container manager passes to PID 1 ends up all the way in the console login shown. */
2018 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
2019 term
= getenv("TERM");
2020 else if (tty_path
&& in_charset(skip_dev_prefix(tty_path
), ALPHANUMERICAL
)) {
2021 _cleanup_free_
char *key
= NULL
;
2023 key
= strjoin("systemd.tty.term.", skip_dev_prefix(tty_path
));
2027 r
= proc_cmdline_get_key(key
, 0, &cmdline
);
2029 log_debug_errno(r
, "Failed to read %s from kernel cmdline, ignoring: %m", key
);
2035 term
= default_term_for_tty(tty_path
);
2037 x
= strjoin("TERM=", term
);
2040 our_env
[n_env
++] = x
;
2043 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
2044 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
2047 our_env
[n_env
++] = x
;
2050 if (c
->log_namespace
) {
2051 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
2055 our_env
[n_env
++] = x
;
2058 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2059 _cleanup_free_
char *joined
= NULL
;
2065 if (c
->directories
[t
].n_items
== 0)
2068 n
= exec_directory_env_name_to_string(t
);
2072 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
2073 _cleanup_free_
char *prefixed
= NULL
;
2075 prefixed
= path_join(p
->prefix
[t
], c
->directories
[t
].items
[i
].path
);
2079 if (!strextend_with_separator(&joined
, ":", prefixed
))
2083 x
= strjoin(n
, "=", joined
);
2087 our_env
[n_env
++] = x
;
2090 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
2091 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
2095 our_env
[n_env
++] = x
;
2098 if (asprintf(&x
, "SYSTEMD_EXEC_PID=" PID_FMT
, getpid_cached()) < 0)
2101 our_env
[n_env
++] = x
;
2103 if (memory_pressure_path
) {
2104 x
= strjoin("MEMORY_PRESSURE_WATCH=", memory_pressure_path
);
2108 our_env
[n_env
++] = x
;
2110 if (cgroup_context
&& !path_equal(memory_pressure_path
, "/dev/null")) {
2111 _cleanup_free_
char *b
= NULL
, *e
= NULL
;
2113 if (asprintf(&b
, "%s " USEC_FMT
" " USEC_FMT
,
2114 MEMORY_PRESSURE_DEFAULT_TYPE
,
2115 cgroup_context
->memory_pressure_threshold_usec
== USEC_INFINITY
? MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC
:
2116 CLAMP(cgroup_context
->memory_pressure_threshold_usec
, 1U, MEMORY_PRESSURE_DEFAULT_WINDOW_USEC
),
2117 MEMORY_PRESSURE_DEFAULT_WINDOW_USEC
) < 0)
2120 if (base64mem(b
, strlen(b
) + 1, &e
) < 0)
2123 x
= strjoin("MEMORY_PRESSURE_WRITE=", e
);
2127 our_env
[n_env
++] = x
;
2131 assert(n_env
< N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
2134 *ret
= TAKE_PTR(our_env
);
2139 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
2140 _cleanup_strv_free_
char **pass_env
= NULL
;
2143 STRV_FOREACH(i
, c
->pass_environment
) {
2144 _cleanup_free_
char *x
= NULL
;
2150 x
= strjoin(*i
, "=", v
);
2154 if (!GREEDY_REALLOC(pass_env
, n_env
+ 2))
2157 pass_env
[n_env
++] = TAKE_PTR(x
);
2158 pass_env
[n_env
] = NULL
;
2161 *ret
= TAKE_PTR(pass_env
);
2166 bool exec_needs_network_namespace(const ExecContext
*context
) {
2169 return context
->private_network
|| context
->network_namespace_path
;
2172 static bool exec_needs_ipc_namespace(const ExecContext
*context
) {
2175 return context
->private_ipc
|| context
->ipc_namespace_path
;
2178 bool exec_needs_mount_namespace(
2179 const ExecContext
*context
,
2180 const ExecParameters
*params
,
2181 const ExecRuntime
*runtime
) {
2185 if (context
->root_image
)
2188 if (!strv_isempty(context
->read_write_paths
) ||
2189 !strv_isempty(context
->read_only_paths
) ||
2190 !strv_isempty(context
->inaccessible_paths
) ||
2191 !strv_isempty(context
->exec_paths
) ||
2192 !strv_isempty(context
->no_exec_paths
))
2195 if (context
->n_bind_mounts
> 0)
2198 if (context
->n_temporary_filesystems
> 0)
2201 if (context
->n_mount_images
> 0)
2204 if (context
->n_extension_images
> 0)
2207 if (!strv_isempty(context
->extension_directories
))
2210 if (!IN_SET(context
->mount_propagation_flag
, 0, MS_SHARED
))
2213 if (context
->private_tmp
&& runtime
&& runtime
->shared
&& (runtime
->shared
->tmp_dir
|| runtime
->shared
->var_tmp_dir
))
2216 if (context
->private_devices
||
2217 context
->private_mounts
> 0 ||
2218 (context
->private_mounts
< 0 && exec_needs_network_namespace(context
)) ||
2219 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2220 context
->protect_home
!= PROTECT_HOME_NO
||
2221 context
->protect_kernel_tunables
||
2222 context
->protect_kernel_modules
||
2223 context
->protect_kernel_logs
||
2224 context
->protect_control_groups
||
2225 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2226 context
->proc_subset
!= PROC_SUBSET_ALL
||
2227 exec_needs_ipc_namespace(context
))
2230 if (context
->root_directory
) {
2231 if (exec_context_get_effective_mount_apivfs(context
))
2234 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2235 if (params
&& !params
->prefix
[t
])
2238 if (context
->directories
[t
].n_items
> 0)
2243 if (context
->dynamic_user
&&
2244 (context
->directories
[EXEC_DIRECTORY_STATE
].n_items
> 0 ||
2245 context
->directories
[EXEC_DIRECTORY_CACHE
].n_items
> 0 ||
2246 context
->directories
[EXEC_DIRECTORY_LOGS
].n_items
> 0))
2249 if (context
->log_namespace
)
2255 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2256 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2257 _cleanup_close_pair_
int errno_pipe
[2] = PIPE_EBADF
;
2258 _cleanup_close_
int unshare_ready_fd
= -EBADF
;
2259 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2264 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2265 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2266 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2267 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2268 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2269 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2270 * continues execution normally.
2271 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2272 * does not need CAP_SETUID to write the single line mapping to itself. */
2274 /* Can only set up multiple mappings with CAP_SETUID. */
2275 if (have_effective_cap(CAP_SETUID
) > 0 && uid
!= ouid
&& uid_is_valid(uid
))
2276 r
= asprintf(&uid_map
,
2277 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2278 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2279 ouid
, ouid
, uid
, uid
);
2281 r
= asprintf(&uid_map
,
2282 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2288 /* Can only set up multiple mappings with CAP_SETGID. */
2289 if (have_effective_cap(CAP_SETGID
) > 0 && gid
!= ogid
&& gid_is_valid(gid
))
2290 r
= asprintf(&gid_map
,
2291 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2292 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2293 ogid
, ogid
, gid
, gid
);
2295 r
= asprintf(&gid_map
,
2296 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2302 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2304 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2305 if (unshare_ready_fd
< 0)
2308 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2310 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2313 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2317 _cleanup_close_
int fd
= -EBADF
;
2321 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2322 * here, after the parent opened its own user namespace. */
2325 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2327 /* Wait until the parent unshared the user namespace */
2328 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2333 /* Disable the setgroups() system call in the child user namespace, for good. */
2334 a
= procfs_file_alloca(ppid
, "setgroups");
2335 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2337 if (errno
!= ENOENT
) {
2342 /* If the file is missing the kernel is too old, let's continue anyway. */
2344 if (write(fd
, "deny\n", 5) < 0) {
2349 fd
= safe_close(fd
);
2352 /* First write the GID map */
2353 a
= procfs_file_alloca(ppid
, "gid_map");
2354 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2359 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2363 fd
= safe_close(fd
);
2365 /* The write the UID map */
2366 a
= procfs_file_alloca(ppid
, "uid_map");
2367 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2372 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2377 _exit(EXIT_SUCCESS
);
2380 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2381 _exit(EXIT_FAILURE
);
2384 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2386 if (unshare(CLONE_NEWUSER
) < 0)
2389 /* Let the child know that the namespace is ready now */
2390 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2393 /* Try to read an error code from the child */
2394 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2397 if (n
== sizeof(r
)) { /* an error code was sent to us */
2402 if (n
!= 0) /* on success we should have read 0 bytes */
2405 r
= wait_for_terminate_and_check("(sd-userns)", TAKE_PID(pid
), 0);
2408 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2414 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2417 if (!context
->dynamic_user
)
2420 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2423 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2429 static int create_many_symlinks(const char *root
, const char *source
, char **symlinks
) {
2430 _cleanup_free_
char *src_abs
= NULL
;
2435 src_abs
= path_join(root
, source
);
2439 STRV_FOREACH(dst
, symlinks
) {
2440 _cleanup_free_
char *dst_abs
= NULL
;
2442 dst_abs
= path_join(root
, *dst
);
2446 r
= mkdir_parents_label(dst_abs
, 0755);
2450 r
= symlink_idempotent(src_abs
, dst_abs
, true);
2458 static int setup_exec_directory(
2459 const ExecContext
*context
,
2460 const ExecParameters
*params
,
2463 ExecDirectoryType type
,
2464 bool needs_mount_namespace
,
2467 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2468 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2469 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2470 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2471 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2472 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2478 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2479 assert(exit_status
);
2481 if (!params
->prefix
[type
])
2484 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2485 if (!uid_is_valid(uid
))
2487 if (!gid_is_valid(gid
))
2491 for (size_t i
= 0; i
< context
->directories
[type
].n_items
; i
++) {
2492 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2494 p
= path_join(params
->prefix
[type
], context
->directories
[type
].items
[i
].path
);
2500 r
= mkdir_parents_label(p
, 0755);
2504 if (exec_directory_is_private(context
, type
)) {
2505 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2506 * case we want to avoid leaving a directory around fully accessible that is owned by
2507 * a dynamic user whose UID is later on reused. To lock this down we use the same
2508 * trick used by container managers to prohibit host users to get access to files of
2509 * the same UID in containers: we place everything inside a directory that has an
2510 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2511 * for unprivileged host code. We then use fs namespacing to make this directory
2512 * permeable for the service itself.
2514 * Specifically: for a service which wants a special directory "foo/" we first create
2515 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2516 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2517 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2518 * unprivileged host users can't look into it. Inside of the namespace of the unit
2519 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2520 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2521 * for the service and making sure it only gets access to the dirs it needs but no
2522 * others. Tricky? Yes, absolutely, but it works!
2524 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2525 * to be owned by the service itself.
2527 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2528 * for sharing files or sockets with other services. */
2530 pp
= path_join(params
->prefix
[type
], "private");
2536 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2537 r
= mkdir_safe_label(pp
, 0700, 0, 0, MKDIR_WARN_MODE
);
2541 if (!path_extend(&pp
, context
->directories
[type
].items
[i
].path
)) {
2546 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2547 r
= mkdir_parents_label(pp
, 0755);
2551 if (is_dir(p
, false) > 0 &&
2552 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2554 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2555 * it over. Most likely the service has been upgraded from one that didn't use
2556 * DynamicUser=1, to one that does. */
2558 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2559 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2560 exec_directory_type_to_string(type
), p
, pp
);
2562 if (rename(p
, pp
) < 0) {
2567 /* Otherwise, create the actual directory for the service */
2569 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2570 if (r
< 0 && r
!= -EEXIST
)
2574 if (!context
->directories
[type
].items
[i
].only_create
) {
2575 /* And link it up from the original place.
2577 * 1) If a mount namespace is going to be used, then this symlink remains on
2578 * the host, and a new one for the child namespace will be created later.
2579 * 2) It is not necessary to create this symlink when one of its parent
2580 * directories is specified and already created. E.g.
2581 * StateDirectory=foo foo/bar
2582 * In that case, the inode points to pp and p for "foo/bar" are the same:
2583 * pp = "/var/lib/private/foo/bar"
2584 * p = "/var/lib/foo/bar"
2585 * and, /var/lib/foo is a symlink to /var/lib/private/foo. So, not only
2586 * we do not need to create the symlink, but we cannot create the symlink.
2587 * See issue #24783. */
2588 r
= symlink_idempotent(pp
, p
, true);
2594 _cleanup_free_
char *target
= NULL
;
2596 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2597 readlink_and_make_absolute(p
, &target
) >= 0) {
2598 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2600 /* This already exists and is a symlink? Interesting. Maybe it's one created
2601 * by DynamicUser=1 (see above)?
2603 * We do this for all directory types except for ConfigurationDirectory=,
2604 * since they all support the private/ symlink logic at least in some
2605 * configurations, see above. */
2607 r
= chase(target
, NULL
, 0, &target_resolved
, NULL
);
2611 q
= path_join(params
->prefix
[type
], "private", context
->directories
[type
].items
[i
].path
);
2617 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2618 r
= chase(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2622 if (path_equal(q_resolved
, target_resolved
)) {
2624 /* Hmm, apparently DynamicUser= was once turned on for this service,
2625 * but is no longer. Let's move the directory back up. */
2627 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2628 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2629 exec_directory_type_to_string(type
), q
, p
);
2631 if (unlink(p
) < 0) {
2636 if (rename(q
, p
) < 0) {
2643 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2648 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2651 /* Don't change the owner/access mode of the configuration directory,
2652 * as in the common case it is not written to by a service, and shall
2653 * not be writable. */
2655 if (stat(p
, &st
) < 0) {
2660 /* Still complain if the access mode doesn't match */
2661 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2662 log_warning("%s \'%s\' already exists but the mode is different. "
2663 "(File system: %o %sMode: %o)",
2664 exec_directory_type_to_string(type
), context
->directories
[type
].items
[i
].path
,
2665 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2672 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2673 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2674 * current UID/GID ownership.) */
2675 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2679 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2680 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2681 * assignments to exist. */
2682 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2687 /* If we are not going to run in a namespace, set up the symlinks - otherwise
2688 * they are set up later, to allow configuring empty var/run/etc. */
2689 if (!needs_mount_namespace
)
2690 for (size_t i
= 0; i
< context
->directories
[type
].n_items
; i
++) {
2691 r
= create_many_symlinks(params
->prefix
[type
],
2692 context
->directories
[type
].items
[i
].path
,
2693 context
->directories
[type
].items
[i
].symlinks
);
2701 *exit_status
= exit_status_table
[type
];
2705 static int write_credential(
2711 bool ownership_ok
) {
2713 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2714 _cleanup_close_
int fd
= -EBADF
;
2717 r
= tempfn_random_child("", "cred", &tmp
);
2721 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2727 r
= loop_write(fd
, data
, size
, /* do_poll = */ false);
2731 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2734 if (uid_is_valid(uid
) && uid
!= getuid()) {
2735 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2737 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2740 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2741 * to express: that the user gets read access and nothing
2742 * else. But if the backing fs can't support that (e.g. ramfs)
2743 * then we can use file ownership instead. But that's only safe if
2744 * we can then re-mount the whole thing read-only, so that the
2745 * user can no longer chmod() the file to gain write access. */
2748 if (fchown(fd
, uid
, GID_INVALID
) < 0)
2753 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2760 static char **credential_search_path(
2761 const ExecParameters
*params
,
2764 _cleanup_strv_free_
char **l
= NULL
;
2768 /* Assemble a search path to find credentials in. We'll look in /etc/credstore/ (and similar
2769 * directories in /usr/lib/ + /run/) for all types of credentials. If we are looking for encrypted
2770 * credentials, also look in /etc/credstore.encrypted/ (and similar dirs). */
2773 if (strv_extend(&l
, params
->received_encrypted_credentials_directory
) < 0)
2776 if (strv_extend_strv(&l
, CONF_PATHS_STRV("credstore.encrypted"), /* filter_duplicates= */ true) < 0)
2780 if (params
->received_credentials_directory
)
2781 if (strv_extend(&l
, params
->received_credentials_directory
) < 0)
2784 if (strv_extend_strv(&l
, CONF_PATHS_STRV("credstore"), /* filter_duplicates= */ true) < 0)
2787 if (DEBUG_LOGGING
) {
2788 _cleanup_free_
char *t
= strv_join(l
, ":");
2790 log_debug("Credential search path is: %s", strempty(t
));
2796 static int load_credential(
2797 const ExecContext
*context
,
2798 const ExecParameters
*params
,
2809 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
|READ_FULL_FILE_FAIL_WHEN_LARGER
;
2810 _cleanup_strv_free_
char **search_path
= NULL
;
2811 _cleanup_(erase_and_freep
) char *data
= NULL
;
2812 _cleanup_free_
char *bindname
= NULL
;
2813 const char *source
= NULL
;
2814 bool missing_ok
= true;
2815 size_t size
, add
, maxsz
;
2823 assert(read_dfd
>= 0 || read_dfd
== AT_FDCWD
);
2824 assert(write_dfd
>= 0);
2827 if (read_dfd
>= 0) {
2828 /* If a directory fd is specified, then read the file directly from that dir. In this case we
2829 * won't do AF_UNIX stuff (we simply don't want to recursively iterate down a tree of AF_UNIX
2830 * IPC sockets). It's OK if a file vanishes here in the time we enumerate it and intend to
2833 if (!filename_is_valid(path
)) /* safety check */
2839 } else if (path_is_absolute(path
)) {
2840 /* If this is an absolute path, read the data directly from it, and support AF_UNIX
2843 if (!path_is_valid(path
)) /* safety check */
2846 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2848 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2849 * via the source socket address in case we read off an AF_UNIX socket. */
2850 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, id
) < 0)
2856 } else if (credential_name_valid(path
)) {
2857 /* If this is a relative path, take it as credential name relative to the credentials
2858 * directory we received ourselves. We don't support the AF_UNIX stuff in this mode, since we
2859 * are operating on a credential store, i.e. this is guaranteed to be regular files. */
2861 search_path
= credential_search_path(params
, encrypted
);
2870 flags
|= READ_FULL_FILE_UNBASE64
;
2872 maxsz
= encrypted
? CREDENTIAL_ENCRYPTED_SIZE_MAX
: CREDENTIAL_SIZE_MAX
;
2875 STRV_FOREACH(d
, search_path
) {
2876 _cleanup_free_
char *j
= NULL
;
2878 j
= path_join(*d
, path
);
2882 r
= read_full_file_full(
2883 AT_FDCWD
, j
, /* path is absolute, hence pass AT_FDCWD as nop dir fd here */
2893 r
= read_full_file_full(
2903 if (r
== -ENOENT
&& (missing_ok
|| hashmap_contains(context
->set_credentials
, id
))) {
2904 /* Make a missing inherited credential non-fatal, let's just continue. After all apps
2905 * will get clear errors if we don't pass such a missing credential on as they
2906 * themselves will get ENOENT when trying to read them, which should not be much
2907 * worse than when we handle the error here and make it fatal.
