1 /* SPDX-License-Identifier: LGPL-2.1+ */
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"
45 #include "capability-util.h"
46 #include "cgroup-setup.h"
47 #include "chown-recursive.h"
48 #include "cpu-set-util.h"
52 #include "errno-list.h"
54 #include "exit-status.h"
57 #include "format-util.h"
59 #include "glob-util.h"
60 #include "hexdecoct.h"
67 #include "memory-util.h"
68 #include "missing_fs.h"
70 #include "mount-util.h"
71 #include "mountpoint-util.h"
72 #include "namespace.h"
73 #include "parse-util.h"
74 #include "path-util.h"
75 #include "process-util.h"
76 #include "rlimit-util.h"
79 #include "seccomp-util.h"
81 #include "securebits-util.h"
82 #include "selinux-util.h"
83 #include "signal-util.h"
84 #include "smack-util.h"
85 #include "socket-util.h"
87 #include "stat-util.h"
88 #include "string-table.h"
89 #include "string-util.h"
91 #include "syslog-util.h"
92 #include "terminal-util.h"
93 #include "tmpfile-util.h"
94 #include "umask-util.h"
96 #include "user-util.h"
97 #include "utmp-wtmp.h"
99 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
100 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
102 #define SNDBUF_SIZE (8*1024*1024)
104 static int shift_fds(int fds
[], size_t n_fds
) {
108 /* Modifies the fds array! (sorts it) */
112 for (int start
= 0;;) {
113 int restart_from
= -1;
115 for (int i
= start
; i
< (int) n_fds
; i
++) {
118 /* Already at right index? */
122 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
129 /* Hmm, the fd we wanted isn't free? Then
130 * let's remember that and try again from here */
131 if (nfd
!= i
+3 && restart_from
< 0)
135 if (restart_from
< 0)
138 start
= restart_from
;
144 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
148 n_fds
= n_socket_fds
+ n_storage_fds
;
154 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
155 * O_NONBLOCK only applies to socket activation though. */
157 for (size_t i
= 0; i
< n_fds
; i
++) {
159 if (i
< n_socket_fds
) {
160 r
= fd_nonblock(fds
[i
], nonblock
);
165 /* We unconditionally drop FD_CLOEXEC from the fds,
166 * since after all we want to pass these fds to our
169 r
= fd_cloexec(fds
[i
], false);
177 static const char *exec_context_tty_path(const ExecContext
*context
) {
180 if (context
->stdio_as_fds
)
183 if (context
->tty_path
)
184 return context
->tty_path
;
186 return "/dev/console";
189 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
194 path
= exec_context_tty_path(context
);
196 if (context
->tty_vhangup
) {
197 if (p
&& p
->stdin_fd
>= 0)
198 (void) terminal_vhangup_fd(p
->stdin_fd
);
200 (void) terminal_vhangup(path
);
203 if (context
->tty_reset
) {
204 if (p
&& p
->stdin_fd
>= 0)
205 (void) reset_terminal_fd(p
->stdin_fd
, true);
207 (void) reset_terminal(path
);
210 if (context
->tty_vt_disallocate
&& path
)
211 (void) vt_disallocate(path
);
214 static bool is_terminal_input(ExecInput i
) {
217 EXEC_INPUT_TTY_FORCE
,
218 EXEC_INPUT_TTY_FAIL
);
221 static bool is_terminal_output(ExecOutput o
) {
224 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
225 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
228 static bool is_kmsg_output(ExecOutput o
) {
231 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
234 static bool exec_context_needs_term(const ExecContext
*c
) {
237 /* Return true if the execution context suggests we should set $TERM to something useful. */
239 if (is_terminal_input(c
->std_input
))
242 if (is_terminal_output(c
->std_output
))
245 if (is_terminal_output(c
->std_error
))
248 return !!c
->tty_path
;
251 static int open_null_as(int flags
, int nfd
) {
256 fd
= open("/dev/null", flags
|O_NOCTTY
);
260 return move_fd(fd
, nfd
, false);
263 static int connect_journal_socket(
265 const char *log_namespace
,
269 union sockaddr_union sa
;
271 uid_t olduid
= UID_INVALID
;
272 gid_t oldgid
= GID_INVALID
;
277 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
278 "/run/systemd/journal/stdout";
279 r
= sockaddr_un_set_path(&sa
.un
, j
);
284 if (gid_is_valid(gid
)) {
287 if (setegid(gid
) < 0)
291 if (uid_is_valid(uid
)) {
294 if (seteuid(uid
) < 0) {
300 r
= connect(fd
, &sa
.sa
, sa_len
) < 0 ? -errno
: 0;
302 /* If we fail to restore the uid or gid, things will likely
303 fail later on. This should only happen if an LSM interferes. */
305 if (uid_is_valid(uid
))
306 (void) seteuid(olduid
);
309 if (gid_is_valid(gid
))
310 (void) setegid(oldgid
);
315 static int connect_logger_as(
317 const ExecContext
*context
,
318 const ExecParameters
*params
,
325 _cleanup_close_
int fd
= -1;
330 assert(output
< _EXEC_OUTPUT_MAX
);
334 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
338 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
342 if (shutdown(fd
, SHUT_RD
) < 0)
345 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
355 context
->syslog_identifier
?: ident
,
356 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
357 context
->syslog_priority
,
358 !!context
->syslog_level_prefix
,
360 is_kmsg_output(output
),
361 is_terminal_output(output
)) < 0)
364 return move_fd(TAKE_FD(fd
), nfd
, false);
367 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
373 fd
= open_terminal(path
, flags
| O_NOCTTY
);
377 return move_fd(fd
, nfd
, false);
380 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
381 union sockaddr_union sa
;
383 _cleanup_close_
int fd
= -1;
388 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
391 fd
= open(path
, flags
|O_NOCTTY
, mode
);
395 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
398 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
400 r
= sockaddr_un_set_path(&sa
.un
, path
);
402 return r
== -EINVAL
? -ENXIO
: r
;