2909 * Also, if the source file doesn't exist, but a fallback is set via SetCredentials=
2910 * we are fine, too. */
2911 log_debug_errno(r
, "Couldn't read inherited credential '%s', skipping: %m", path
);
2915 return log_debug_errno(r
, "Failed to read credential '%s': %m", path
);
2918 _cleanup_free_
void *plaintext
= NULL
;
2919 size_t plaintext_size
= 0;
2921 r
= decrypt_credential_and_warn(id
, now(CLOCK_REALTIME
), NULL
, NULL
, data
, size
, &plaintext
, &plaintext_size
);
2925 free_and_replace(data
, plaintext
);
2926 size
= plaintext_size
;
2929 add
= strlen(id
) + size
;
2933 r
= write_credential(write_dfd
, id
, data
, size
, uid
, ownership_ok
);
2935 return log_debug_errno(r
, "Failed to write credential '%s': %m", id
);
2941 struct load_cred_args
{
2942 const ExecContext
*context
;
2943 const ExecParameters
*params
;
2952 static int load_cred_recurse_dir_cb(
2953 RecurseDirEvent event
,
2957 const struct dirent
*de
,
2958 const struct statx
*sx
,
2961 struct load_cred_args
*args
= ASSERT_PTR(userdata
);
2962 _cleanup_free_
char *sub_id
= NULL
;
2965 if (event
!= RECURSE_DIR_ENTRY
)
2966 return RECURSE_DIR_CONTINUE
;
2968 if (!IN_SET(de
->d_type
, DT_REG
, DT_SOCK
))
2969 return RECURSE_DIR_CONTINUE
;
2971 sub_id
= strreplace(path
, "/", "_");
2975 if (!credential_name_valid(sub_id
))
2976 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential would get ID %s, which is not valid, refusing", sub_id
);
2978 if (faccessat(args
->dfd
, sub_id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0) {
2979 log_debug("Skipping credential with duplicated ID %s at %s", sub_id
, path
);
2980 return RECURSE_DIR_CONTINUE
;
2982 if (errno
!= ENOENT
)
2983 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sub_id
);
2985 r
= load_credential(
3000 return RECURSE_DIR_CONTINUE
;
3003 static int acquire_credentials(
3004 const ExecContext
*context
,
3005 const ExecParameters
*params
,
3009 bool ownership_ok
) {
3011 uint64_t left
= CREDENTIALS_TOTAL_SIZE_MAX
;
3012 _cleanup_close_
int dfd
= -EBADF
;
3013 ExecLoadCredential
*lc
;
3014 ExecSetCredential
*sc
;
3020 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
3024 /* First, load credentials off disk (or acquire via AF_UNIX socket) */
3025 HASHMAP_FOREACH(lc
, context
->load_credentials
) {
3026 _cleanup_close_
int sub_fd
= -EBADF
;
3028 /* If this is an absolute path, then try to open it as a directory. If that works, then we'll
3029 * recurse into it. If it is an absolute path but it isn't a directory, then we'll open it as
3030 * a regular file. Finally, if it's a relative path we will use it as a credential name to
3031 * propagate a credential passed to us from further up. */
3033 if (path_is_absolute(lc
->path
)) {
3034 sub_fd
= open(lc
->path
, O_DIRECTORY
|O_CLOEXEC
|O_RDONLY
);
3035 if (sub_fd
< 0 && !IN_SET(errno
,
3036 ENOTDIR
, /* Not a directory */
3037 ENOENT
)) /* Doesn't exist? */
3038 return log_debug_errno(errno
, "Failed to open '%s': %m", lc
->path
);
3042 /* Regular file (incl. a credential passed in from higher up) */
3043 r
= load_credential(
3059 /* path= */ lc
->id
, /* recurse_dir() will suffix the subdir paths from here to the top-level id */
3060 /* statx_mask= */ 0,
3061 /* n_depth_max= */ UINT_MAX
,
3062 RECURSE_DIR_SORT
|RECURSE_DIR_IGNORE_DOT
|RECURSE_DIR_ENSURE_TYPE
,
3063 load_cred_recurse_dir_cb
,
3064 &(struct load_cred_args
) {
3067 .encrypted
= lc
->encrypted
,
3071 .ownership_ok
= ownership_ok
,
3078 /* Second, we add in literally specified credentials. If the credentials already exist, we'll not add
3079 * them, so that they can act as a "default" if the same credential is specified multiple times. */
3080 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
3081 _cleanup_(erase_and_freep
) void *plaintext
= NULL
;
3085 /* Note that we check ahead of time here instead of relying on O_EXCL|O_CREAT later to return
3086 * EEXIST if the credential already exists. That's because the TPM2-based decryption is kinda
3087 * slow and involved, hence it's nice to be able to skip that if the credential already
3089 if (faccessat(dfd
, sc
->id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0)
3091 if (errno
!= ENOENT
)
3092 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sc
->id
);
3094 if (sc
->encrypted
) {
3095 r
= decrypt_credential_and_warn(sc
->id
, now(CLOCK_REALTIME
), NULL
, NULL
, sc
->data
, sc
->size
, &plaintext
, &size
);
3105 add
= strlen(sc
->id
) + size
;
3109 r
= write_credential(dfd
, sc
->id
, data
, size
, uid
, ownership_ok
);
3116 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
3119 /* After we created all keys with the right perms, also make sure the credential store as a whole is
3122 if (uid_is_valid(uid
) && uid
!= getuid()) {
3123 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
3125 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
3131 if (fchown(dfd
, uid
, GID_INVALID
) < 0)
3139 static int setup_credentials_internal(
3140 const ExecContext
*context
,
3141 const ExecParameters
*params
,
3143 const char *final
, /* This is where the credential store shall eventually end up at */
3144 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
3145 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
3146 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
3149 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
3150 * if we mounted something; false if we definitely can't mount anything */
3158 if (reuse_workspace
) {
3159 r
= path_is_mount_point(workspace
, NULL
, 0);
3163 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
3165 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
3167 workspace_mounted
= -1; /* ditto */
3169 r
= path_is_mount_point(final
, NULL
, 0);
3173 /* If the final place already has something mounted, we use that. If the workspace also has
3174 * something mounted we assume it's actually the same mount (but with MS_RDONLY
3176 final_mounted
= true;
3178 if (workspace_mounted
< 0) {
3179 /* If the final place is mounted, but the workspace isn't, then let's bind mount
3180 * the final version to the workspace, and make it writable, so that we can make
3183 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
3187 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3191 workspace_mounted
= true;
3194 final_mounted
= false;
3196 if (workspace_mounted
< 0) {
3197 /* Nothing is mounted on the workspace yet, let's try to mount something now */
3198 for (int try = 0;; try++) {
3201 /* Try "ramfs" first, since it's not swap backed */
3202 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
3204 workspace_mounted
= true;
3208 } else if (try == 1) {
3209 _cleanup_free_
char *opts
= NULL
;
3211 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%zu", (size_t) CREDENTIALS_TOTAL_SIZE_MAX
) < 0)
3214 /* Fall back to "tmpfs" otherwise */
3215 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
3217 workspace_mounted
= true;
3222 /* If that didn't work, try to make a bind mount from the final to the workspace, so that we can make it writable there. */
3223 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
3225 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
3228 if (must_mount
) /* If we it's not OK to use the plain directory
3229 * fallback, propagate all errors too */
3232 /* If we lack privileges to bind mount stuff, then let's gracefully
3233 * proceed for compat with container envs, and just use the final dir
3236 workspace_mounted
= false;
3240 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
3241 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3245 workspace_mounted
= true;
3251 assert(!must_mount
|| workspace_mounted
> 0);
3252 where
= workspace_mounted
? workspace
: final
;
3254 (void) label_fix_full(AT_FDCWD
, where
, final
, 0);
3256 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
3260 if (workspace_mounted
) {
3263 /* Determine if we should actually install the prepared mount in the final location by bind
3264 * mounting it there. We do so only if the mount is not established there already, and if the
3265 * mount is actually non-empty (i.e. carries at least one credential). Not that in the best
3266 * case we are doing all this in a mount namespace, thus no one else will see that we
3267 * allocated a file system we are getting rid of again here. */
3269 install
= false; /* already installed */
3271 r
= dir_is_empty(where
, /* ignore_hidden_or_backup= */ false);
3275 install
= r
== 0; /* install only if non-empty */
3279 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
3280 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3284 /* And mount it to the final place, read-only */
3285 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
3287 /* Otherwise get rid of it */
3288 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
3292 _cleanup_free_
char *parent
= NULL
;
3294 /* If we do not have our own mount put used the plain directory fallback, then we need to
3295 * open access to the top-level credential directory and the per-service directory now */
3297 r
= path_extract_directory(final
, &parent
);
3300 if (chmod(parent
, 0755) < 0)
3307 static int setup_credentials(
3308 const ExecContext
*context
,
3309 const ExecParameters
*params
,
3313 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
3319 if (!exec_context_has_credentials(context
))
3322 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
3325 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
3326 * and the subdir we mount over with a read-only file system readable by the service's user */
3327 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
3331 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
3332 if (r
< 0 && r
!= -EEXIST
)
3335 p
= path_join(q
, unit
);
3339 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
3340 if (r
< 0 && r
!= -EEXIST
)
3343 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
3345 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
3347 /* If this is not a privilege or support issue then propagate the error */
3348 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
3351 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
3352 * it into place, so that users can't access half-initialized credential stores. */
3353 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
3357 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
3358 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
3359 * after it is fully set up */
3360 u
= path_join(t
, unit
);
3364 FOREACH_STRING(i
, t
, u
) {
3365 r
= mkdir_label(i
, 0700);
3366 if (r
< 0 && r
!= -EEXIST
)
3370 r
= setup_credentials_internal(
3374 p
, /* final mount point */
3375 u
, /* temporary workspace to overmount */
3376 true, /* reuse the workspace if it is already a mount */
3377 false, /* it's OK to fall back to a plain directory if we can't mount anything */
3380 (void) rmdir(u
); /* remove the workspace again if we can. */
3385 } else if (r
== 0) {
3387 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
3388 * we can use the same directory for all cases, after turning off propagation. Question
3389 * though is: where do we turn off propagation exactly, and where do we place the workspace
3390 * directory? We need some place that is guaranteed to be a mount point in the host, and
3391 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
3392 * since we ultimately want to move the resulting file system there, i.e. we need propagation
3393 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
3394 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
3395 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
3396 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
3397 * propagation on the former, and then overmount the latter.