405 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
409 if (connect(fd
, &sa
.sa
, sa_len
) < 0)
410 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
411 * indication that this wasn't an AF_UNIX socket after all */
413 if ((flags
& O_ACCMODE
) == O_RDONLY
)
414 r
= shutdown(fd
, SHUT_WR
);
415 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
416 r
= shutdown(fd
, SHUT_RD
);
425 static int fixup_input(
426 const ExecContext
*context
,
428 bool apply_tty_stdin
) {
434 std_input
= context
->std_input
;
436 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
437 return EXEC_INPUT_NULL
;
439 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
440 return EXEC_INPUT_NULL
;
442 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
443 return EXEC_INPUT_NULL
;
448 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
450 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
451 return EXEC_OUTPUT_INHERIT
;
456 static int setup_input(
457 const ExecContext
*context
,
458 const ExecParameters
*params
,
460 const int named_iofds
[static 3]) {
468 if (params
->stdin_fd
>= 0) {
469 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
472 /* Try to make this the controlling tty, if it is a tty, and reset it */
473 if (isatty(STDIN_FILENO
)) {
474 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
475 (void) reset_terminal_fd(STDIN_FILENO
, true);
481 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
485 case EXEC_INPUT_NULL
:
486 return open_null_as(O_RDONLY
, STDIN_FILENO
);
489 case EXEC_INPUT_TTY_FORCE
:
490 case EXEC_INPUT_TTY_FAIL
: {
493 fd
= acquire_terminal(exec_context_tty_path(context
),
494 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
495 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
496 ACQUIRE_TERMINAL_WAIT
,
501 return move_fd(fd
, STDIN_FILENO
, false);
504 case EXEC_INPUT_SOCKET
:
505 assert(socket_fd
>= 0);
507 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
509 case EXEC_INPUT_NAMED_FD
:
510 assert(named_iofds
[STDIN_FILENO
] >= 0);
512 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
513 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
515 case EXEC_INPUT_DATA
: {
518 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
522 return move_fd(fd
, STDIN_FILENO
, false);
525 case EXEC_INPUT_FILE
: {
529 assert(context
->stdio_file
[STDIN_FILENO
]);
531 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
532 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
534 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
538 return move_fd(fd
, STDIN_FILENO
, false);
542 assert_not_reached("Unknown input type");
546 static bool can_inherit_stderr_from_stdout(
547 const ExecContext
*context
,
553 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
556 if (e
== EXEC_OUTPUT_INHERIT
)
561 if (e
== EXEC_OUTPUT_NAMED_FD
)
562 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
564 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
565 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
570 static int setup_output(
572 const ExecContext
*context
,
573 const ExecParameters
*params
,
576 const int named_iofds
[static 3],
580 dev_t
*journal_stream_dev
,
581 ino_t
*journal_stream_ino
) {
591 assert(journal_stream_dev
);
592 assert(journal_stream_ino
);
594 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
596 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
599 return STDOUT_FILENO
;
602 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
603 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
606 return STDERR_FILENO
;
609 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
610 o
= fixup_output(context
->std_output
, socket_fd
);
612 if (fileno
== STDERR_FILENO
) {
614 e
= fixup_output(context
->std_error
, socket_fd
);
616 /* This expects the input and output are already set up */
618 /* Don't change the stderr file descriptor if we inherit all
619 * the way and are not on a tty */
620 if (e
== EXEC_OUTPUT_INHERIT
&&
621 o
== EXEC_OUTPUT_INHERIT
&&
622 i
== EXEC_INPUT_NULL
&&
623 !is_terminal_input(context
->std_input
) &&
627 /* Duplicate from stdout if possible */
628 if (can_inherit_stderr_from_stdout(context
, o
, e
))
629 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
633 } else if (o
== EXEC_OUTPUT_INHERIT
) {
634 /* If input got downgraded, inherit the original value */
635 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
636 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
638 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
639 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
640 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
642 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
646 /* We need to open /dev/null here anew, to get the right access mode. */
647 return open_null_as(O_WRONLY
, fileno
);
652 case EXEC_OUTPUT_NULL
:
653 return open_null_as(O_WRONLY
, fileno
);
655 case EXEC_OUTPUT_TTY
:
656 if (is_terminal_input(i
))
657 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
659 /* We don't reset the terminal if this is just about output */
660 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
662 case EXEC_OUTPUT_KMSG
:
663 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
664 case EXEC_OUTPUT_JOURNAL
:
665 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
666 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
668 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
669 r
= open_null_as(O_WRONLY
, fileno
);
673 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
674 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
675 * services to detect whether they are connected to the journal or not.