3399 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
3400 * for this purpose, but there are few other candidates that work equally well for us, and
3401 * given that the we do this in a privately namespaced short-lived single-threaded process
3402 * that no one else sees this should be OK to do. */
3404 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
3408 r
= setup_credentials_internal(
3412 p
, /* final mount point */
3413 "/dev/shm", /* temporary workspace to overmount */
3414 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
3415 true, /* insist that something is mounted, do not allow fallback to plain directory */
3420 _exit(EXIT_SUCCESS
);
3423 _exit(EXIT_FAILURE
);
3426 /* If the credentials dir is empty and not a mount point, then there's no point in having it. Let's
3427 * try to remove it. This matters in particular if we created the dir as mount point but then didn't
3428 * actually end up mounting anything on it. In that case we'd rather have ENOENT than EACCESS being
3429 * seen by users when trying access this inode. */
3435 static int setup_smack(
3436 const Manager
*manager
,
3437 const ExecContext
*context
,
3438 int executable_fd
) {
3442 assert(executable_fd
>= 0);
3444 if (context
->smack_process_label
) {
3445 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
3448 } else if (manager
->default_smack_process_label
) {
3449 _cleanup_free_
char *exec_label
= NULL
;
3451 r
= mac_smack_read_fd(executable_fd
, SMACK_ATTR_EXEC
, &exec_label
);
3452 if (r
< 0 && !ERRNO_IS_XATTR_ABSENT(r
))
3455 r
= mac_smack_apply_pid(0, exec_label
?: manager
->default_smack_process_label
);
3464 static int compile_bind_mounts(
3465 const ExecContext
*context
,
3466 const ExecParameters
*params
,
3467 BindMount
**ret_bind_mounts
,
3468 size_t *ret_n_bind_mounts
,
3469 char ***ret_empty_directories
) {
3471 _cleanup_strv_free_
char **empty_directories
= NULL
;
3472 BindMount
*bind_mounts
= NULL
;
3478 assert(ret_bind_mounts
);
3479 assert(ret_n_bind_mounts
);
3480 assert(ret_empty_directories
);
3482 CLEANUP_ARRAY(bind_mounts
, h
, bind_mount_free_many
);
3484 n
= context
->n_bind_mounts
;
3485 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3486 if (!params
->prefix
[t
])
3489 for (size_t i
= 0; i
< context
->directories
[t
].n_items
; i
++)
3490 n
+= !context
->directories
[t
].items
[i
].only_create
;
3494 *ret_bind_mounts
= NULL
;
3495 *ret_n_bind_mounts
= 0;
3496 *ret_empty_directories
= NULL
;
3500 bind_mounts
= new(BindMount
, n
);
3504 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
3505 BindMount
*item
= context
->bind_mounts
+ i
;
3506 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
3508 s
= strdup(item
->source
);
3512 d
= strdup(item
->destination
);
3516 bind_mounts
[h
++] = (BindMount
) {
3517 .source
= TAKE_PTR(s
),
3518 .destination
= TAKE_PTR(d
),
3519 .read_only
= item
->read_only
,
3520 .recursive
= item
->recursive
,
3521 .ignore_enoent
= item
->ignore_enoent
,
3525 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3526 if (!params
->prefix
[t
])
3529 if (context
->directories
[t
].n_items
== 0)
3532 if (exec_directory_is_private(context
, t
) &&
3533 !exec_context_with_rootfs(context
)) {
3536 /* So this is for a dynamic user, and we need to make sure the process can access its own
3537 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3538 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3540 private_root
= path_join(params
->prefix
[t
], "private");
3544 r
= strv_consume(&empty_directories
, private_root
);
3549 for (size_t i
= 0; i
< context
->directories
[t
].n_items
; i
++) {
3550 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
3552 /* When one of the parent directories is in the list, we cannot create the symlink
3553 * for the child directory. See also the comments in setup_exec_directory(). */
3554 if (context
->directories
[t
].items
[i
].only_create
)
3557 if (exec_directory_is_private(context
, t
))
3558 s
= path_join(params
->prefix
[t
], "private", context
->directories
[t
].items
[i
].path
);
3560 s
= path_join(params
->prefix
[t
], context
->directories
[t
].items
[i
].path
);
3564 if (exec_directory_is_private(context
, t
) &&
3565 exec_context_with_rootfs(context
))
3566 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3567 * directory is not created on the root directory. So, let's bind-mount the directory
3568 * on the 'non-private' place. */
3569 d
= path_join(params
->prefix
[t
], context
->directories
[t
].items
[i
].path
);
3575 bind_mounts
[h
++] = (BindMount
) {
3576 .source
= TAKE_PTR(s
),
3577 .destination
= TAKE_PTR(d
),
3579 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3581 .ignore_enoent
= false,
3588 *ret_bind_mounts
= TAKE_PTR(bind_mounts
);
3589 *ret_n_bind_mounts
= n
;
3590 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3595 /* ret_symlinks will contain a list of pairs src:dest that describes
3596 * the symlinks to create later on. For example, the symlinks needed
3597 * to safely give private directories to DynamicUser=1 users. */
3598 static int compile_symlinks(
3599 const ExecContext
*context
,
3600 const ExecParameters
*params
,
3601 char ***ret_symlinks
) {
3603 _cleanup_strv_free_
char **symlinks
= NULL
;
3608 assert(ret_symlinks
);
3610 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3611 for (size_t i
= 0; i
< context
->directories
[dt
].n_items
; i
++) {
3612 _cleanup_free_
char *private_path
= NULL
, *path
= NULL
;
3614 STRV_FOREACH(symlink
, context
->directories
[dt
].items
[i
].symlinks
) {
3615 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
3617 src_abs
= path_join(params
->prefix
[dt
], context
->directories
[dt
].items
[i
].path
);
3618 dst_abs
= path_join(params
->prefix
[dt
], *symlink
);
3619 if (!src_abs
|| !dst_abs
)
3622 r
= strv_consume_pair(&symlinks
, TAKE_PTR(src_abs
), TAKE_PTR(dst_abs
));
3627 if (!exec_directory_is_private(context
, dt
) ||
3628 exec_context_with_rootfs(context
) ||
3629 context
->directories
[dt
].items
[i
].only_create
)
3632 private_path
= path_join(params
->prefix
[dt
], "private", context
->directories
[dt
].items
[i
].path
);
3636 path
= path_join(params
->prefix
[dt
], context
->directories
[dt
].items
[i
].path
);
3640 r
= strv_consume_pair(&symlinks
, TAKE_PTR(private_path
), TAKE_PTR(path
));
3646 *ret_symlinks
= TAKE_PTR(symlinks
);
3651 static bool insist_on_sandboxing(
3652 const ExecContext
*context
,
3653 const char *root_dir
,
3654 const char *root_image
,
3655 const BindMount
*bind_mounts
,
3656 size_t n_bind_mounts
) {
3659 assert(n_bind_mounts
== 0 || bind_mounts
);
3661 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3662 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3663 * rearrange stuff in a way we cannot ignore gracefully. */
3665 if (context
->n_temporary_filesystems
> 0)
3668 if (root_dir
|| root_image
)
3671 if (context
->n_mount_images
> 0)
3674 if (context
->dynamic_user
)
3677 if (context
->n_extension_images
> 0 || !strv_isempty(context
->extension_directories
))
3680 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3682 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3683 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3686 if (context
->log_namespace
)
3692 static int apply_mount_namespace(
3694 ExecCommandFlags command_flags
,
3695 const ExecContext
*context
,
3696 const ExecParameters
*params
,
3697 const ExecRuntime
*runtime
,
3698 const char *memory_pressure_path
,
3699 char **error_path
) {
3701 _cleanup_strv_free_
char **empty_directories
= NULL
, **symlinks
= NULL
,
3702 **read_write_paths_cleanup
= NULL
;
3703 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3704 const char *root_dir
= NULL
, *root_image
= NULL
;
3705 _cleanup_free_
char *creds_path
= NULL
, *incoming_dir
= NULL
, *propagate_dir
= NULL
,
3706 *extension_dir
= NULL
;
3707 char **read_write_paths
;
3708 NamespaceInfo ns_info
;
3709 bool needs_sandboxing
;
3710 BindMount
*bind_mounts
= NULL
;
3711 size_t n_bind_mounts
= 0;
3716 CLEANUP_ARRAY(bind_mounts
, n_bind_mounts
, bind_mount_free_many
);
3718 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3719 root_image
= context
->root_image
;
3722 root_dir
= context
->root_directory
;
3725 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3729 /* Symlinks for exec dirs are set up after other mounts, before they are made read-only. */
3730 r
= compile_symlinks(context
, params
, &symlinks
);
3734 /* We need to make the pressure path writable even if /sys/fs/cgroups is made read-only, as the
3735 * service will need to write to it in order to start the notifications. */
3736 if (context
->protect_control_groups
&& memory_pressure_path
&& !streq(memory_pressure_path
, "/dev/null")) {
3737 read_write_paths_cleanup
= strv_copy(context
->read_write_paths
);
3738 if (!read_write_paths_cleanup
)
3741 r
= strv_extend(&read_write_paths_cleanup
, memory_pressure_path
);
3745 read_write_paths
= read_write_paths_cleanup
;
3747 read_write_paths
= context
->read_write_paths
;
3749 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3750 if (needs_sandboxing
) {
3751 /* The runtime struct only contains the parent of the private /tmp,
3752 * which is non-accessible to world users. Inside of it there's a /tmp
3753 * that is sticky, and that's the one we want to use here.
3754 * This does not apply when we are using /run/systemd/empty as fallback. */
3756 if (context
->private_tmp
&& runtime
&& runtime
->shared
) {
3757 if (streq_ptr(runtime
->shared
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3758 tmp_dir
= runtime
->shared
->tmp_dir
;
3759 else if (runtime
->shared
->tmp_dir
)
3760 tmp_dir
= strjoina(runtime
->shared
->tmp_dir
, "/tmp");
3762 if (streq_ptr(runtime
->shared
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3763 var_tmp_dir
= runtime
->shared
->var_tmp_dir
;
3764 else if (runtime
->shared
->var_tmp_dir
)
3765 var_tmp_dir
= strjoina(runtime
->shared
->var_tmp_dir
, "/tmp");
3768 ns_info
= (NamespaceInfo
) {
3769 .ignore_protect_paths
= false,
3770 .private_dev
= context
->private_devices
,
3771 .protect_control_groups
= context
->protect_control_groups
,
3772 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3773 .protect_kernel_modules
= context
->protect_kernel_modules
,
3774 .protect_kernel_logs
= context
->protect_kernel_logs
,
3775 .protect_hostname
= context
->protect_hostname
,
3776 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3777 .protect_home
= context
->protect_home
,
3778 .protect_system
= context
->protect_system
,
3779 .protect_proc
= context
->protect_proc
,
3780 .proc_subset
= context
->proc_subset
,
3781 .private_network
= exec_needs_network_namespace(context
),
3782 .private_ipc
= exec_needs_ipc_namespace(context
),
3783 /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
3784 .mount_nosuid
= context
->no_new_privileges
&& !mac_selinux_use(),
3786 } else if (!context
->dynamic_user
&& root_dir
)
3788 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3789 * sandbox info, otherwise enforce it, don't ignore protected paths and
3790 * fail if we are enable to apply the sandbox inside the mount namespace.
3792 ns_info
= (NamespaceInfo
) {
3793 .ignore_protect_paths
= true,
3796 ns_info
= (NamespaceInfo
) {};
3798 if (context
->mount_propagation_flag
== MS_SHARED
)
3799 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3801 if (exec_context_has_credentials(context
) &&
3802 params
->prefix
[EXEC_DIRECTORY_RUNTIME
] &&
3803 FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
3804 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3809 if (MANAGER_IS_SYSTEM(u
->manager
)) {
3810 propagate_dir
= path_join("/run/systemd/propagate/", u
->id
);
3814 incoming_dir
= strdup("/run/systemd/incoming");
3818 extension_dir
= strdup("/run/systemd/unit-extensions");
3822 if (asprintf(&extension_dir
, "/run/user/" UID_FMT
"/systemd/unit-extensions", geteuid()) < 0)
3825 r
= setup_namespace(
3828 context
->root_image_options
,
3829 context
->root_image_policy
?: &image_policy_service
,
3832 needs_sandboxing
? context
->read_only_paths
: NULL
,
3833 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3834 needs_sandboxing
? context
->exec_paths
: NULL
,
3835 needs_sandboxing
? context
->no_exec_paths
: NULL
,
3840 context
->temporary_filesystems
,
3841 context
->n_temporary_filesystems
,
3842 context
->mount_images
,
3843 context
->n_mount_images
,
3844 context
->mount_image_policy
?: &image_policy_service
,
3848 context
->log_namespace
,
3849 context
->mount_propagation_flag
,
3850 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3851 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3852 context
->root_verity
,
3853 context
->extension_images
,
3854 context
->n_extension_images
,
3855 context
->extension_image_policy
?: &image_policy_sysext
,
3856 context
->extension_directories
,
3860 root_dir
|| root_image
? params
->notify_socket
: NULL
,
3863 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3864 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3865 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3866 * completely different execution environment. */
3868 if (insist_on_sandboxing(
3870 root_dir
, root_image
,
3873 return log_unit_debug_errno(u
,
3874 SYNTHETIC_ERRNO(EOPNOTSUPP
),
3875 "Failed to set up namespace, and refusing to continue since "
3876 "the selected namespacing options alter mount environment non-trivially.\n"
3877 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3879 context
->n_temporary_filesystems
,
3882 yes_no(context
->dynamic_user
));
3884 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3891 static int apply_working_directory(
3892 const ExecContext
*context
,
3893 const ExecParameters
*params
,
3900 assert(exit_status
);
3902 if (context
->working_directory_home
) {
3905 *exit_status
= EXIT_CHDIR
;
3912 wd
= empty_to_root(context
->working_directory
);
3914 if (params
->flags
& EXEC_APPLY_CHROOT
)
3917 d
= prefix_roota(context
->root_directory
, wd
);
3919 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3920 *exit_status
= EXIT_CHDIR
;
3927 static int apply_root_directory(
3928 const ExecContext
*context
,
3929 const ExecParameters
*params
,
3930 const bool needs_mount_ns
,
3934 assert(exit_status
);
3936 if (params
->flags
& EXEC_APPLY_CHROOT
)
3937 if (!needs_mount_ns
&& context
->root_directory
)
3938 if (chroot(context
->root_directory
) < 0) {
3939 *exit_status
= EXIT_CHROOT
;
3946 static int setup_keyring(
3948 const ExecContext
*context
,
3949 const ExecParameters
*p
,
3950 uid_t uid
, gid_t gid
) {
3952 key_serial_t keyring
;
3961 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3962 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3963 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3964 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3965 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3966 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3968 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3971 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3972 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3973 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3974 * & group is just as nasty as acquiring a reference to the user keyring. */
3976 saved_uid
= getuid();
3977 saved_gid
= getgid();
3979 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3980 if (setregid(gid
, -1) < 0)
3981 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3984 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3985 if (setreuid(uid
, -1) < 0) {
3986 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3991 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3992 if (keyring
== -1) {
3993 if (errno
== ENOSYS
)
3994 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3995 else if (ERRNO_IS_PRIVILEGE(errno
))
3996 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3997 else if (errno
== EDQUOT
)
3998 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
4000 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
4005 /* When requested link the user keyring into the session keyring. */
4006 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
4008 if (keyctl(KEYCTL_LINK
,
4009 KEY_SPEC_USER_KEYRING
,
4010 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
4011 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
4016 /* Restore uid/gid back */
4017 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
4018 if (setreuid(saved_uid
, -1) < 0) {
4019 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
4024 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
4025 if (setregid(saved_gid
, -1) < 0)
4026 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
4029 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
4030 if (!sd_id128_is_null(u
->invocation_id
)) {
4033 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
4035 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
4037 if (keyctl(KEYCTL_SETPERM
, key
,
4038 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
4039 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
4040 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
4045 /* Revert back uid & gid for the last time, and exit */
4046 /* no extra logging, as only the first already reported error matters */
4047 if (getuid() != saved_uid
)
4048 (void) setreuid(saved_uid
, -1);
4050 if (getgid() != saved_gid
)
4051 (void) setregid(saved_gid
, -1);
4056 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
4062 array
[(*n
)++] = pair
[0];
4064 array
[(*n
)++] = pair
[1];
4067 static int close_remaining_fds(
4068 const ExecParameters
*params
,
4069 const ExecRuntime
*runtime
,
4072 const int *fds
, size_t n_fds
) {
4074 size_t n_dont_close
= 0;
4075 int dont_close
[n_fds
+ 12];
4079 if (params
->stdin_fd
>= 0)
4080 dont_close
[n_dont_close
++] = params
->stdin_fd
;
4081 if (params
->stdout_fd
>= 0)
4082 dont_close
[n_dont_close
++] = params
->stdout_fd
;
4083 if (params
->stderr_fd
>= 0)
4084 dont_close
[n_dont_close
++] = params
->stderr_fd
;
4087 dont_close
[n_dont_close
++] = socket_fd
;
4089 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
4090 n_dont_close
+= n_fds
;
4093 if (runtime
&& runtime
->shared
) {
4094 append_socket_pair(dont_close
, &n_dont_close
, runtime
->shared
->netns_storage_socket
);
4095 append_socket_pair(dont_close
, &n_dont_close
, runtime
->shared
->ipcns_storage_socket
);
4098 if (runtime
&& runtime
->dynamic_creds
) {
4099 if (runtime
->dynamic_creds
->user
)
4100 append_socket_pair(dont_close
, &n_dont_close
, runtime
->dynamic_creds
->user
->storage_socket
);
4101 if (runtime
->dynamic_creds
->group
)
4102 append_socket_pair(dont_close
, &n_dont_close
, runtime
->dynamic_creds
->group
->storage_socket
);
4105 if (user_lookup_fd
>= 0)
4106 dont_close
[n_dont_close
++] = user_lookup_fd
;
4108 return close_all_fds(dont_close
, n_dont_close
);
4111 static int send_user_lookup(
4119 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
4120 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
4123 if (user_lookup_fd
< 0)
4126 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
4129 if (writev(user_lookup_fd
,
4131 IOVEC_MAKE(&uid
, sizeof(uid
)),
4132 IOVEC_MAKE(&gid
, sizeof(gid
)),
4133 IOVEC_MAKE_STRING(unit
->id
) }, 3) < 0)
4139 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
4146 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
4151 if (!c
->working_directory_home
)
4154 r
= get_home_dir(buf
);
4162 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
4163 _cleanup_strv_free_
char ** list
= NULL
;
4170 assert(c
->dynamic_user
);
4172 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
4173 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
4176 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
4177 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
4183 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
4186 if (exec_directory_is_private(c
, t
))
4187 e
= path_join(p
->prefix
[t
], "private", c
->directories
[t
].items
[i
].path
);
4189 e
= path_join(p
->prefix
[t
], c
->directories
[t
].items
[i
].path
);
4193 r
= strv_consume(&list
, e
);
4199 *ret
= TAKE_PTR(list
);
4204 static int exec_parameters_get_cgroup_path(
4205 const ExecParameters
*params
,
4206 const CGroupContext
*c
,
4209 const char *subgroup
= NULL
;
4215 if (!params
->cgroup_path
)
4218 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
4219 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
4220 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
4221 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
4222 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
4223 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
4224 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
4225 * flag, which is only passed for the former statements, not for the latter. */
4227 if (FLAGS_SET(params
->flags
, EXEC_CGROUP_DELEGATE
) && (FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
) || c
->delegate_subgroup
)) {
4228 if (FLAGS_SET(params
->flags
, EXEC_IS_CONTROL
))
4229 subgroup
= ".control";
4231 subgroup
= c
->delegate_subgroup
;
4235 p
= path_join(params
->cgroup_path
, subgroup
);
4237 p
= strdup(params
->cgroup_path
);
4245 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
4246 _cleanup_(cpu_set_reset
) CPUSet s
= {};
4252 if (!c
->numa_policy
.nodes
.set
) {
4253 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
4257 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
4263 return cpu_set_add_all(ret
, &s
);
4266 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
4269 return c
->cpu_affinity_from_numa
;
4272 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
4277 assert(*n_fds
< fds_size
);
4285 if (fd
< 3 + (int) *n_fds
) {
4286 /* Let's move the fd up, so that it's outside of the fd range we will use to store
4287 * the fds we pass to the process (or which are closed only during execve). */
4289 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
4293 close_and_replace(fd
, r
);
4296 *ret_fd
= fds
[*n_fds
] = fd
;
4301 static int connect_unix_harder(Unit
*u
, const OpenFile
*of
, int ofd
) {
4302 union sockaddr_union addr
= {
4303 .un
.sun_family
= AF_UNIX
,
4306 static const int socket_types
[] = { SOCK_DGRAM
, SOCK_STREAM
, SOCK_SEQPACKET
};
4313 r
= sockaddr_un_set_path(&addr
.un
, FORMAT_PROC_FD_PATH(ofd
));
4315 return log_unit_error_errno(u
, r
, "Failed to set sockaddr for %s: %m", of
->path
);
4319 for (size_t i
= 0; i
< ELEMENTSOF(socket_types
); i
++) {
4320 _cleanup_close_
int fd
= -EBADF
;
4322 fd
= socket(AF_UNIX
, socket_types
[i
] | SOCK_CLOEXEC
, 0);
4324 return log_unit_error_errno(u
, errno
, "Failed to create socket for %s: %m", of
->path
);
4326 r
= RET_NERRNO(connect(fd
, &addr
.sa
, sa_len
));
4327 if (r
== -EPROTOTYPE
)
4330 return log_unit_error_errno(u
, r
, "Failed to connect socket for %s: %m", of
->path
);
4335 return log_unit_error_errno(u
, SYNTHETIC_ERRNO(EPROTOTYPE
), "Failed to connect socket for \"%s\".", of
->path
);
4338 static int get_open_file_fd(Unit
*u
, const OpenFile
*of
) {
4340 _cleanup_close_
int fd
= -EBADF
, ofd
= -EBADF
;
4345 ofd
= open(of
->path
, O_PATH
| O_CLOEXEC
);
4347 return log_unit_error_errno(u
, errno
, "Could not open \"%s\": %m", of
->path
);
4349 if (fstat(ofd
, &st
) < 0)
4350 return log_unit_error_errno(u
, errno
, "Failed to stat %s: %m", of
->path
);
4352 if (S_ISSOCK(st
.st_mode
)) {
4353 fd
= connect_unix_harder(u
, of
, ofd
);
4357 if (FLAGS_SET(of
->flags
, OPENFILE_READ_ONLY
) && shutdown(fd
, SHUT_WR
) < 0)
4358 return log_unit_error_errno(u
, errno
, "Failed to shutdown send for socket %s: %m",
4361 log_unit_debug(u
, "socket %s opened (fd=%d)", of
->path
, fd
);
4363 int flags
= FLAGS_SET(of
->flags
, OPENFILE_READ_ONLY
) ? O_RDONLY
: O_RDWR
;
4364 if (FLAGS_SET(of
->flags
, OPENFILE_APPEND
))
4366 else if (FLAGS_SET(of
->flags
, OPENFILE_TRUNCATE
))
4369 fd
= fd_reopen(ofd
, flags
| O_CLOEXEC
);
4371 return log_unit_error_errno(u
, fd
, "Failed to open file %s: %m", of
->path
);
4373 log_unit_debug(u
, "file %s opened (fd=%d)", of
->path
, fd
);
4379 static int collect_open_file_fds(
4381 OpenFile
* open_files
,
4392 LIST_FOREACH(open_files
, of
, open_files
) {
4393 _cleanup_close_
int fd
= -EBADF
;
4395 fd
= get_open_file_fd(u
, of
);
4397 if (FLAGS_SET(of
->flags
, OPENFILE_GRACEFUL
)) {
4398 log_unit_debug_errno(u
, fd
, "Failed to get OpenFile= file descriptor for %s, ignoring: %m", of
->path
);
4405 if (!GREEDY_REALLOC(*fds
, *n_fds
+ 1))
4408 r
= strv_extend(fdnames
, of
->fdname
);
4412 (*fds
)[*n_fds
] = TAKE_FD(fd
);
4420 static void log_command_line(Unit
*unit
, const char *msg
, const char *executable
, char **argv
) {
4428 _cleanup_free_
char *cmdline
= quote_command_line(argv
, SHELL_ESCAPE_EMPTY
);
4430 log_unit_struct(unit
, LOG_DEBUG
,
4431 "EXECUTABLE=%s", executable
,
4432 LOG_UNIT_MESSAGE(unit
, "%s: %s", msg
, strnull(cmdline
)),
4433 LOG_UNIT_INVOCATION_ID(unit
));
4436 static bool exec_context_need_unprivileged_private_users(const ExecContext
*context
, const Manager
*manager
) {
4440 /* These options require PrivateUsers= when used in user units, as we need to be in a user namespace
4441 * to have permission to enable them when not running as root. If we have effective CAP_SYS_ADMIN
4442 * (system manager) then we have privileges and don't need this. */
4443 if (MANAGER_IS_SYSTEM(manager
))
4446 return context
->private_users
||
4447 context
->private_tmp
||
4448 context
->private_devices
||
4449 context
->private_network
||
4450 context
->network_namespace_path
||
4451 context
->private_ipc
||
4452 context
->ipc_namespace_path
||
4453 context
->private_mounts
||
4454 context
->mount_apivfs
||
4455 context
->n_bind_mounts
> 0 ||
4456 context
->n_temporary_filesystems
> 0 ||
4457 context
->root_directory
||
4458 !strv_isempty(context
->extension_directories
) ||
4459 context
->protect_system
!= PROTECT_SYSTEM_NO
||
4460 context
->protect_home
!= PROTECT_HOME_NO
||
4461 context
->protect_kernel_tunables
||
4462 context
->protect_kernel_modules
||
4463 context
->protect_kernel_logs
||
4464 context
->protect_control_groups
||
4465 context
->protect_clock
||
4466 context
->protect_hostname
||
4467 !strv_isempty(context
->read_write_paths
) ||
4468 !strv_isempty(context
->read_only_paths
) ||
4469 !strv_isempty(context
->inaccessible_paths
) ||
4470 !strv_isempty(context
->exec_paths
) ||
4471 !strv_isempty(context
->no_exec_paths
);
4474 static int exec_child(
4476 const ExecCommand
*command
,
4477 const ExecContext
*context
,
4478 const ExecParameters
*params
,
4479 ExecRuntime
*runtime
,
4480 const CGroupContext
*cgroup_context
,
4482 const int named_iofds
[static 3],
4484 size_t n_socket_fds
,
4485 size_t n_storage_fds
,
4490 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **joined_exec_search_path
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
4491 int r
, ngids
= 0, exec_fd
;
4492 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
4493 const char *username
= NULL
, *groupname
= NULL
;
4494 _cleanup_free_
char *home_buffer
= NULL
, *memory_pressure_path
= NULL
;
4495 const char *home
= NULL
, *shell
= NULL
;
4496 char **final_argv
= NULL
;
4497 dev_t journal_stream_dev
= 0;
4498 ino_t journal_stream_ino
= 0;
4499 bool userns_set_up
= false;
4500 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
4501 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
4502 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
4503 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
4505 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
4506 bool use_selinux
= false;
4509 bool use_smack
= false;
4512 bool use_apparmor
= false;
4514 uid_t saved_uid
= getuid();
4515 gid_t saved_gid
= getgid();
4516 uid_t uid
= UID_INVALID
;
4517 gid_t gid
= GID_INVALID
;
4518 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
4519 n_keep_fds
; /* total number of fds not to close */
4521 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
4522 int ngids_after_pam
= 0;
4523 _cleanup_free_
int *fds
= NULL
;
4524 _cleanup_strv_free_
char **fdnames
= NULL
;
4530 assert(exit_status
);
4532 /* Explicitly test for CVE-2021-4034 inspired invocations */
4533 assert(command
->path
);
4534 assert(!strv_isempty(command
->argv
));
4536 rename_process_from_path(command
->path
);
4538 /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main
4539 * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially,
4540 * both of which will be demoted to SIG_DFL. */
4541 (void) default_signals(SIGNALS_CRASH_HANDLER
,
4544 if (context
->ignore_sigpipe
)
4545 (void) ignore_signals(SIGPIPE
);
4547 r
= reset_signal_mask();
4549 *exit_status
= EXIT_SIGNAL_MASK
;
4550 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
4553 if (params
->idle_pipe
)
4554 do_idle_pipe_dance(params
->idle_pipe
);
4556 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
4557 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
4558 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
4559 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
4562 log_set_open_when_needed(true);
4563 log_settle_target();
4565 /* In case anything used libc syslog(), close this here, too */
4568 fds
= newdup(int, params_fds
, n_fds
);
4570 *exit_status
= EXIT_MEMORY
;
4574 fdnames
= strv_copy((char**) params
->fd_names
);
4576 *exit_status
= EXIT_MEMORY
;
4580 r
= collect_open_file_fds(unit
, params
->open_files
, &fds
, &fdnames
, &n_fds
);
4582 *exit_status
= EXIT_FDS
;
4583 return log_unit_error_errno(unit
, r
, "Failed to get OpenFile= file descriptors: %m");
4586 int keep_fds
[n_fds
+ 3];
4587 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
4590 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
4592 *exit_status
= EXIT_FDS
;
4593 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4597 if (unit
->manager
->restrict_fs
) {
4598 int bpf_map_fd
= lsm_bpf_map_restrict_fs_fd(unit
);
4599 if (bpf_map_fd
< 0) {
4600 *exit_status
= EXIT_FDS
;
4601 return log_unit_error_errno(unit
, bpf_map_fd
, "Failed to get restrict filesystems BPF map fd: %m");
4604 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, bpf_map_fd
, &bpf_map_fd
);
4606 *exit_status
= EXIT_FDS
;
4607 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4612 r
= close_remaining_fds(params
, runtime
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
4614 *exit_status
= EXIT_FDS
;
4615 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
4618 if (!context
->same_pgrp
&&
4620 *exit_status
= EXIT_SETSID
;
4621 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
4624 exec_context_tty_reset(context
, params
);
4626 if (unit_shall_confirm_spawn(unit
)) {
4627 _cleanup_free_
char *cmdline
= NULL
;
4629 cmdline
= quote_command_line(command
->argv
, SHELL_ESCAPE_EMPTY
);
4631 *exit_status
= EXIT_MEMORY
;
4635 r
= ask_for_confirmation(context
, params
->confirm_spawn
, unit
, cmdline
);
4636 if (r
!= CONFIRM_EXECUTE
) {
4637 if (r
== CONFIRM_PRETEND_SUCCESS
) {
4638 *exit_status
= EXIT_SUCCESS
;
4641 *exit_status
= EXIT_CONFIRM
;
4642 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
4643 "Execution cancelled by the user");
4647 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
4648 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
4649 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
4650 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
4651 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
4652 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
4653 setenv("SYSTEMD_ACTIVATION_SCOPE", runtime_scope_to_string(unit
->manager
->runtime_scope
), true) != 0) {
4654 *exit_status
= EXIT_MEMORY
;
4655 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
4658 if (context
->dynamic_user
&& runtime
&& runtime
->dynamic_creds
) {
4659 _cleanup_strv_free_
char **suggested_paths
= NULL
;
4661 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
4662 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
4663 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
4664 *exit_status
= EXIT_USER
;
4665 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
4668 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
4670 *exit_status
= EXIT_MEMORY
;
4674 r
= dynamic_creds_realize(runtime
->dynamic_creds
, suggested_paths
, &uid
, &gid
);
4676 *exit_status
= EXIT_USER
;
4678 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4679 "Failed to update dynamic user credentials: User or group with specified name already exists.");
4680 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
4683 if (!uid_is_valid(uid
)) {
4684 *exit_status
= EXIT_USER
;
4685 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "UID validation failed for \""UID_FMT
"\"", uid
);
4688 if (!gid_is_valid(gid
)) {
4689 *exit_status
= EXIT_USER
;
4690 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "GID validation failed for \""GID_FMT
"\"", gid
);
4693 if (runtime
->dynamic_creds
->user
)
4694 username
= runtime
->dynamic_creds
->user
->name
;
4697 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
4699 *exit_status
= EXIT_USER
;
4700 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
4703 r
= get_fixed_group(context
, &groupname
, &gid
);
4705 *exit_status
= EXIT_GROUP
;
4706 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
4710 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
4711 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
4712 &supplementary_gids
, &ngids
);
4714 *exit_status
= EXIT_GROUP
;
4715 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
4718 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
4720 *exit_status
= EXIT_USER
;
4721 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
4724 user_lookup_fd
= safe_close(user_lookup_fd
);
4726 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
4728 *exit_status
= EXIT_CHDIR
;
4729 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
4732 /* If a socket is connected to STDIN/STDOUT/STDERR, we must drop O_NONBLOCK */
4734 (void) fd_nonblock(socket_fd
, false);
4736 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
4737 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
4738 if (params
->cgroup_path
) {
4739 _cleanup_free_
char *p
= NULL
;
4741 r
= exec_parameters_get_cgroup_path(params
, cgroup_context
, &p
);
4743 *exit_status
= EXIT_CGROUP
;
4744 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
4747 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
4748 if (r
== -EUCLEAN
) {
4749 *exit_status
= EXIT_CGROUP
;
4750 return log_unit_error_errno(unit
, r
, "Failed to attach process to cgroup %s "
4751 "because the cgroup or one of its parents or "
4752 "siblings is in the threaded mode: %m", p
);
4755 *exit_status
= EXIT_CGROUP
;
4756 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
4760 if (context
->network_namespace_path
&& runtime
&& runtime
->shared
&& runtime
->shared
->netns_storage_socket
[0] >= 0) {
4761 r
= open_shareable_ns_path(runtime
->shared
->netns_storage_socket
, context
->network_namespace_path
, CLONE_NEWNET
);
4763 *exit_status
= EXIT_NETWORK
;
4764 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
4768 if (context
->ipc_namespace_path
&& runtime
&& runtime
->shared
&& runtime
->shared
->ipcns_storage_socket
[0] >= 0) {
4769 r
= open_shareable_ns_path(runtime
->shared
->ipcns_storage_socket
, context
->ipc_namespace_path
, CLONE_NEWIPC
);
4771 *exit_status
= EXIT_NAMESPACE
;
4772 return log_unit_error_errno(unit
, r
, "Failed to open IPC namespace path %s: %m", context
->ipc_namespace_path
);
4776 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
4778 *exit_status
= EXIT_STDIN
;
4779 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
4782 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
4784 *exit_status
= EXIT_STDOUT
;
4785 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
4788 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
4790 *exit_status
= EXIT_STDERR
;
4791 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
4794 if (context
->oom_score_adjust_set
) {
4795 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
4796 * prohibit write access to this file, and we shouldn't trip up over that. */
4797 r
= set_oom_score_adjust(context
->oom_score_adjust
);
4798 if (ERRNO_IS_PRIVILEGE(r
))
4799 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
4801 *exit_status
= EXIT_OOM_ADJUST
;
4802 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
4806 if (context
->coredump_filter_set
) {
4807 r
= set_coredump_filter(context
->coredump_filter
);
4808 if (ERRNO_IS_PRIVILEGE(r
))
4809 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
4811 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
4814 if (context
->nice_set
) {
4815 r
= setpriority_closest(context
->nice
);
4817 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
4820 if (context
->cpu_sched_set
) {
4821 struct sched_param param
= {
4822 .sched_priority
= context
->cpu_sched_priority
,
4825 r
= sched_setscheduler(0,
4826 context
->cpu_sched_policy
|
4827 (context
->cpu_sched_reset_on_fork
?
4828 SCHED_RESET_ON_FORK
: 0),
4831 *exit_status
= EXIT_SETSCHEDULER
;
4832 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
4836 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
4837 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
4838 const CPUSet
*cpu_set
;
4840 if (context
->cpu_affinity_from_numa
) {
4841 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
4843 *exit_status
= EXIT_CPUAFFINITY
;
4844 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
4847 cpu_set
= &converted_cpu_set
;
4849 cpu_set
= &context
->cpu_set
;
4851 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
4852 *exit_status
= EXIT_CPUAFFINITY
;
4853 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
4857 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
4858 r
= apply_numa_policy(&context
->numa_policy
);
4860 if (ERRNO_IS_NOT_SUPPORTED(r
))
4861 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
4863 *exit_status
= EXIT_NUMA_POLICY
;
4864 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
4869 if (context
->ioprio_set
)
4870 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
4871 *exit_status
= EXIT_IOPRIO
;
4872 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
4875 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
4876 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
4877 *exit_status
= EXIT_TIMERSLACK
;
4878 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
4881 if (context
->personality
!= PERSONALITY_INVALID
) {
4882 r
= safe_personality(context
->personality
);
4884 *exit_status
= EXIT_PERSONALITY
;
4885 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
4889 if (context
->utmp_id
) {
4890 const char *line
= context
->tty_path
?