677 * If both stdout and stderr are connected to a stream then let's make sure to store the data
678 * about STDERR as that's usually the best way to do logging. */
680 if (fstat(fileno
, &st
) >= 0 &&
681 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
682 *journal_stream_dev
= st
.st_dev
;
683 *journal_stream_ino
= st
.st_ino
;
688 case EXEC_OUTPUT_SOCKET
:
689 assert(socket_fd
>= 0);
691 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
693 case EXEC_OUTPUT_NAMED_FD
:
694 assert(named_iofds
[fileno
] >= 0);
696 (void) fd_nonblock(named_iofds
[fileno
], false);
697 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
699 case EXEC_OUTPUT_FILE
:
700 case EXEC_OUTPUT_FILE_APPEND
: {
704 assert(context
->stdio_file
[fileno
]);
706 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
707 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
710 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
713 if (o
== EXEC_OUTPUT_FILE_APPEND
)
716 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
720 return move_fd(fd
, fileno
, 0);
724 assert_not_reached("Unknown error type");
728 static int chown_terminal(int fd
, uid_t uid
) {
733 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
734 if (isatty(fd
) < 1) {
735 if (IN_SET(errno
, EINVAL
, ENOTTY
))
736 return 0; /* not a tty */
741 /* This might fail. What matters are the results. */
742 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, -1);
749 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
750 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
753 assert(_saved_stdin
);
754 assert(_saved_stdout
);
756 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
760 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
761 if (saved_stdout
< 0)
764 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
768 r
= chown_terminal(fd
, getuid());
772 r
= reset_terminal_fd(fd
, true);
776 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
781 *_saved_stdin
= saved_stdin
;
782 *_saved_stdout
= saved_stdout
;
784 saved_stdin
= saved_stdout
= -1;
789 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
792 if (err
== -ETIMEDOUT
)
793 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
796 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
800 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
801 _cleanup_close_
int fd
= -1;
805 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
809 write_confirm_error_fd(err
, fd
, u
);
812 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
816 assert(saved_stdout
);
820 if (*saved_stdin
>= 0)
821 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
824 if (*saved_stdout
>= 0)
825 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
828 *saved_stdin
= safe_close(*saved_stdin
);
829 *saved_stdout
= safe_close(*saved_stdout
);
835 CONFIRM_PRETEND_FAILURE
= -1,
836 CONFIRM_PRETEND_SUCCESS
= 0,
840 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
841 int saved_stdout
= -1, saved_stdin
= -1, r
;
842 _cleanup_free_
char *e
= NULL
;
845 /* For any internal errors, assume a positive response. */
846 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
848 write_confirm_error(r
, vc
, u
);
849 return CONFIRM_EXECUTE
;
852 /* confirm_spawn might have been disabled while we were sleeping. */
853 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
858 e
= ellipsize(cmdline
, 60, 100);
866 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
868 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
875 printf("Resuming normal execution.\n");
876 manager_disable_confirm_spawn();
880 unit_dump(u
, stdout
, " ");
881 continue; /* ask again */
883 printf("Failing execution.\n");
884 r
= CONFIRM_PRETEND_FAILURE
;
887 printf(" c - continue, proceed without asking anymore\n"
888 " D - dump, show the state of the unit\n"
889 " f - fail, don't execute the command and pretend it failed\n"
891 " i - info, show a short summary of the unit\n"
892 " j - jobs, show jobs that are in progress\n"
893 " s - skip, don't execute the command and pretend it succeeded\n"
894 " y - yes, execute the command\n");
895 continue; /* ask again */
897 printf(" Description: %s\n"
900 u
->id
, u
->description
, cmdline
);
901 continue; /* ask again */
903 manager_dump_jobs(u
->manager
, stdout
, " ");
904 continue; /* ask again */
906 /* 'n' was removed in favor of 'f'. */
907 printf("Didn't understand 'n', did you mean 'f'?\n");
908 continue; /* ask again */
910 printf("Skipping execution.\n");
911 r
= CONFIRM_PRETEND_SUCCESS
;
917 assert_not_reached("Unhandled choice");
923 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
927 static int get_fixed_user(const ExecContext
*c
, const char **user
,
928 uid_t
*uid
, gid_t
*gid
,
929 const char **home
, const char **shell
) {
938 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
939 * (i.e. are "/" or "/bin/nologin"). */
942 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
950 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
960 r
= get_group_creds(&name
, gid
, 0);
968 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
969 const char *group
, gid_t gid
,
970 gid_t
**supplementary_gids
, int *ngids
) {
974 bool keep_groups
= false;
975 gid_t
*groups
= NULL
;
976 _cleanup_free_ gid_t
*l_gids
= NULL
;
981 * If user is given, then lookup GID and supplementary groups list.
982 * We avoid NSS lookups for gid=0. Also we have to initialize groups
983 * here and as early as possible so we keep the list of supplementary
984 * groups of the caller.
986 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
987 /* First step, initialize groups from /etc/groups */
988 if (initgroups(user
, gid
) < 0)
994 if (strv_isempty(c
->supplementary_groups
))
998 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
999 * be positive, otherwise fail.
1002 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1003 if (ngroups_max
<= 0)
1004 return errno_or_else(EOPNOTSUPP
);
1006 l_gids
= new(gid_t
, ngroups_max
);
1012 * Lookup the list of groups that the user belongs to, we
1013 * avoid NSS lookups here too for gid=0.
1016 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1021 STRV_FOREACH(i
, c
->supplementary_groups
) {
1024 if (k
>= ngroups_max
)
1028 r
= get_group_creds(&g
, l_gids
+k
, 0);
1036 * Sets ngids to zero to drop all supplementary groups, happens
1037 * when we are under root and SupplementaryGroups= is empty.