4891 (path_startswith(context
->tty_path
, "/dev/") ?: context
->tty_path
) :
4893 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
4895 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
4896 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
4901 if (uid_is_valid(uid
)) {
4902 r
= chown_terminal(STDIN_FILENO
, uid
);
4904 *exit_status
= EXIT_STDIN
;
4905 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
4909 if (params
->cgroup_path
) {
4910 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4911 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4912 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4913 * touch a single hierarchy too. */
4915 if (params
->flags
& EXEC_CGROUP_DELEGATE
) {
4916 _cleanup_free_
char *p
= NULL
;
4918 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
4920 *exit_status
= EXIT_CGROUP
;
4921 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4924 r
= exec_parameters_get_cgroup_path(params
, cgroup_context
, &p
);
4926 *exit_status
= EXIT_CGROUP
;
4927 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
4930 r
= cg_set_access_recursive(SYSTEMD_CGROUP_CONTROLLER
, p
, uid
, gid
);
4932 *exit_status
= EXIT_CGROUP
;
4933 return log_unit_error_errno(unit
, r
, "Failed to adjust control subgroup access: %m");
4938 if (cgroup_context
&& cg_unified() > 0 && is_pressure_supported() > 0) {
4939 if (cgroup_context_want_memory_pressure(cgroup_context
)) {
4940 r
= cg_get_path("memory", params
->cgroup_path
, "memory.pressure", &memory_pressure_path
);
4942 *exit_status
= EXIT_MEMORY
;
4946 r
= chmod_and_chown(memory_pressure_path
, 0644, uid
, gid
);
4948 log_unit_full_errno(unit
, r
== -ENOENT
|| ERRNO_IS_PRIVILEGE(r
) ? LOG_DEBUG
: LOG_WARNING
, r
,
4949 "Failed to adjust ownership of '%s', ignoring: %m", memory_pressure_path
);
4950 memory_pressure_path
= mfree(memory_pressure_path
);
4952 } else if (cgroup_context
->memory_pressure_watch
== CGROUP_PRESSURE_WATCH_OFF
) {
4953 memory_pressure_path
= strdup("/dev/null"); /* /dev/null is explicit indicator for turning of memory pressure watch */
4954 if (!memory_pressure_path
) {
4955 *exit_status
= EXIT_MEMORY
;
4962 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4964 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4965 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, needs_mount_namespace
, exit_status
);
4967 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4970 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4971 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4973 *exit_status
= EXIT_CREDENTIALS
;
4974 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4978 r
= build_environment(
4990 memory_pressure_path
,
4993 *exit_status
= EXIT_MEMORY
;
4997 r
= build_pass_environment(context
, &pass_env
);
4999 *exit_status
= EXIT_MEMORY
;
5003 /* The $PATH variable is set to the default path in params->environment. However, this is overridden
5004 * if user-specified fields have $PATH set. The intention is to also override $PATH if the unit does
5005 * not specify PATH but the unit has ExecSearchPath. */
5006 if (!strv_isempty(context
->exec_search_path
)) {
5007 _cleanup_free_
char *joined
= NULL
;
5009 joined
= strv_join(context
->exec_search_path
, ":");
5011 *exit_status
= EXIT_MEMORY
;
5015 r
= strv_env_assign(&joined_exec_search_path
, "PATH", joined
);
5017 *exit_status
= EXIT_MEMORY
;
5022 accum_env
= strv_env_merge(params
->environment
,
5024 joined_exec_search_path
,
5026 context
->environment
,
5029 *exit_status
= EXIT_MEMORY
;
5032 accum_env
= strv_env_clean(accum_env
);
5034 (void) umask(context
->umask
);
5036 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
5038 *exit_status
= EXIT_KEYRING
;
5039 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
5042 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted
5044 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
5046 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked
5047 * for it, and the kernel doesn't actually support ambient caps. */
5048 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
5050 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly
5051 * excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not
5053 if (needs_ambient_hack
)
5054 needs_setuid
= false;
5056 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
5058 uint64_t capability_ambient_set
= context
->capability_ambient_set
;
5060 if (needs_sandboxing
) {
5061 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on
5062 * /sys being present. The actual MAC context application will happen later, as late as
5063 * possible, to avoid impacting our own code paths. */
5066 use_selinux
= mac_selinux_use();
5069 use_smack
= mac_smack_use();
5072 use_apparmor
= mac_apparmor_use();
5076 if (needs_sandboxing
) {
5079 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
5080 * is set here. (See below.) */
5082 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
5084 *exit_status
= EXIT_LIMITS
;
5085 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
5089 if (needs_setuid
&& context
->pam_name
&& username
) {
5090 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
5091 * wins here. (See above.) */
5093 /* All fds passed in the fds array will be closed in the pam child process. */
5094 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
5096 *exit_status
= EXIT_PAM
;
5097 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
5100 if (ambient_capabilities_supported()) {
5101 uint64_t ambient_after_pam
;
5103 /* PAM modules might have set some ambient caps. Query them here and merge them into
5104 * the caps we want to set in the end, so that we don't end up unsetting them. */
5105 r
= capability_get_ambient(&ambient_after_pam
);
5107 *exit_status
= EXIT_CAPABILITIES
;
5108 return log_unit_error_errno(unit
, r
, "Failed to query ambient caps: %m");
5111 capability_ambient_set
|= ambient_after_pam
;
5114 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
5115 if (ngids_after_pam
< 0) {
5116 *exit_status
= EXIT_MEMORY
;
5117 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
5121 if (needs_sandboxing
&& exec_context_need_unprivileged_private_users(context
, unit
->manager
)) {
5122 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
5123 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
5124 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
5126 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
5127 /* If it was requested explicitly and we can't set it up, fail early. Otherwise, continue and let
5128 * the actual requested operations fail (or silently continue). */
5129 if (r
< 0 && context
->private_users
) {
5130 *exit_status
= EXIT_USER
;
5131 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
5134 log_unit_info_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user, ignoring: %m");
5136 userns_set_up
= true;
5139 if (exec_needs_network_namespace(context
) && runtime
&& runtime
->shared
&& runtime
->shared
->netns_storage_socket
[0] >= 0) {
5141 if (ns_type_supported(NAMESPACE_NET
)) {
5142 r
= setup_shareable_ns(runtime
->shared
->netns_storage_socket
, CLONE_NEWNET
);
5144 if (ERRNO_IS_PRIVILEGE(r
))
5145 log_unit_warning_errno(unit
, r
,
5146 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
5148 *exit_status
= EXIT_NETWORK
;
5149 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
5152 } else if (context
->network_namespace_path
) {
5153 *exit_status
= EXIT_NETWORK
;
5154 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
5155 "NetworkNamespacePath= is not supported, refusing.");
5157 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
5160 if (exec_needs_ipc_namespace(context
) && runtime
&& runtime
->shared
&& runtime
->shared
->ipcns_storage_socket
[0] >= 0) {
5162 if (ns_type_supported(NAMESPACE_IPC
)) {
5163 r
= setup_shareable_ns(runtime
->shared
->ipcns_storage_socket
, CLONE_NEWIPC
);
5165 log_unit_warning_errno(unit
, r
,
5166 "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
5168 *exit_status
= EXIT_NAMESPACE
;
5169 return log_unit_error_errno(unit
, r
, "Failed to set up IPC namespacing: %m");
5171 } else if (context
->ipc_namespace_path
) {
5172 *exit_status
= EXIT_NAMESPACE
;
5173 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
5174 "IPCNamespacePath= is not supported, refusing.");
5176 log_unit_warning(unit
, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
5179 if (needs_mount_namespace
) {
5180 _cleanup_free_
char *error_path
= NULL
;
5182 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, memory_pressure_path
, &error_path
);
5184 *exit_status
= EXIT_NAMESPACE
;
5185 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
5186 error_path
? ": " : "", strempty(error_path
));
5190 if (needs_sandboxing
) {
5191 r
= apply_protect_hostname(unit
, context
, exit_status
);
5196 if (context
->memory_ksm
>= 0)
5197 if (prctl(PR_SET_MEMORY_MERGE
, context
->memory_ksm
) < 0) {
5198 if (ERRNO_IS_NOT_SUPPORTED(errno
))
5199 log_unit_debug_errno(unit
, errno
, "KSM support not available, ignoring.");
5201 *exit_status
= EXIT_KSM
;
5202 return log_unit_error_errno(unit
, errno
, "Failed to set KSM: %m");
5206 /* Drop groups as early as possible.
5207 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
5208 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
5210 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
5211 int ngids_to_enforce
= 0;
5213 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
5218 if (ngids_to_enforce
< 0) {
5219 *exit_status
= EXIT_MEMORY
;
5220 return log_unit_error_errno(unit
,
5222 "Failed to merge group lists. Group membership might be incorrect: %m");
5225 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
5227 *exit_status
= EXIT_GROUP
;
5228 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
5232 /* If the user namespace was not set up above, try to do it now.
5233 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
5234 * restricted by rules pertaining to combining user namespaces with other namespaces (e.g. in the
5235 * case of mount namespaces being less privileged when the mount point list is copied from a
5236 * different user namespace). */
5238 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
5239 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
5241 *exit_status
= EXIT_USER
;
5242 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
5246 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
5249 _cleanup_free_
char *executable
= NULL
;
5250 _cleanup_close_
int executable_fd
= -EBADF
;
5251 r
= find_executable_full(command
->path
, /* root= */ NULL
, context
->exec_search_path
, false, &executable
, &executable_fd
);
5253 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
5254 log_unit_struct_errno(unit
, LOG_INFO
, r
,
5255 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
5256 LOG_UNIT_INVOCATION_ID(unit
),
5257 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
5259 "EXECUTABLE=%s", command
->path
);
5263 *exit_status
= EXIT_EXEC
;
5265 return log_unit_struct_errno(unit
, LOG_INFO
, r
,
5266 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
5267 LOG_UNIT_INVOCATION_ID(unit
),
5268 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
5270 "EXECUTABLE=%s", command
->path
);
5273 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, executable_fd
, &executable_fd
);
5275 *exit_status
= EXIT_FDS
;
5276 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
5280 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
) {
5285 else if (params
->n_socket_fds
== 1)
5286 /* If stdin is not connected to a socket but we are triggered by exactly one socket unit then we
5287 * use context from that fd to compute the label. */
5288 fd
= params
->fds
[0];
5291 r
= mac_selinux_get_child_mls_label(fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
5293 if (!context
->selinux_context_ignore
) {
5294 *exit_status
= EXIT_SELINUX_CONTEXT
;
5295 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
5297 log_unit_debug_errno(unit
, r
, "Failed to determine SELinux context, ignoring: %m");
5303 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that
5304 * we are more aggressive this time, since we don't need socket_fd and the netns and ipcns fds any
5305 * more. We do keep exec_fd however, if we have it, since we need to keep it open until the final
5308 r
= close_all_fds(keep_fds
, n_keep_fds
);
5310 r
= shift_fds(fds
, n_fds
);
5312 r
= flags_fds(fds
, n_socket_fds
, n_fds
, context
->non_blocking
);
5314 *exit_status
= EXIT_FDS
;
5315 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
5318 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
5319 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
5320 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
5323 secure_bits
= context
->secure_bits
;
5325 if (needs_sandboxing
) {
5328 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested.
5329 * (Note this is placed after the general resource limit initialization, see above, in order
5330 * to take precedence.) */
5331 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
5332 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
5333 *exit_status
= EXIT_LIMITS
;
5334 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
5339 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
5340 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
5342 r
= setup_smack(unit
->manager
, context
, executable_fd
);
5343 if (r
< 0 && !context
->smack_process_label_ignore
) {
5344 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
5345 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
5350 bset
= context
->capability_bounding_set
;
5351 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
5352 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
5353 * instead of us doing that */
5354 if (needs_ambient_hack
)
5355 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
5356 (UINT64_C(1) << CAP_SETUID
) |
5357 (UINT64_C(1) << CAP_SETGID
);
5359 if (!cap_test_all(bset
)) {
5360 r
= capability_bounding_set_drop(bset
, /* right_now= */ false);
5362 *exit_status
= EXIT_CAPABILITIES
;
5363 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
5367 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
5370 * To be able to raise the ambient capabilities after setresuid() they have to be added to
5371 * the inherited set and keep caps has to be set (done in enforce_user()). After setresuid()
5372 * the ambient capabilities can be raised as they are present in the permitted and
5373 * inhertiable set. However it is possible that someone wants to set ambient capabilities
5374 * without changing the user, so we also set the ambient capabilities here.
5376 * The requested ambient capabilities are raised in the inheritable set if the second
5377 * argument is true. */
5378 if (!needs_ambient_hack
) {
5379 r
= capability_ambient_set_apply(capability_ambient_set
, /* also_inherit= */ true);
5381 *exit_status
= EXIT_CAPABILITIES
;
5382 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
5387 /* chroot to root directory first, before we lose the ability to chroot */
5388 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
5390 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
5393 if (uid_is_valid(uid
)) {
5394 r
= enforce_user(context
, uid
, capability_ambient_set
);
5396 *exit_status
= EXIT_USER
;
5397 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
5400 if (!needs_ambient_hack
&& capability_ambient_set
!= 0) {
5402 /* Raise the ambient capabilities after user change. */
5403 r
= capability_ambient_set_apply(capability_ambient_set
, /* also_inherit= */ false);
5405 *exit_status
= EXIT_CAPABILITIES
;
5406 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
5412 /* Apply working directory here, because the working directory might be on NFS and only the user running
5413 * this service might have the correct privilege to change to the working directory */
5414 r
= apply_working_directory(context
, params
, home
, exit_status
);
5416 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
5418 if (needs_sandboxing
) {
5419 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
5420 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
5421 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
5422 * are restricted. */
5426 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
5429 r
= setexeccon(exec_context
);
5431 if (!context
->selinux_context_ignore
) {
5432 *exit_status
= EXIT_SELINUX_CONTEXT
;
5433 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
5435 log_unit_debug_errno(unit
, r
, "Failed to change SELinux context to %s, ignoring: %m", exec_context
);
5442 if (use_apparmor
&& context
->apparmor_profile
) {
5443 r
= aa_change_onexec(context
->apparmor_profile
);
5444 if (r
< 0 && !context
->apparmor_profile_ignore
) {
5445 *exit_status
= EXIT_APPARMOR_PROFILE
;
5446 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
5451 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential
5452 * EPERMs we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits
5453 * requires CAP_SETPCAP. */
5454 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
5455 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
5456 * effective set here.