1044 /* Otherwise get the final list of supplementary groups */
1045 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1049 *supplementary_gids
= groups
;
1057 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1060 /* Handle SupplementaryGroups= if it is not empty */
1062 r
= maybe_setgroups(ngids
, supplementary_gids
);
1067 if (gid_is_valid(gid
)) {
1068 /* Then set our gids */
1069 if (setresgid(gid
, gid
, gid
) < 0)
1076 static int set_securebits(int bits
, int mask
) {
1077 int current
, applied
;
1078 current
= prctl(PR_GET_SECUREBITS
);
1081 /* Clear all securebits defined in mask and set bits */
1082 applied
= (current
& ~mask
) | bits
;
1083 if (current
== applied
)
1085 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1090 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1094 if (!uid_is_valid(uid
))
1097 /* Sets (but doesn't look up) the uid and make sure we keep the
1098 * capabilities while doing so. For setting secure bits the capability CAP_SETPCAP is
1099 * required, so we also need keep-caps in this case.
1102 if (context
->capability_ambient_set
!= 0 || context
->secure_bits
!= 0) {
1104 /* First step: If we need to keep capabilities but
1105 * drop privileges we need to make sure we keep our
1106 * caps, while we drop privileges. */
1108 /* Add KEEP_CAPS to the securebits */
1109 r
= set_securebits(1<<SECURE_KEEP_CAPS
, 0);
1115 /* Second step: actually set the uids */
1116 if (setresuid(uid
, uid
, uid
) < 0)
1119 /* At this point we should have all necessary capabilities but
1120 are otherwise a normal user. However, the caps might got
1121 corrupted due to the setresuid() so we need clean them up
1122 later. This is done outside of this call. */
1129 static int null_conv(
1131 const struct pam_message
**msg
,
1132 struct pam_response
**resp
,
1133 void *appdata_ptr
) {
1135 /* We don't support conversations */
1137 return PAM_CONV_ERR
;
1142 static int setup_pam(
1149 const int fds
[], size_t n_fds
) {
1153 static const struct pam_conv conv
= {
1158 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1159 pam_handle_t
*handle
= NULL
;
1161 int pam_code
= PAM_SUCCESS
, r
;
1162 char **nv
, **e
= NULL
;
1163 bool close_session
= false;
1164 pid_t pam_pid
= 0, parent_pid
;
1171 /* We set up PAM in the parent process, then fork. The child
1172 * will then stay around until killed via PR_GET_PDEATHSIG or
1173 * systemd via the cgroup logic. It will then remove the PAM
1174 * session again. The parent process will exec() the actual
1175 * daemon. We do things this way to ensure that the main PID
1176 * of the daemon is the one we initially fork()ed. */
1178 r
= barrier_create(&barrier
);
1182 if (log_get_max_level() < LOG_DEBUG
)
1183 flags
|= PAM_SILENT
;
1185 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1186 if (pam_code
!= PAM_SUCCESS
) {
1192 _cleanup_free_
char *q
= NULL
;
1194 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1195 * out if that's the case, and read the TTY off it. */
1197 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1198 tty
= strjoina("/dev/", q
);
1202 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1203 if (pam_code
!= PAM_SUCCESS
)
1207 STRV_FOREACH(nv
, *env
) {
1208 pam_code
= pam_putenv(handle
, *nv
);
1209 if (pam_code
!= PAM_SUCCESS
)
1213 pam_code
= pam_acct_mgmt(handle
, flags
);
1214 if (pam_code
!= PAM_SUCCESS
)
1217 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1218 if (pam_code
!= PAM_SUCCESS
)
1219 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1221 pam_code
= pam_open_session(handle
, flags
);
1222 if (pam_code
!= PAM_SUCCESS
)
1225 close_session
= true;
1227 e
= pam_getenvlist(handle
);
1229 pam_code
= PAM_BUF_ERR
;
1233 /* Block SIGTERM, so that we know that it won't get lost in
1236 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1238 parent_pid
= getpid_cached();
1240 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1244 int sig
, ret
= EXIT_PAM
;
1246 /* The child's job is to reset the PAM session on
1248 barrier_set_role(&barrier
, BARRIER_CHILD
);
1250 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1251 * those fds are open here that have been opened by PAM. */
1252 (void) close_many(fds
, n_fds
);
1254 /* Drop privileges - we don't need any to pam_close_session
1255 * and this will make PR_SET_PDEATHSIG work in most cases.
1256 * If this fails, ignore the error - but expect sd-pam threads
1257 * to fail to exit normally */
1259 r
= maybe_setgroups(0, NULL
);
1261 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1262 if (setresgid(gid
, gid
, gid
) < 0)
1263 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1264 if (setresuid(uid
, uid
, uid
) < 0)
1265 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1267 (void) ignore_signals(SIGPIPE
, -1);
1269 /* Wait until our parent died. This will only work if
1270 * the above setresuid() succeeds, otherwise the kernel
1271 * will not allow unprivileged parents kill their privileged
1272 * children this way. We rely on the control groups kill logic
1273 * to do the rest for us. */
1274 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1277 /* Tell the parent that our setup is done. This is especially
1278 * important regarding dropping privileges. Otherwise, unit
1279 * setup might race against our setresuid(2) call.
1281 * If the parent aborted, we'll detect this below, hence ignore
1282 * return failure here. */
1283 (void) barrier_place(&barrier
);
1285 /* Check if our parent process might already have died? */
1286 if (getppid() == parent_pid
) {
1289 assert_se(sigemptyset(&ss
) >= 0);
1290 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1293 if (sigwait(&ss
, &sig
) < 0) {
1300 assert(sig
== SIGTERM
);
1305 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1306 if (pam_code
!= PAM_SUCCESS
)
1309 /* If our parent died we'll end the session */
1310 if (getppid() != parent_pid
) {
1311 pam_code
= pam_close_session(handle
, flags
);
1312 if (pam_code
!= PAM_SUCCESS
)
1319 pam_end(handle
, pam_code
| flags
);
1323 barrier_set_role(&barrier
, BARRIER_PARENT
);
1325 /* If the child was forked off successfully it will do all the
1326 * cleanups, so forget about the handle here. */
1329 /* Unblock SIGTERM again in the parent */
1330 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1332 /* We close the log explicitly here, since the PAM modules
1333 * might have opened it, but we don't want this fd around. */
1336 /* Synchronously wait for the child to initialize. We don't care for
1337 * errors as we cannot recover. However, warn loudly if it happens. */
1338 if (!barrier_place_and_sync(&barrier
))
1339 log_error("PAM initialization failed");
1341 return strv_free_and_replace(*env
, e
);
1344 if (pam_code
!= PAM_SUCCESS
) {
1345 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1346 r
= -EPERM
; /* PAM errors do not map to errno */
1348 log_error_errno(r
, "PAM failed: %m");
1352 pam_code
= pam_close_session(handle
, flags
);
1354 pam_end(handle
, pam_code
| flags
);
1366 static void rename_process_from_path(const char *path
) {
1367 char process_name
[11];
1371 /* This resulting string must fit in 10 chars (i.e. the length
1372 * of "/sbin/init") to look pretty in /bin/ps */
1376 rename_process("(...)");
1382 /* The end of the process name is usually more
1383 * interesting, since the first bit might just be
1389 process_name
[0] = '(';
1390 memcpy(process_name
+1, p
, l
);
1391 process_name
[1+l
] = ')';
1392 process_name
[1+l
+1] = 0;
1394 rename_process(process_name
);
1397 static bool context_has_address_families(const ExecContext
*c
) {
1400 return c
->address_families_allow_list
||
1401 !set_isempty(c
->address_families
);
1404 static bool context_has_syscall_filters(const ExecContext
*c
) {
1407 return c
->syscall_allow_list
||
1408 !hashmap_isempty(c
->syscall_filter
);
1411 static bool context_has_syscall_logs(const ExecContext
*c
) {
1414 return c
->syscall_log_allow_list
||
1415 !hashmap_isempty(c
->syscall_log
);
1418 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1421 if (c
->no_new_privileges
)
1424 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1427 /* We need NNP if we have any form of seccomp and are unprivileged */
1428 return context_has_address_families(c
) ||
1429 c
->memory_deny_write_execute
||
1430 c
->restrict_realtime
||
1431 c
->restrict_suid_sgid
||
1432 exec_context_restrict_namespaces_set(c
) ||
1434 c
->protect_kernel_tunables
||
1435 c
->protect_kernel_modules
||
1436 c
->protect_kernel_logs
||
1437 c
->private_devices
||
1438 context_has_syscall_filters(c
) ||
1439 context_has_syscall_logs(c
) ||
1440 !set_isempty(c
->syscall_archs
) ||
1441 c
->lock_personality
||
1442 c
->protect_hostname
;
1445 static bool exec_context_has_credentials(const ExecContext
*context
) {
1449 return !hashmap_isempty(context
->set_credentials
) ||
1450 context
->load_credentials
;
1455 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1457 if (is_seccomp_available())
1460 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1464 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1465 uint32_t negative_action
, default_action
, action
;
1471 if (!context_has_syscall_filters(c
))
1474 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1477 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1479 if (c
->syscall_allow_list
) {
1480 default_action
= negative_action
;
1481 action
= SCMP_ACT_ALLOW
;
1483 default_action
= SCMP_ACT_ALLOW
;
1484 action
= negative_action
;
1487 if (needs_ambient_hack
) {
1488 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1493 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1496 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1498 uint32_t default_action
, action
;
1504 if (!context_has_syscall_logs(c
))
1508 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1511 if (c
->syscall_log_allow_list
) {
1512 /* Log nothing but the ones listed */
1513 default_action
= SCMP_ACT_ALLOW
;
1514 action
= SCMP_ACT_LOG
;
1516 /* Log everything but the ones listed */
1517 default_action
= SCMP_ACT_LOG
;
1518 action
= SCMP_ACT_ALLOW
;
1521 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1523 /* old libseccomp */
1524 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1529 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1533 if (set_isempty(c
->syscall_archs
))
1536 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1539 return seccomp_restrict_archs(c
->syscall_archs
);
1542 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1546 if (!context_has_address_families(c
))
1549 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1552 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1555 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1559 if (!c
->memory_deny_write_execute
)
1562 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1565 return seccomp_memory_deny_write_execute();
1568 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1572 if (!c
->restrict_realtime
)
1575 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1578 return seccomp_restrict_realtime();
1581 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1585 if (!c
->restrict_suid_sgid
)
1588 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1591 return seccomp_restrict_suid_sgid();
1594 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1598 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1599 * let's protect even those systems where this is left on in the kernel. */
1601 if (!c
->protect_kernel_tunables
)
1604 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1607 return seccomp_protect_sysctl();
1610 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1614 /* Turn off module syscalls on ProtectKernelModules=yes */
1616 if (!c
->protect_kernel_modules
)
1619 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1622 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1625 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1629 if (!c
->protect_kernel_logs
)
1632 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1635 return seccomp_protect_syslog();
1638 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1642 if (!c
->protect_clock
)
1645 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1648 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1651 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1655 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1657 if (!c
->private_devices
)
1660 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1663 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1666 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1670 if (!exec_context_restrict_namespaces_set(c
))
1673 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1676 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1679 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1680 unsigned long personality