5458 * The effective set is overwritten during execve() with the following values:
5460 * - ambient set (for non-root processes)
5462 * - (inheritable | bounding) set for root processes)
5464 * Hence there is no security impact to raise it in the effective set before execve
5466 r
= capability_gain_cap_setpcap(/* return_caps= */ NULL
);
5468 *exit_status
= EXIT_CAPABILITIES
;
5469 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
5471 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
5472 *exit_status
= EXIT_SECUREBITS
;
5473 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
5477 if (context_has_no_new_privileges(context
))
5478 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
5479 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
5480 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
5484 r
= apply_address_families(unit
, context
);
5486 *exit_status
= EXIT_ADDRESS_FAMILIES
;
5487 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
5490 r
= apply_memory_deny_write_execute(unit
, context
);
5492 *exit_status
= EXIT_SECCOMP
;
5493 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
5496 r
= apply_restrict_realtime(unit
, context
);
5498 *exit_status
= EXIT_SECCOMP
;
5499 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
5502 r
= apply_restrict_suid_sgid(unit
, context
);
5504 *exit_status
= EXIT_SECCOMP
;
5505 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
5508 r
= apply_restrict_namespaces(unit
, context
);
5510 *exit_status
= EXIT_SECCOMP
;
5511 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
5514 r
= apply_protect_sysctl(unit
, context
);
5516 *exit_status
= EXIT_SECCOMP
;
5517 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
5520 r
= apply_protect_kernel_modules(unit
, context
);
5522 *exit_status
= EXIT_SECCOMP
;
5523 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
5526 r
= apply_protect_kernel_logs(unit
, context
);
5528 *exit_status
= EXIT_SECCOMP
;
5529 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
5532 r
= apply_protect_clock(unit
, context
);
5534 *exit_status
= EXIT_SECCOMP
;
5535 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
5538 r
= apply_private_devices(unit
, context
);
5540 *exit_status
= EXIT_SECCOMP
;
5541 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
5544 r
= apply_syscall_archs(unit
, context
);
5546 *exit_status
= EXIT_SECCOMP
;
5547 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
5550 r
= apply_lock_personality(unit
, context
);
5552 *exit_status
= EXIT_SECCOMP
;
5553 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
5556 r
= apply_syscall_log(unit
, context
);
5558 *exit_status
= EXIT_SECCOMP
;
5559 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
5562 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
5563 * by the filter as little as possible. */
5564 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
5566 *exit_status
= EXIT_SECCOMP
;
5567 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
5572 r
= apply_restrict_filesystems(unit
, context
);
5574 *exit_status
= EXIT_BPF
;
5575 return log_unit_error_errno(unit
, r
, "Failed to restrict filesystems: %m");
5581 if (!strv_isempty(context
->unset_environment
)) {
5584 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
5586 *exit_status
= EXIT_MEMORY
;
5590 strv_free_and_replace(accum_env
, ee
);
5593 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
5594 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
5595 if (!replaced_argv
) {
5596 *exit_status
= EXIT_MEMORY
;
5599 final_argv
= replaced_argv
;
5601 final_argv
= command
->argv
;
5603 log_command_line(unit
, "Executing", executable
, final_argv
);
5608 /* We have finished with all our initializations. Let's now let the manager know that. From this point
5609 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
5611 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
5612 *exit_status
= EXIT_EXEC
;
5613 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
5617 r
= fexecve_or_execve(executable_fd
, executable
, final_argv
, accum_env
);
5622 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
5623 * that POLLHUP on it no longer means execve() succeeded. */
5625 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
5626 *exit_status
= EXIT_EXEC
;
5627 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
5631 *exit_status
= EXIT_EXEC
;
5632 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
5635 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
5636 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
5638 int exec_spawn(Unit
*unit
,
5639 ExecCommand
*command
,
5640 const ExecContext
*context
,
5641 const ExecParameters
*params
,
5642 ExecRuntime
*runtime
,
5643 const CGroupContext
*cgroup_context
,
5646 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
5647 _cleanup_free_
char *subcgroup_path
= NULL
;
5648 _cleanup_strv_free_
char **files_env
= NULL
;
5649 size_t n_storage_fds
= 0, n_socket_fds
= 0;
5657 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
5659 LOG_CONTEXT_PUSH_UNIT(unit
);
5661 if (context
->std_input
== EXEC_INPUT_SOCKET
||
5662 context
->std_output
== EXEC_OUTPUT_SOCKET
||
5663 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
5665 if (params
->n_socket_fds
> 1)
5666 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got more than one socket.");
5668 if (params
->n_socket_fds
== 0)
5669 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got no socket.");
5671 socket_fd
= params
->fds
[0];
5675 n_socket_fds
= params
->n_socket_fds
;
5676 n_storage_fds
= params
->n_storage_fds
;
5679 r
= exec_context_named_iofds(context
, params
, named_iofds
);
5681 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
5683 r
= exec_context_load_environment(unit
, context
, &files_env
);
5685 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
5687 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
5688 and, until the next SELinux policy changes, we save further reloads in future children. */
5689 mac_selinux_maybe_reload();
5691 /* We won't know the real executable path until we create the mount namespace in the child, but we
5692 want to log from the parent, so we use the possibly inaccurate path here. */
5693 log_command_line(unit
, "About to execute", command
->path
, command
->argv
);
5695 if (params
->cgroup_path
) {
5696 r
= exec_parameters_get_cgroup_path(params
, cgroup_context
, &subcgroup_path
);
5698 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
5700 /* If there's a subcgroup, then let's create it here now (the main cgroup was already
5701 * realized by the unit logic) */
5703 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
5705 return log_unit_error_errno(unit
, r
, "Failed to create subcgroup '%s': %m", subcgroup_path
);
5711 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
5714 int exit_status
= EXIT_SUCCESS
;
5716 r
= exec_child(unit
,
5728 unit
->manager
->user_lookup_fds
[1],
5732 const char *status
=
5733 exit_status_to_string(exit_status
,
5734 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
5736 log_unit_struct_errno(unit
, LOG_ERR
, r
,
5737 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
5738 LOG_UNIT_INVOCATION_ID(unit
),
5739 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
5740 status
, command
->path
),
5741 "EXECUTABLE=%s", command
->path
);
5747 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
5749 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
5750 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
5751 * process will be killed too). */
5753 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
5755 exec_status_start(&command
->exec_status
, pid
);
5761 void exec_context_init(ExecContext
*c
) {
5765 c
->ioprio
= IOPRIO_DEFAULT_CLASS_AND_PRIO
;
5766 c
->cpu_sched_policy
= SCHED_OTHER
;
5767 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
5768 c
->syslog_level_prefix
= true;
5769 c
->ignore_sigpipe
= true;
5770 c
->timer_slack_nsec
= NSEC_INFINITY
;
5771 c
->personality
= PERSONALITY_INVALID
;
5772 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5773 c
->directories
[t
].mode
= 0755;
5774 c
->timeout_clean_usec
= USEC_INFINITY
;
5775 c
->capability_bounding_set
= CAP_MASK_UNSET
;
5776 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
5777 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
5778 c
->log_level_max
= -1;
5780 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
5782 c
->tty_rows
= UINT_MAX
;
5783 c
->tty_cols
= UINT_MAX
;
5784 numa_policy_reset(&c
->numa_policy
);
5785 c
->private_mounts
= -1;
5789 void exec_context_done(ExecContext
*c
) {
5792 c
->environment
= strv_free(c
->environment
);
5793 c
->environment_files
= strv_free(c
->environment_files
);
5794 c
->pass_environment
= strv_free(c
->pass_environment
);
5795 c
->unset_environment
= strv_free(c
->unset_environment
);
5797 rlimit_free_all(c
->rlimit
);
5799 for (size_t l
= 0; l
< 3; l
++) {
5800 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
5801 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
5804 c
->working_directory
= mfree(c
->working_directory
);
5805 c
->root_directory
= mfree(c
->root_directory
);
5806 c
->root_image
= mfree(c
->root_image
);
5807 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
5808 c
->root_hash
= mfree(c
->root_hash
);
5809 c
->root_hash_size
= 0;
5810 c
->root_hash_path
= mfree(c
->root_hash_path
);
5811 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
5812 c
->root_hash_sig_size
= 0;
5813 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
5814 c
->root_verity
= mfree(c
->root_verity
);
5815 c
->extension_images
= mount_image_free_many(c
->extension_images
, &c
->n_extension_images
);
5816 c
->extension_directories
= strv_free(c
->extension_directories
);
5817 c
->tty_path
= mfree(c
->tty_path
);
5818 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
5819 c
->user
= mfree(c
->user
);
5820 c
->group
= mfree(c
->group
);
5822 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
5824 c
->pam_name
= mfree(c
->pam_name
);
5826 c
->read_only_paths
= strv_free(c
->read_only_paths
);
5827 c
->read_write_paths
= strv_free(c
->read_write_paths
);
5828 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
5829 c
->exec_paths
= strv_free(c
->exec_paths
);
5830 c
->no_exec_paths
= strv_free(c
->no_exec_paths
);
5831 c
->exec_search_path
= strv_free(c
->exec_search_path
);
5833 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
5834 c
->bind_mounts
= NULL
;
5835 c
->n_bind_mounts
= 0;
5836 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
5837 c
->temporary_filesystems
= NULL
;
5838 c
->n_temporary_filesystems
= 0;
5839 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
5841 cpu_set_reset(&c
->cpu_set
);
5842 numa_policy_reset(&c
->numa_policy
);
5844 c
->utmp_id
= mfree(c
->utmp_id
);
5845 c
->selinux_context
= mfree(c
->selinux_context
);
5846 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
5847 c
->smack_process_label
= mfree(c
->smack_process_label
);
5849 c
->restrict_filesystems
= set_free(c
->restrict_filesystems
);
5851 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
5852 c
->syscall_archs
= set_free(c
->syscall_archs
);
5853 c
->address_families
= set_free(c
->address_families
);
5855 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5856 exec_directory_done(&c
->directories
[t
]);
5858 c
->log_level_max
= -1;
5860 exec_context_free_log_extra_fields(c
);
5861 c
->log_filter_allowed_patterns
= set_free(c
->log_filter_allowed_patterns
);
5862 c
->log_filter_denied_patterns
= set_free(c
->log_filter_denied_patterns
);
5864 c
->log_ratelimit_interval_usec
= 0;
5865 c
->log_ratelimit_burst
= 0;
5867 c
->stdin_data
= mfree(c
->stdin_data
);
5868 c
->stdin_data_size
= 0;
5870 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
5871 c
->ipc_namespace_path
= mfree(c
->ipc_namespace_path
);
5873 c
->log_namespace
= mfree(c
->log_namespace
);
5875 c
->load_credentials
= hashmap_free(c
->load_credentials
);
5876 c
->set_credentials
= hashmap_free(c
->set_credentials
);
5878 c
->root_image_policy
= image_policy_free(c
->root_image_policy
);
5879 c
->mount_image_policy
= image_policy_free(c
->mount_image_policy
);
5880 c
->extension_image_policy
= image_policy_free(c
->extension_image_policy
);
5883 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
5886 if (!runtime_prefix
)
5889 for (size_t i
= 0; i
< c
->directories
[EXEC_DIRECTORY_RUNTIME
].n_items
; i
++) {
5890 _cleanup_free_
char *p
= NULL
;
5892 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
5893 p
= path_join(runtime_prefix
, "private", c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].path
);
5895 p
= path_join(runtime_prefix
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].path
);
5899 /* We execute this synchronously, since we need to be sure this is gone when we start the
5901 (void) rm_rf(p
, REMOVE_ROOT
);
5903 STRV_FOREACH(symlink
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].symlinks
) {
5904 _cleanup_free_
char *symlink_abs
= NULL
;
5906 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
5907 symlink_abs
= path_join(runtime_prefix
, "private", *symlink
);
5909 symlink_abs
= path_join(runtime_prefix
, *symlink
);
5913 (void) unlink(symlink_abs
);
5920 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
5921 _cleanup_free_
char *p
= NULL
;
5925 if (!runtime_prefix
|| !unit
)
5928 p
= path_join(runtime_prefix
, "credentials", unit
);
5932 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
5933 * unmount it, and afterwards remove the mount point */
5934 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
5935 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
5940 int exec_context_destroy_mount_ns_dir(Unit
*u
) {
5941 _cleanup_free_
char *p
= NULL
;
5943 if (!u
|| !MANAGER_IS_SYSTEM(u
->manager
))
5946 p
= path_join("/run/systemd/propagate/", u
->id
);
5950 /* This is only filled transiently (see mount_in_namespace()), should be empty or even non-existent*/
5951 if (rmdir(p
) < 0 && errno
!= ENOENT
)
5952 log_unit_debug_errno(u
, errno
, "Unable to remove propagation dir '%s', ignoring: %m", p
);
5957 static void exec_command_done(ExecCommand
*c
) {
5960 c
->path
= mfree(c
->path
);
5961 c
->argv
= strv_free(c
->argv
);
5964 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
5965 for (size_t i
= 0; i
< n
; i
++)
5966 exec_command_done(c
+i
);
5969 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
5973 LIST_REMOVE(command
, c
, i
);
5974 exec_command_done(i
);
5981 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
5982 for (size_t i
= 0; i
< n
; i
++)
5983 c
[i
] = exec_command_free_list(c
[i
]);
5986 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
5987 for (size_t i
= 0; i
< n
; i
++)
5988 exec_status_reset(&c
[i
].exec_status
);
5991 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
5992 for (size_t i
= 0; i
< n
; i
++)
5993 LIST_FOREACH(command
, z
, c
[i
])
5994 exec_status_reset(&z
->exec_status
);
5997 typedef struct InvalidEnvInfo
{
6002 static void invalid_env(const char *p
, void *userdata
) {
6003 InvalidEnvInfo
*info
= userdata
;
6005 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
6008 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
6014 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
6017 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
6020 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
6023 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
6026 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
6029 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
6036 static int exec_context_named_iofds(
6037 const ExecContext
*c
,
6038 const ExecParameters
*p
,
6039 int named_iofds
[static 3]) {
6042 const char* stdio_fdname
[3];
6047 assert(named_iofds
);
6049 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
6050 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
6051 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
6053 for (size_t i
= 0; i
< 3; i
++)
6054 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
6056 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
6058 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
6059 if (named_iofds
[STDIN_FILENO
] < 0 &&
6060 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
6061 stdio_fdname
[STDIN_FILENO
] &&
6062 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
6064 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
6067 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
6068 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
6069 stdio_fdname
[STDOUT_FILENO
] &&
6070 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
6072 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
6075 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
6076 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
6077 stdio_fdname
[STDERR_FILENO
] &&
6078 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
6080 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
6084 return targets
== 0 ? 0 : -ENOENT
;
6087 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***ret
) {
6088 _cleanup_strv_free_
char **v
= NULL
;
6094 STRV_FOREACH(i
, c
->environment_files
) {
6095 _cleanup_globfree_ glob_t pglob
= {};
6096 bool ignore
= false;
6104 if (!path_is_absolute(fn
)) {
6110 /* Filename supports globbing, take all matching files */
6111 r
= safe_glob(fn
, 0, &pglob
);
6118 /* When we don't match anything, -ENOENT should be returned */
6119 assert(pglob
.gl_pathc
> 0);
6121 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
6122 _cleanup_strv_free_
char **p
= NULL
;
6124 r
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
6131 /* Log invalid environment variables with filename */
6133 InvalidEnvInfo info
= {
6135 .path
= pglob
.gl_pathv
[n
]
6138 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
6144 char **m
= strv_env_merge(v
, p
);
6148 strv_free_and_replace(v
, m
);
6158 static bool tty_may_match_dev_console(const char *tty
) {
6159 _cleanup_free_
char *resolved
= NULL
;
6164 tty
= skip_dev_prefix(tty
);
6166 /* trivial identity? */
6167 if (streq(tty
, "console"))
6170 if (resolve_dev_console(&resolved
) < 0)
6171 return true; /* if we could not resolve, assume it may */
6173 /* "tty0" means the active VC, so it may be the same sometimes */
6174 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
6177 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
6180 return ec
->tty_reset
||
6182 ec
->tty_vt_disallocate
||
6183 is_terminal_input(ec
->std_input
) ||
6184 is_terminal_output(ec
->std_output
) ||
6185 is_terminal_output(ec
->std_error
);
6188 bool exec_context_may_touch_console(const ExecContext
*ec
) {
6190 return exec_context_may_touch_tty(ec
) &&
6191 tty_may_match_dev_console(exec_context_tty_path(ec
));
6194 static void strv_fprintf(FILE *f
, char **l
) {
6198 fprintf(f
, " %s", *g
);
6201 static void strv_dump(FILE* f
, const char *prefix
, const char *name
, char **strv
) {
6206 if (!strv_isempty(strv
)) {
6207 fprintf(f
, "%s%s:", prefix
, name
);
6208 strv_fprintf(f
, strv
);
6213 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
6219 prefix
= strempty(prefix
);
6223 "%sWorkingDirectory: %s\n"
6224 "%sRootDirectory: %s\n"
6225 "%sNonBlocking: %s\n"
6226 "%sPrivateTmp: %s\n"
6227 "%sPrivateDevices: %s\n"
6228 "%sProtectKernelTunables: %s\n"
6229 "%sProtectKernelModules: %s\n"
6230 "%sProtectKernelLogs: %s\n"
6231 "%sProtectClock: %s\n"
6232 "%sProtectControlGroups: %s\n"
6233 "%sPrivateNetwork: %s\n"
6234 "%sPrivateUsers: %s\n"
6235 "%sProtectHome: %s\n"
6236 "%sProtectSystem: %s\n"
6237 "%sMountAPIVFS: %s\n"
6238 "%sIgnoreSIGPIPE: %s\n"
6239 "%sMemoryDenyWriteExecute: %s\n"
6240 "%sRestrictRealtime: %s\n"
6241 "%sRestrictSUIDSGID: %s\n"
6242 "%sKeyringMode: %s\n"
6243 "%sProtectHostname: %s\n"
6244 "%sProtectProc: %s\n"
6245 "%sProcSubset: %s\n",
6247 prefix
, empty_to_root(c
->working_directory
),
6248 prefix
, empty_to_root(c
->root_directory
),
6249 prefix
, yes_no(c
->non_blocking
),
6250 prefix
, yes_no(c
->private_tmp
),
6251 prefix
, yes_no(c
->private_devices
),
6252 prefix
, yes_no(c
->protect_kernel_tunables
),
6253 prefix
, yes_no(c
->protect_kernel_modules
),
6254 prefix
, yes_no(c
->protect_kernel_logs
),
6255 prefix
, yes_no(c
->protect_clock
),
6256 prefix
, yes_no(c
->protect_control_groups
),
6257 prefix
, yes_no(c
->private_network
),
6258 prefix
, yes_no(c
->private_users
),
6259 prefix
, protect_home_to_string(c
->protect_home
),
6260 prefix
, protect_system_to_string(c
->protect_system
),
6261 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
6262 prefix
, yes_no(c
->ignore_sigpipe
),
6263 prefix
, yes_no(c
->memory_deny_write_execute
),
6264 prefix
, yes_no(c
->restrict_realtime
),
6265 prefix
, yes_no(c
->restrict_suid_sgid
),
6266 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
6267 prefix
, yes_no(c
->protect_hostname
),
6268 prefix
, protect_proc_to_string(c
->protect_proc
),
6269 prefix
, proc_subset_to_string(c
->proc_subset
));
6272 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
6274 if (c
->root_image_options
) {
6275 fprintf(f
, "%sRootImageOptions:", prefix
);
6276 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
6277 if (!isempty(o
->options
))
6278 fprintf(f
, " %s:%s",
6279 partition_designator_to_string(o
->partition_designator
),
6285 _cleanup_free_
char *encoded
= NULL
;
6286 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
6288 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
6291 if (c
->root_hash_path
)
6292 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
6294 if (c
->root_hash_sig
) {
6295 _cleanup_free_
char *encoded
= NULL
;
6297 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
6299 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
6302 if (c
->root_hash_sig_path
)
6303 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
6306 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
6308 STRV_FOREACH(e
, c
->environment
)
6309 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
6311 STRV_FOREACH(e
, c
->environment_files
)
6312 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
6314 STRV_FOREACH(e
, c
->pass_environment
)
6315 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
6317 STRV_FOREACH(e
, c
->unset_environment
)
6318 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