;
1686 if (!c
->lock_personality
)
1689 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1692 personality
= c
->personality
;
1694 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1695 if (personality
== PERSONALITY_INVALID
) {
1697 r
= opinionated_personality(&personality
);
1702 return seccomp_lock_personality(personality
);
1707 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1711 if (!c
->protect_hostname
)
1714 if (ns_type_supported(NAMESPACE_UTS
)) {
1715 if (unshare(CLONE_NEWUTS
) < 0) {
1716 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1717 *ret_exit_status
= EXIT_NAMESPACE
;
1718 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1721 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1724 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1729 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1732 r
= seccomp_protect_hostname();
1734 *ret_exit_status
= EXIT_SECCOMP
;
1735 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1742 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1745 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1746 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1748 if (idle_pipe
[0] >= 0) {
1751 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1753 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1756 /* Signal systemd that we are bored and want to continue. */
1757 n
= write(idle_pipe
[3], "x", 1);
1759 /* Wait for systemd to react to the signal above. */
1760 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1763 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1767 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1770 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1772 static int build_environment(
1774 const ExecContext
*c
,
1775 const ExecParameters
*p
,
1778 const char *username
,
1780 dev_t journal_stream_dev
,
1781 ino_t journal_stream_ino
,
1784 _cleanup_strv_free_
char **our_env
= NULL
;
1793 #define N_ENV_VARS 16
1794 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1799 _cleanup_free_
char *joined
= NULL
;
1801 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1803 our_env
[n_env
++] = x
;
1805 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1807 our_env
[n_env
++] = x
;
1809 joined
= strv_join(p
->fd_names
, ":");
1813 x
= strjoin("LISTEN_FDNAMES=", joined
);
1816 our_env
[n_env
++] = x
;
1819 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1820 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1822 our_env
[n_env
++] = x
;
1824 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1826 our_env
[n_env
++] = x
;
1829 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1830 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1831 * check the database directly. */
1832 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1833 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1836 our_env
[n_env
++] = x
;
1840 x
= strjoin("HOME=", home
);
1844 path_simplify(x
+ 5, true);
1845 our_env
[n_env
++] = x
;
1849 x
= strjoin("LOGNAME=", username
);
1852 our_env
[n_env
++] = x
;
1854 x
= strjoin("USER=", username
);
1857 our_env
[n_env
++] = x
;
1861 x
= strjoin("SHELL=", shell
);
1865 path_simplify(x
+ 6, true);
1866 our_env
[n_env
++] = x
;
1869 if (!sd_id128_is_null(u
->invocation_id
)) {
1870 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1873 our_env
[n_env
++] = x
;
1876 if (exec_context_needs_term(c
)) {
1877 const char *tty_path
, *term
= NULL
;
1879 tty_path
= exec_context_tty_path(c
);
1881 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1882 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1883 * container manager passes to PID 1 ends up all the way in the console login shown. */
1885 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1886 term
= getenv("TERM");
1889 term
= default_term_for_tty(tty_path
);
1891 x
= strjoin("TERM=", term
);
1894 our_env
[n_env
++] = x
;
1897 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1898 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1901 our_env
[n_env
++] = x
;
1904 if (c
->log_namespace
) {
1905 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1909 our_env
[n_env
++] = x
;
1912 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1913 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1919 if (strv_isempty(c
->directories
[t
].paths
))
1922 n
= exec_directory_env_name_to_string(t
);
1926 pre
= strjoin(p
->prefix
[t
], "/");
1930 joined
= strv_join_full(c
->directories
[t
].paths
, ":", pre
, true);
1934 x
= strjoin(n
, "=", joined
);
1938 our_env
[n_env
++] = x
;
1941 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
1942 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
1946 our_env
[n_env
++] = x
;
1949 our_env
[n_env
++] = NULL
;
1950 assert(n_env
<= N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1953 *ret
= TAKE_PTR(our_env
);
1958 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1959 _cleanup_strv_free_
char **pass_env
= NULL
;
1960 size_t n_env
= 0, n_bufsize
= 0;
1963 STRV_FOREACH(i
, c
->pass_environment
) {
1964 _cleanup_free_
char *x
= NULL
;
1970 x
= strjoin(*i
, "=", v
);
1974 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1977 pass_env
[n_env
++] = TAKE_PTR(x
);
1978 pass_env
[n_env
] = NULL
;
1981 *ret
= TAKE_PTR(pass_env
);
1986 static bool exec_needs_mount_namespace(
1987 const ExecContext
*context
,
1988 const ExecParameters
*params
,
1989 const ExecRuntime
*runtime
) {
1994 if (context
->root_image
)
1997 if (!strv_isempty(context
->read_write_paths
) ||
1998 !strv_isempty(context
->read_only_paths
) ||
1999 !strv_isempty(context
->inaccessible_paths
))
2002 if (context
->n_bind_mounts
> 0)
2005 if (context
->n_temporary_filesystems
> 0)
2008 if (context
->n_mount_images
> 0)
2011 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
2014 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
2017 if (context
->private_devices
||
2018 context
->private_mounts
||
2019 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2020 context
->protect_home
!= PROTECT_HOME_NO
||
2021 context
->protect_kernel_tunables
||
2022 context
->protect_kernel_modules
||
2023 context
->protect_kernel_logs
||
2024 context
->protect_control_groups
||
2025 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2026 context
->proc_subset
!= PROC_SUBSET_ALL
)
2029 if (context
->root_directory
) {
2030 if (exec_context_get_effective_mount_apivfs(context
))
2033 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2034 if (!params
->prefix
[t
])
2037 if (!strv_isempty(context
->directories
[t
].paths
))
2042 if (context
->dynamic_user
&&
2043 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
2044 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
2045 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
2048 if (context
->log_namespace
)
2054 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2055 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2056 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
2057 _cleanup_close_
int unshare_ready_fd
= -1;
2058 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2063 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2064 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2065 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2066 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2067 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2068 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2069 * continues execution normally.