6320 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
6322 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
6323 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
6325 for (size_t i
= 0; i
< c
->directories
[dt
].n_items
; i
++) {
6326 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].items
[i
].path
);
6328 STRV_FOREACH(d
, c
->directories
[dt
].items
[i
].symlinks
)
6329 fprintf(f
, "%s%s: %s:%s\n", prefix
, exec_directory_type_symlink_to_string(dt
), c
->directories
[dt
].items
[i
].path
, *d
);
6333 fprintf(f
, "%sTimeoutCleanSec: %s\n", prefix
, FORMAT_TIMESPAN(c
->timeout_clean_usec
, USEC_PER_SEC
));
6336 fprintf(f
, "%sNice: %i\n", prefix
, c
->nice
);
6338 if (c
->oom_score_adjust_set
)
6339 fprintf(f
, "%sOOMScoreAdjust: %i\n", prefix
, c
->oom_score_adjust
);
6341 if (c
->coredump_filter_set
)
6342 fprintf(f
, "%sCoredumpFilter: 0x%"PRIx64
"\n", prefix
, c
->coredump_filter
);
6344 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
6346 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
6347 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
6348 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
6349 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
6352 if (c
->ioprio_set
) {
6353 _cleanup_free_
char *class_str
= NULL
;
6355 r
= ioprio_class_to_string_alloc(ioprio_prio_class(c
->ioprio
), &class_str
);
6357 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
6359 fprintf(f
, "%sIOPriority: %d\n", prefix
, ioprio_prio_data(c
->ioprio
));
6362 if (c
->cpu_sched_set
) {
6363 _cleanup_free_
char *policy_str
= NULL
;
6365 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
6367 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
6370 "%sCPUSchedulingPriority: %i\n"
6371 "%sCPUSchedulingResetOnFork: %s\n",
6372 prefix
, c
->cpu_sched_priority
,
6373 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
6376 if (c
->cpu_set
.set
) {
6377 _cleanup_free_
char *affinity
= NULL
;
6379 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
6380 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
6383 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
6384 _cleanup_free_
char *nodes
= NULL
;
6386 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
6387 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
6388 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
6391 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
6392 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
6395 "%sStandardInput: %s\n"
6396 "%sStandardOutput: %s\n"
6397 "%sStandardError: %s\n",
6398 prefix
, exec_input_to_string(c
->std_input
),
6399 prefix
, exec_output_to_string(c
->std_output
),
6400 prefix
, exec_output_to_string(c
->std_error
));
6402 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
6403 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
6404 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
6405 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
6406 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
6407 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
6409 if (c
->std_input
== EXEC_INPUT_FILE
)
6410 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
6411 if (c
->std_output
== EXEC_OUTPUT_FILE
)
6412 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
6413 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
6414 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
6415 if (c
->std_output
== EXEC_OUTPUT_FILE_TRUNCATE
)
6416 fprintf(f
, "%sStandardOutputFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
6417 if (c
->std_error
== EXEC_OUTPUT_FILE
)
6418 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
6419 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
6420 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
6421 if (c
->std_error
== EXEC_OUTPUT_FILE_TRUNCATE
)
6422 fprintf(f
, "%sStandardErrorFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
6428 "%sTTYVHangup: %s\n"
6429 "%sTTYVTDisallocate: %s\n"
6431 "%sTTYColumns: %u\n",
6432 prefix
, c
->tty_path
,
6433 prefix
, yes_no(c
->tty_reset
),
6434 prefix
, yes_no(c
->tty_vhangup
),
6435 prefix
, yes_no(c
->tty_vt_disallocate
),
6436 prefix
, c
->tty_rows
,
6437 prefix
, c
->tty_cols
);
6439 if (IN_SET(c
->std_output
,
6441 EXEC_OUTPUT_JOURNAL
,
6442 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
6443 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
6444 IN_SET(c
->std_error
,
6446 EXEC_OUTPUT_JOURNAL
,
6447 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
6448 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
6450 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
6452 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
6454 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
6456 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
6458 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
6461 if (c
->log_level_max
>= 0) {
6462 _cleanup_free_
char *t
= NULL
;
6464 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
6466 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
6469 if (c
->log_ratelimit_interval_usec
> 0)
6471 "%sLogRateLimitIntervalSec: %s\n",
6472 prefix
, FORMAT_TIMESPAN(c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
6474 if (c
->log_ratelimit_burst
> 0)
6475 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
6477 if (!set_isempty(c
->log_filter_allowed_patterns
) || !set_isempty(c
->log_filter_denied_patterns
)) {
6478 fprintf(f
, "%sLogFilterPatterns:", prefix
);
6481 SET_FOREACH(pattern
, c
->log_filter_allowed_patterns
)
6482 fprintf(f
, " %s", pattern
);
6483 SET_FOREACH(pattern
, c
->log_filter_denied_patterns
)
6484 fprintf(f
, " ~%s", pattern
);
6488 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
6489 fprintf(f
, "%sLogExtraFields: ", prefix
);
6490 fwrite(c
->log_extra_fields
[j
].iov_base
,
6491 1, c
->log_extra_fields
[j
].iov_len
,
6496 if (c
->log_namespace
)
6497 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
6499 if (c
->secure_bits
) {
6500 _cleanup_free_
char *str
= NULL
;
6502 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
6504 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
6507 if (c
->capability_bounding_set
!= CAP_MASK_UNSET
) {
6508 _cleanup_free_
char *str
= NULL
;
6510 r
= capability_set_to_string(c
->capability_bounding_set
, &str
);
6512 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
6515 if (c
->capability_ambient_set
!= 0) {
6516 _cleanup_free_
char *str
= NULL
;
6518 r
= capability_set_to_string(c
->capability_ambient_set
, &str
);
6520 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
6524 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
6526 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
6528 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
6530 strv_dump(f
, prefix
, "SupplementaryGroups", c
->supplementary_groups
);
6533 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
6535 strv_dump(f
, prefix
, "ReadWritePaths", c
->read_write_paths
);
6536 strv_dump(f
, prefix
, "ReadOnlyPaths", c
->read_only_paths
);
6537 strv_dump(f
, prefix
, "InaccessiblePaths", c
->inaccessible_paths
);
6538 strv_dump(f
, prefix
, "ExecPaths", c
->exec_paths
);
6539 strv_dump(f
, prefix
, "NoExecPaths", c
->no_exec_paths
);
6540 strv_dump(f
, prefix
, "ExecSearchPath", c
->exec_search_path
);
6542 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
6543 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
6544 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
6545 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
6546 c
->bind_mounts
[i
].source
,
6547 c
->bind_mounts
[i
].destination
,
6548 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
6550 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
6551 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
6553 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
6555 isempty(t
->options
) ? "" : ":",
6556 strempty(t
->options
));
6561 "%sUtmpIdentifier: %s\n",
6562 prefix
, c
->utmp_id
);
6564 if (c
->selinux_context
)
6566 "%sSELinuxContext: %s%s\n",
6567 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
6569 if (c
->apparmor_profile
)
6571 "%sAppArmorProfile: %s%s\n",
6572 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
6574 if (c
->smack_process_label
)
6576 "%sSmackProcessLabel: %s%s\n",
6577 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
6579 if (c
->personality
!= PERSONALITY_INVALID
)
6581 "%sPersonality: %s\n",
6582 prefix
, strna(personality_to_string(c
->personality
)));
6585 "%sLockPersonality: %s\n",
6586 prefix
, yes_no(c
->lock_personality
));
6588 if (c
->syscall_filter
) {
6590 "%sSystemCallFilter: ",
6593 if (!c
->syscall_allow_list
)
6599 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
6600 _cleanup_free_
char *name
= NULL
;
6601 const char *errno_name
= NULL
;
6602 int num
= PTR_TO_INT(val
);
6609 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
6610 fputs(strna(name
), f
);
6613 errno_name
= seccomp_errno_or_action_to_string(num
);
6615 fprintf(f
, ":%s", errno_name
);
6617 fprintf(f
, ":%d", num
);
6625 if (c
->syscall_archs
) {
6627 "%sSystemCallArchitectures:",
6632 SET_FOREACH(id
, c
->syscall_archs
)
6633 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
6638 if (exec_context_restrict_namespaces_set(c
)) {
6639 _cleanup_free_
char *s
= NULL
;
6641 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
6643 fprintf(f
, "%sRestrictNamespaces: %s\n",
6648 if (exec_context_restrict_filesystems_set(c
)) {
6650 SET_FOREACH(fs
, c
->restrict_filesystems
)
6651 fprintf(f
, "%sRestrictFileSystems: %s\n", prefix
, fs
);
6655 if (c
->network_namespace_path
)
6657 "%sNetworkNamespacePath: %s\n",
6658 prefix
, c
->network_namespace_path
);
6660 if (c
->syscall_errno
> 0) {
6661 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
6664 const char *errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
6666 fputs(errno_name
, f
);
6668 fprintf(f
, "%d", c
->syscall_errno
);
6673 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
6674 fprintf(f
, "%sMountImages: %s%s:%s", prefix
,
6675 c
->mount_images
[i
].ignore_enoent
? "-": "",
6676 c
->mount_images
[i
].source
,
6677 c
->mount_images
[i
].destination
);
6678 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
6679 fprintf(f
, ":%s:%s",
6680 partition_designator_to_string(o
->partition_designator
),
6681 strempty(o
->options
));
6685 for (size_t i
= 0; i
< c
->n_extension_images
; i
++) {
6686 fprintf(f
, "%sExtensionImages: %s%s", prefix
,
6687 c
->extension_images
[i
].ignore_enoent
? "-": "",
6688 c
->extension_images
[i
].source
);
6689 LIST_FOREACH(mount_options
, o
, c
->extension_images
[i
].mount_options
)
6690 fprintf(f
, ":%s:%s",
6691 partition_designator_to_string(o
->partition_designator
),
6692 strempty(o
->options
));
6696 strv_dump(f
, prefix
, "ExtensionDirectories", c
->extension_directories
);
6699 bool exec_context_maintains_privileges(const ExecContext
*c
) {
6702 /* Returns true if the process forked off would run under
6703 * an unchanged UID or as root. */
6708 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
6714 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
6722 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
6724 return IOPRIO_DEFAULT_CLASS_AND_PRIO
;
6726 return ioprio_normalize(p
);
6729 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
6732 /* Explicit setting wins */
6733 if (c
->mount_apivfs_set
)
6734 return c
->mount_apivfs
;
6736 /* Default to "yes" if root directory or image are specified */
6737 if (exec_context_with_rootfs(c
))
6743 void exec_context_free_log_extra_fields(ExecContext
*c
) {
6746 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
6747 free(c
->log_extra_fields
[l
].iov_base
);
6748 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
6749 c
->n_log_extra_fields
= 0;
6752 void exec_context_revert_tty(ExecContext
*c
) {
6753 _cleanup_close_
int fd
= -EBADF
;
6760 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
6761 exec_context_tty_reset(c
, NULL
);
6763 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
6764 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
6765 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
6766 if (!exec_context_may_touch_tty(c
))
6769 path
= exec_context_tty_path(c
);
6773 fd
= open(path
, O_PATH
|O_CLOEXEC
);
6775 return (void) log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
6776 "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m",
6779 if (fstat(fd
, &st
) < 0)
6780 return (void) log_warning_errno(errno
, "Failed to stat TTY '%s', ignoring: %m", path
);
6782 /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check
6783 * if things are a character device, since a proper check either means we'd have to open the TTY and
6784 * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects
6785 * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother
6786 * with this at all? → https://github.com/systemd/systemd/issues/19213 */
6787 if (!S_ISCHR(st
.st_mode
))
6788 return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path
);
6790 r
= fchmod_and_chown(fd
, TTY_MODE
, 0, TTY_GID
);
6792 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
6795 int exec_context_get_clean_directories(
6801 _cleanup_strv_free_
char **l
= NULL
;
6808 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
6809 if (!FLAGS_SET(mask
, 1U << t
))
6815 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
6818 j
= path_join(prefix
[t
], c
->directories
[t
].items
[i
].path
);
6822 r
= strv_consume(&l
, j
);
6826 /* Also remove private directories unconditionally. */
6827 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
6828 j
= path_join(prefix
[t
], "private", c
->directories
[t
].items
[i
].path
);
6832 r
= strv_consume(&l
, j
);
6837 STRV_FOREACH(symlink
, c
->directories
[t
].items
[i
].symlinks
) {
6838 j
= path_join(prefix
[t
], *symlink
);
6842 r
= strv_consume(&l
, j
);
6853 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
6854 ExecCleanMask mask
= 0;
6859 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
6860 if (c
->directories
[t
].n_items
> 0)
6867 bool exec_context_has_encrypted_credentials(ExecContext
*c
) {
6868 ExecLoadCredential
*load_cred
;
6869 ExecSetCredential
*set_cred
;
6873 HASHMAP_FOREACH(load_cred
, c
->load_credentials
)
6874 if (load_cred
->encrypted
)
6877 HASHMAP_FOREACH(set_cred
, c
->set_credentials
)
6878 if (set_cred
->encrypted
)
6884 int exec_context_add_default_dependencies(Unit
*u
, const ExecContext
*c
) {
6886 assert(u
->default_dependencies
);
6888 if (c
&& exec_context_needs_term(c
))
6889 return unit_add_dependency_by_name(u
, UNIT_AFTER
, SPECIAL_VCONSOLE_SETUP_SERVICE
,
6890 /* add_reference= */ true, UNIT_DEPENDENCY_DEFAULT
);
6894 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
6901 dual_timestamp_get(&s
->start_timestamp
);
6904 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
6912 dual_timestamp_get(&s
->exit_timestamp
);
6917 if (context
&& context
->utmp_id
)
6918 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
6921 void exec_status_reset(ExecStatus
*s
) {
6924 *s
= (ExecStatus
) {};
6927 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
6934 prefix
= strempty(prefix
);
6937 "%sPID: "PID_FMT
"\n",
6940 if (dual_timestamp_is_set(&s
->start_timestamp
))
6942 "%sStart Timestamp: %s\n",
6943 prefix
, FORMAT_TIMESTAMP(s
->start_timestamp
.realtime
));
6945 if (dual_timestamp_is_set(&s
->exit_timestamp
))
6947 "%sExit Timestamp: %s\n"
6949 "%sExit Status: %i\n",
6950 prefix
, FORMAT_TIMESTAMP(s
->exit_timestamp
.realtime
),
6951 prefix
, sigchld_code_to_string(s
->code
),
6955 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6956 _cleanup_free_
char *cmd
= NULL
;
6957 const char *prefix2
;
6962 prefix
= strempty(prefix
);
6963 prefix2
= strjoina(prefix
, "\t");
6965 cmd
= quote_command_line(c
->argv
, SHELL_ESCAPE_EMPTY
);
6968 "%sCommand Line: %s\n",
6969 prefix
, strnull(cmd
));
6971 exec_status_dump(&c
->exec_status
, f
, prefix2
);
6974 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6977 prefix
= strempty(prefix
);
6979 LIST_FOREACH(command
, i
, c
)
6980 exec_command_dump(i
, f
, prefix
);
6983 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
6990 /* It's kind of important, that we keep the order here */
6991 end
= LIST_FIND_TAIL(command
, *l
);
6992 LIST_INSERT_AFTER(command
, *l
, end
, e
);
6997 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
7005 l
= strv_new_ap(path
, ap
);
7017 free_and_replace(c
->path
, p
);
7019 return strv_free_and_replace(c
->argv
, l
);
7022 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
7023 _cleanup_strv_free_
char **l
= NULL
;
7031 l
= strv_new_ap(path
, ap
);
7037 r
= strv_extend_strv(&c
->argv
, l
, false);
7044 static void *remove_tmpdir_thread(void *p
) {
7045 _cleanup_free_
char *path
= p
;
7047 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
7051 static ExecSharedRuntime
* exec_shared_runtime_free(ExecSharedRuntime
*rt
) {
7056 (void) hashmap_remove(rt
->manager
->exec_shared_runtime_by_id
, rt
->id
);
7058 rt
->id
= mfree(rt
->id
);
7059 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
7060 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
7061 safe_close_pair(rt
->netns_storage_socket
);
7062 safe_close_pair(rt
->ipcns_storage_socket
);
7066 DEFINE_TRIVIAL_UNREF_FUNC(ExecSharedRuntime
, exec_shared_runtime
, exec_shared_runtime_free
);
7067 DEFINE_TRIVIAL_CLEANUP_FUNC(ExecSharedRuntime
*, exec_shared_runtime_free
);
7069 ExecSharedRuntime
* exec_shared_runtime_destroy(ExecSharedRuntime
*rt
) {
7075 assert(rt
->n_ref
> 0);
7081 if (rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
7082 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
7084 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
7086 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
7091 if (rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
7092 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
7094 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
7096 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
7098 rt
->var_tmp_dir
= NULL
;
7101 return exec_shared_runtime_free(rt
);
7104 static int exec_shared_runtime_allocate(ExecSharedRuntime
**ret
, const char *id
) {
7105 _cleanup_free_
char *id_copy
= NULL
;
7106 ExecSharedRuntime
*n
;
7110 id_copy
= strdup(id
);
7114 n
= new(ExecSharedRuntime
, 1);
7118 *n
= (ExecSharedRuntime
) {
7119 .id
= TAKE_PTR(id_copy
),
7120 .netns_storage_socket
= PIPE_EBADF
,
7121 .ipcns_storage_socket
= PIPE_EBADF
,
7128 static int exec_shared_runtime_add(
7133 int netns_storage_socket
[2],
7134 int ipcns_storage_socket
[2],
7135 ExecSharedRuntime
**ret
) {
7137 _cleanup_(exec_shared_runtime_freep
) ExecSharedRuntime
*rt
= NULL
;
7143 /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */
7145 r
= exec_shared_runtime_allocate(&rt
, id
);
7149 r
= hashmap_ensure_put(&m
->exec_shared_runtime_by_id
, &string_hash_ops
, rt
->id
, rt
);
7153 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
7154 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
7155 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
7157 if (netns_storage_socket
) {
7158 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
7159 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
7162 if (ipcns_storage_socket
) {
7163 rt
->ipcns_storage_socket
[0] = TAKE_FD(ipcns_storage_socket
[0]);
7164 rt
->ipcns_storage_socket
[1] = TAKE_FD(ipcns_storage_socket
[1]);
7171 /* do not remove created ExecSharedRuntime object when the operation succeeds. */
7176 static int exec_shared_runtime_make(
7178 const ExecContext
*c
,
7180 ExecSharedRuntime
**ret
) {
7182 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
7183 _cleanup_close_pair_
int netns_storage_socket
[2] = PIPE_EBADF
, ipcns_storage_socket
[2] = PIPE_EBADF
;
7190 /* It is not necessary to create ExecSharedRuntime object. */
7191 if (!exec_needs_network_namespace(c
) && !exec_needs_ipc_namespace(c
) && !c
->private_tmp
) {
7196 if (c
->private_tmp
&&
7197 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
7198 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
7199 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
7200 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
7205 if (exec_needs_network_namespace(c
)) {
7206 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
7210 if (exec_needs_ipc_namespace(c
)) {
7211 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, ipcns_storage_socket
) < 0)
7215 r
= exec_shared_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ipcns_storage_socket
, ret
);
7222 int exec_shared_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecSharedRuntime
**ret
) {
7223 ExecSharedRuntime
*rt
;
7230 rt
= hashmap_get(m
->exec_shared_runtime_by_id
, id
);
7232 /* We already have an ExecSharedRuntime object, let's increase the ref count and reuse it */
7240 /* If not found, then create a new object. */
7241 r
= exec_shared_runtime_make(m
, c
, id
, &rt
);
7245 /* When r == 0, it is not necessary to create ExecSharedRuntime object. */
7251 /* increment reference counter. */
7257 int exec_shared_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
7258 ExecSharedRuntime
*rt
;
7264 HASHMAP_FOREACH(rt
, m
->exec_shared_runtime_by_id
) {
7265 fprintf(f
, "exec-runtime=%s", rt
->id
);
7268 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
7270 if (rt
->var_tmp_dir
)
7271 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
7273 if (rt
->netns_storage_socket
[0] >= 0) {
7276 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
7280 fprintf(f
, " netns-socket-0=%i", copy
);
7283 if (rt
->netns_storage_socket
[1] >= 0) {
7286 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
7290 fprintf(f
, " netns-socket-1=%i", copy
);
7293 if (rt
->ipcns_storage_socket
[0] >= 0) {
7296 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[0]);
7300 fprintf(f
, " ipcns-socket-0=%i", copy
);
7303 if (rt
->ipcns_storage_socket
[1] >= 0) {
7306 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[1]);
7310 fprintf(f
, " ipcns-socket-1=%i", copy
);
7319 int exec_shared_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
7320 _cleanup_(exec_shared_runtime_freep
) ExecSharedRuntime
*rt_create
= NULL
;
7321 ExecSharedRuntime
*rt
;
7324 /* This is for the migration from old (v237 or earlier) deserialization text.
7325 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
7326 * Even if the ExecSharedRuntime object originally created by the other unit, we cannot judge
7327 * so or not from the serialized text, then we always creates a new object owned by this. */
7333 /* Manager manages ExecSharedRuntime objects by the unit id.
7334 * So, we omit the serialized text when the unit does not have id (yet?)... */
7335 if (isempty(u
->id
)) {
7336 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
7340 if (hashmap_ensure_allocated(&u
->manager
->exec_shared_runtime_by_id
, &string_hash_ops
) < 0)
7343 rt
= hashmap_get(u
->manager
->exec_shared_runtime_by_id
, u
->id
);
7345 if (exec_shared_runtime_allocate(&rt_create
, u
->id
) < 0)
7351 if (streq(key
, "tmp-dir")) {
7352 if (free_and_strdup_warn(&rt
->tmp_dir
, value
) < 0)
7355 } else if (streq(key
, "var-tmp-dir")) {
7356 if (free_and_strdup_warn(&rt
->var_tmp_dir
, value
) < 0)
7359 } else if (streq(key
, "netns-socket-0")) {
7362 if ((fd
= parse_fd(value
)) < 0 || !fdset_contains(fds
, fd
)) {
7363 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
7367 safe_close(rt
->netns_storage_socket
[0]);
7368 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
7370 } else if (streq(key
, "netns-socket-1")) {
7373 if ((fd
= parse_fd(value
)) < 0 || !fdset_contains(fds
, fd
)) {
7374 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
7378 safe_close(rt
->netns_storage_socket
[1]);
7379 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
7384 /* If the object is newly created, then put it to the hashmap which manages ExecSharedRuntime objects. */
7386 r
= hashmap_put(u
->manager
->exec_shared_runtime_by_id
, rt_create
->id
, rt_create
);
7388 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
7392 rt_create
->manager
= u
->manager
;
7395 TAKE_PTR(rt_create
);
7401 int exec_shared_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
7402 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
7404 int r
, netns_fdpair
[] = {-1, -1}, ipcns_fdpair
[] = {-1, -1};
7405 const char *p
, *v
= ASSERT_PTR(value
);
7411 n
= strcspn(v
, " ");
7412 id
= strndupa_safe(v
, n
);
7417 v
= startswith(p
, "tmp-dir=");
7419 n
= strcspn(v
, " ");
7420 tmp_dir
= strndup(v
, n
);
7428 v
= startswith(p
, "var-tmp-dir=");
7430 n
= strcspn(v
, " ");
7431 var_tmp_dir
= strndup(v
, n
);
7439 v
= startswith(p
, "netns-socket-0=");
7443 n
= strcspn(v
, " ");
7444 buf
= strndupa_safe(v
, n
);
7446 netns_fdpair
[0] = parse_fd(buf
);
7447 if (netns_fdpair
[0] < 0)
7448 return log_debug_errno(netns_fdpair
[0], "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
7449 if (!fdset_contains(fds
, netns_fdpair
[0]))
7450 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7451 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair
[0]);
7452 netns_fdpair
[0] = fdset_remove(fds
, netns_fdpair
[0]);
7458 v
= startswith(p
, "netns-socket-1=");
7462 n
= strcspn(v
, " ");
7463 buf
= strndupa_safe(v
, n
);
7465 netns_fdpair
[1] = parse_fd(buf
);
7466 if (netns_fdpair
[1] < 0)
7467 return log_debug_errno(netns_fdpair
[1], "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
7468 if (!fdset_contains(fds
, netns_fdpair
[1]))
7469 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7470 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair
[1]);
7471 netns_fdpair
[1] = fdset_remove(fds
, netns_fdpair
[1]);
7477 v
= startswith(p
, "ipcns-socket-0=");
7481 n
= strcspn(v
, " ");
7482 buf
= strndupa_safe(v
, n
);
7484 ipcns_fdpair
[0] = parse_fd(buf
);
7485 if (ipcns_fdpair
[0] < 0)
7486 return log_debug_errno(ipcns_fdpair
[0], "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf
);
7487 if (!fdset_contains(fds
, ipcns_fdpair
[0]))
7488 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7489 "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair
[0]);
7490 ipcns_fdpair
[0] = fdset_remove(fds
, ipcns_fdpair
[0]);
7496 v
= startswith(p
, "ipcns-socket-1=");
7500 n
= strcspn(v
, " ");
7501 buf
= strndupa_safe(v
, n
);
7503 ipcns_fdpair
[1] = parse_fd(buf
);
7504 if (ipcns_fdpair
[1] < 0)
7505 return log_debug_errno(ipcns_fdpair
[1], "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf
);
7506 if (!fdset_contains(fds
, ipcns_fdpair
[1]))
7507 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7508 "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair
[1]);
7509 ipcns_fdpair
[1] = fdset_remove(fds
, ipcns_fdpair
[1]);
7513 r
= exec_shared_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_fdpair
, ipcns_fdpair
, NULL
);
7515 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
7519 void exec_shared_runtime_vacuum(Manager
*m
) {
7520 ExecSharedRuntime
*rt
;
7524 /* Free unreferenced ExecSharedRuntime objects. This is used after manager deserialization process. */
7526 HASHMAP_FOREACH(rt
, m
->exec_shared_runtime_by_id
) {
7530 (void) exec_shared_runtime_free(rt
);
7534 int exec_runtime_make(ExecSharedRuntime
*shared
, DynamicCreds
*creds
, ExecRuntime
**ret
) {
7535 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
7539 if (!shared
&& !creds
) {
7544 rt
= new(ExecRuntime
, 1);
7548 *rt
= (ExecRuntime
) {
7550 .dynamic_creds
= creds
,
7553 *ret
= TAKE_PTR(rt
);
7557 ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
) {
7561 exec_shared_runtime_unref(rt
->shared
);
7562 dynamic_creds_unref(rt
->dynamic_creds
);
7566 ExecRuntime
* exec_runtime_destroy(ExecRuntime
*rt
) {
7570 rt
->shared
= exec_shared_runtime_destroy(rt
->shared
);
7571 rt
->dynamic_creds
= dynamic_creds_destroy(rt
->dynamic_creds
);
7572 return exec_runtime_free(rt
);
7575 void exec_params_clear(ExecParameters
*p
) {
7579 p
->environment
= strv_free(p
->environment
);
7580 p
->fd_names
= strv_free(p
->fd_names
);
7581 p
->fds
= mfree(p
->fds
);
7582 p
->exec_fd
= safe_close(p
->exec_fd
);
7585 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
7594 ExecLoadCredential
*exec_load_credential_free(ExecLoadCredential
*lc
) {
7603 void exec_directory_done(ExecDirectory
*d
) {
7607 for (size_t i
= 0; i
< d
->n_items
; i
++) {
7608 free(d
->items
[i
].path
);
7609 strv_free(d
->items
[i
].symlinks
);
7612 d
->items
= mfree(d
->items
);
7617 static ExecDirectoryItem
*exec_directory_find(ExecDirectory
*d
, const char *path
) {
7621 for (size_t i
= 0; i
< d
->n_items
; i
++)
7622 if (path_equal(d
->items
[i
].path
, path
))
7623 return &d
->items
[i
];
7628 int exec_directory_add(ExecDirectory
*d
, const char *path
, const char *symlink
) {
7629 _cleanup_strv_free_
char **s
= NULL
;
7630 _cleanup_free_
char *p
= NULL
;
7631 ExecDirectoryItem
*existing
;
7637 existing
= exec_directory_find(d
, path
);
7639 r
= strv_extend(&existing
->symlinks
, symlink
);
7643 return 0; /* existing item is updated */
7651 s
= strv_new(symlink
);
7656 if (!GREEDY_REALLOC(d
->items
, d
->n_items
+ 1))
7659 d
->items
[d
->n_items
++] = (ExecDirectoryItem
) {
7660 .path
= TAKE_PTR(p
),
7661 .symlinks
= TAKE_PTR(s
),
7664 return 1; /* new item is added */
7667 static int exec_directory_item_compare_func(const ExecDirectoryItem
*a
, const ExecDirectoryItem
*b
) {
7671 return path_compare(a
->path
, b
->path
);
7674 void exec_directory_sort(ExecDirectory
*d
) {
7677 /* Sort the exec directories to make always parent directories processed at first in
7678 * setup_exec_directory(), e.g., even if StateDirectory=foo/bar foo, we need to create foo at first,
7679 * then foo/bar. Also, set .only_create flag if one of the parent directories is contained in the
7680 * list. See also comments in setup_exec_directory() and issue #24783. */
7682 if (d
->n_items
<= 1)
7685 typesafe_qsort(d
->items
, d
->n_items
, exec_directory_item_compare_func
);
7687 for (size_t i
= 1; i
< d
->n_items
; i
++)
7688 for (size_t j
= 0; j
< i
; j
++)
7689 if (path_startswith(d
->items
[i
].path
, d
->items
[j
].path
)) {
7690 d
->items
[i
].only_create
= true;
7695 ExecCleanMask
exec_clean_mask_from_string(const char *s
) {
7696 ExecDirectoryType t
;
7700 if (streq(s
, "all"))
7701 return EXEC_CLEAN_ALL
;
7702 if (streq(s
, "fdstore"))
7703 return EXEC_CLEAN_FDSTORE
;
7705 t
= exec_resource_type_from_string(s
);
7707 return (ExecCleanMask
) t
;
7712 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
7713 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_load_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecLoadCredential
, exec_load_credential_free
);
7715 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
7716 [EXEC_INPUT_NULL
] = "null",
7717 [EXEC_INPUT_TTY
] = "tty",
7718 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
7719 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
7720 [EXEC_INPUT_SOCKET
] = "socket",
7721 [EXEC_INPUT_NAMED_FD
] = "fd",
7722 [EXEC_INPUT_DATA
] = "data",
7723 [EXEC_INPUT_FILE
] = "file",
7726 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
7728 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
7729 [EXEC_OUTPUT_INHERIT
] = "inherit",
7730 [EXEC_OUTPUT_NULL
] = "null",
7731 [EXEC_OUTPUT_TTY
] = "tty",
7732 [EXEC_OUTPUT_KMSG
] = "kmsg",
7733 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
7734 [EXEC_OUTPUT_JOURNAL
] = "journal",
7735 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
7736 [EXEC_OUTPUT_SOCKET
] = "socket",
7737 [EXEC_OUTPUT_NAMED_FD
] = "fd",
7738 [EXEC_OUTPUT_FILE
] = "file",
7739 [EXEC_OUTPUT_FILE_APPEND
] = "append",
7740 [EXEC_OUTPUT_FILE_TRUNCATE
] = "truncate",
7743 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
7745 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
7746 [EXEC_UTMP_INIT
] = "init",
7747 [EXEC_UTMP_LOGIN
] = "login",
7748 [EXEC_UTMP_USER
] = "user",
7751 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
7753 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
7754 [EXEC_PRESERVE_NO
] = "no",
7755 [EXEC_PRESERVE_YES
] = "yes",
7756 [EXEC_PRESERVE_RESTART
] = "restart",
7759 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
7761 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
7762 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7763 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
7764 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
7765 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
7766 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
7767 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
7770 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
7772 /* This table maps ExecDirectoryType to the symlink setting it is configured with in the unit */
7773 static const char* const exec_directory_type_symlink_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7774 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectorySymlink",
7775 [EXEC_DIRECTORY_STATE
] = "StateDirectorySymlink",
7776 [EXEC_DIRECTORY_CACHE
] = "CacheDirectorySymlink",
7777 [EXEC_DIRECTORY_LOGS
] = "LogsDirectorySymlink",
7778 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectorySymlink",
7781 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type_symlink
, ExecDirectoryType
);
7783 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
7784 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
7785 * directories, specifically .timer units with their timestamp touch file. */
7786 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7787 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
7788 [EXEC_DIRECTORY_STATE
] = "state",
7789 [EXEC_DIRECTORY_CACHE
] = "cache",
7790 [EXEC_DIRECTORY_LOGS
] = "logs",
7791 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
7794 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
7796 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
7797 * the service payload in. */
7798 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7799 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
7800 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
7801 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
7802 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
7803 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
7806 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
7808 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
7809 [EXEC_KEYRING_INHERIT
] = "inherit",
7810 [EXEC_KEYRING_PRIVATE
] = "private",
7811 [EXEC_KEYRING_SHARED
] = "shared",
7814 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);