2070 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2071 * does not need CAP_SETUID to write the single line mapping to itself. */
2073 /* Can only set up multiple mappings with CAP_SETUID. */
2074 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
2075 r
= asprintf(&uid_map
,
2076 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2077 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2078 ouid
, ouid
, uid
, uid
);
2080 r
= asprintf(&uid_map
,
2081 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2087 /* Can only set up multiple mappings with CAP_SETGID. */
2088 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2089 r
= asprintf(&gid_map
,
2090 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2091 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2092 ogid
, ogid
, gid
, gid
);
2094 r
= asprintf(&gid_map
,
2095 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2101 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2103 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2104 if (unshare_ready_fd
< 0)
2107 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2109 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2112 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2116 _cleanup_close_
int fd
= -1;
2120 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2121 * here, after the parent opened its own user namespace. */
2124 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2126 /* Wait until the parent unshared the user namespace */
2127 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2132 /* Disable the setgroups() system call in the child user namespace, for good. */
2133 a
= procfs_file_alloca(ppid
, "setgroups");
2134 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2136 if (errno
!= ENOENT
) {
2141 /* If the file is missing the kernel is too old, let's continue anyway. */
2143 if (write(fd
, "deny\n", 5) < 0) {
2148 fd
= safe_close(fd
);
2151 /* First write the GID map */
2152 a
= procfs_file_alloca(ppid
, "gid_map");
2153 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2158 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2162 fd
= safe_close(fd
);
2164 /* The write the UID map */
2165 a
= procfs_file_alloca(ppid
, "uid_map");
2166 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2171 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2176 _exit(EXIT_SUCCESS
);
2179 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2180 _exit(EXIT_FAILURE
);
2183 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2185 if (unshare(CLONE_NEWUSER
) < 0)
2188 /* Let the child know that the namespace is ready now */
2189 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2192 /* Try to read an error code from the child */
2193 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2196 if (n
== sizeof(r
)) { /* an error code was sent to us */
2201 if (n
!= 0) /* on success we should have read 0 bytes */
2204 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2208 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2214 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2215 if (!context
->dynamic_user
)
2218 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2221 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2227 static int setup_exec_directory(
2228 const ExecContext
*context
,
2229 const ExecParameters
*params
,
2232 ExecDirectoryType type
,
2235 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2236 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2237 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2238 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2239 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2240 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2247 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2248 assert(exit_status
);
2250 if (!params
->prefix
[type
])
2253 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2254 if (!uid_is_valid(uid
))
2256 if (!gid_is_valid(gid
))
2260 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2261 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2263 p
= path_join(params
->prefix
[type
], *rt
);
2269 r
= mkdir_parents_label(p
, 0755);
2273 if (exec_directory_is_private(context
, type
)) {
2274 _cleanup_free_
char *private_root
= NULL
;
2276 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2277 * case we want to avoid leaving a directory around fully accessible that is owned by
2278 * a dynamic user whose UID is later on reused. To lock this down we use the same
2279 * trick used by container managers to prohibit host users to get access to files of
2280 * the same UID in containers: we place everything inside a directory that has an
2281 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2282 * for unprivileged host code. We then use fs namespacing to make this directory
2283 * permeable for the service itself.
2285 * Specifically: for a service which wants a special directory "foo/" we first create
2286 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2287 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2288 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2289 * unprivileged host users can't look into it. Inside of the namespace of the unit
2290 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2291 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2292 * for the service and making sure it only gets access to the dirs it needs but no
2293 * others. Tricky? Yes, absolutely, but it works!
2295 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2296 * to be owned by the service itself.
2298 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2299 * for sharing files or sockets with other services. */
2301 private_root
= path_join(params
->prefix
[type
], "private");
2302 if (!private_root
) {
2307 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2308 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2312 pp
= path_join(private_root
, *rt
);
2318 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2319 r
= mkdir_parents_label(pp
, 0755);
2323 if (is_dir(p
, false) > 0 &&
2324 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2326 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2327 * it over. Most likely the service has been upgraded from one that didn't use
2328 * DynamicUser=1, to one that does. */
2330 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2331 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2332 exec_directory_type_to_string(type
), p
, pp
);
2334 if (rename(p
, pp
) < 0) {
2339 /* Otherwise, create the actual directory for the service */
2341 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2342 if (r
< 0 && r
!= -EEXIST
)
2346 /* And link it up from the original place */
2347 r
= symlink_idempotent(pp
, p
, true);
2352 _cleanup_free_
char *target
= NULL
;
2354 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2355 readlink_and_make_absolute(p
, &target
) >= 0) {
2356 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2358 /* This already exists and is a symlink? Interesting. Maybe it's one created
2359 * by DynamicUser=1 (see above)?
2361 * We do this for all directory types except for ConfigurationDirectory=,
2362 * since they all support the private/ symlink logic at least in some
2363 * configurations, see above. */
2365 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2369 q
= path_join(params
->prefix
[type
], "private", *rt
);
2375 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2376 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2380 if (path_equal(q_resolved
, target_resolved
)) {
2382 /* Hmm, apparently DynamicUser= was once turned on for this service,
2383 * but is no longer. Let's move the directory back up. */
2385 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2386 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2387 exec_directory_type_to_string(type
), q
, p
);
2389 if (unlink(p
) < 0) {
2394 if (rename(q
, p
) < 0) {
2401 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2406 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2409 /* Don't change the owner/access mode of the configuration directory,
2410 * as in the common case it is not written to by a service, and shall
2411 * not be writable. */
2413 if (stat(p
, &st
) < 0) {
2418 /* Still complain if the access mode doesn't match */
2419 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2420 log_warning("%s \'%s\' already exists but the mode is different. "
2421 "(File system: %o %sMode: %o)",
2422 exec_directory_type_to_string(type
), *rt
,
2423 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2430 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2431 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2432 * current UID/GID ownership.) */
2433 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2437 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2438 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2439 * assignments to exist.*/
2440 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2448 *exit_status
= exit_status_table
[type
];
2452 static int write_credential(
2458 bool ownership_ok
) {
2460 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2461 _cleanup_close_
int fd
= -1;
2464 r
= tempfn_random_child("", "cred", &tmp
);
2468 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2474 r
= loop_write(fd
, data
, size
, /* do_pool = */ false);
2478 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2481 if (uid_is_valid(uid
) && uid
!= getuid()) {
2482 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2484 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2487 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2488 * to express: that the user gets read access and nothing
2489 * else. But if the backing fs can't support that (e.g. ramfs)
2490 * then we can use file ownership instead. But that's only safe if
2491 * we can then re-mount the whole thing read-only, so that the
2492 * user can no longer chmod() the file to gain write access. */
2495 if (fchown(fd
, uid
, (gid_t
) -1) < 0)
2500 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2507 #define CREDENTIALS_BYTES_MAX (1024LU * 1024LU) /* Refuse to pass more than 1M, after all this is unswappable memory */
2509 static int acquire_credentials(
2510 const ExecContext
*context
,
2511 const ExecParameters
*params
,
2514 bool ownership_ok
) {
2516 uint64_t left
= CREDENTIALS_BYTES_MAX
;
2517 _cleanup_close_
int dfd
= -1;
2518 ExecSetCredential
*sc
;
2525 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
2529 /* First we use the literally specified credentials. Note that they might be overridden again below,
2530 * and thus act as a "default" if the same credential is specified multiple times */
2531 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
2534 add
= strlen(sc
->id
) + sc
->size
;
2538 r
= write_credential(dfd
, sc
->id
, sc
->data
, sc
->size
, uid
, ownership_ok
);
2545 /* Then, load credential off disk (or acquire via AF_UNIX socket) */
2546 STRV_FOREACH_PAIR(id
, fn
, context
->load_credentials
) {
2547 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
2548 _cleanup_(erase_and_freep
) char *data
= NULL
;
2549 _cleanup_free_
char *j
= NULL
;
2553 if (path_is_absolute(*fn
)) {
2554 /* If this is an absolute path, read the data directly from it, and support AF_UNIX sockets */
2556 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2557 } else if (params
->received_credentials
) {
2558 /* If this is a relative path, take it relative to the credentials we received
2559 * ourselves. We don't support the AF_UNIX stuff in this mode, since we are operating
2560 * on a credential store, i.e. this is guaranteed to be regular files. */
2561 j
= path_join(params
->received_credentials
, *fn
);
2570 r
= read_full_file_full(AT_FDCWD
, source
, flags
, &data
, &size
);
2574 faccessat(dfd
, *id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0) /* If the source file doesn't exist, but we already acquired the key otherwise, then don't fail */
2579 add
= strlen(*id
) + size
;
2583 r
= write_credential(dfd
, *id
, data
, size
, uid
, ownership_ok
);
2590 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
2593 /* After we created all keys with the right perms, also make sure the credential store as a whole is
2596 if (uid_is_valid(uid
) && uid
!= getuid()) {
2597 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
2599 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2605 if (fchown(dfd
, uid
, (gid_t
) -1) < 0)
2613 static int setup_credentials_internal(
2614 const ExecContext
*context
,
2615 const ExecParameters
*params
,
2616 const char *final
, /* This is where the credential store shall eventually end up at */
2617 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
2618 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
2619 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
2622 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
2623 * if we mounted something; false if we definitely can't mount anything */
2631 if (reuse_workspace
) {
2632 r
= path_is_mount_point(workspace
, NULL
, 0);
2636 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
2638 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
2640 workspace_mounted
= -1; /* ditto */
2642 r
= path_is_mount_point(final
, NULL
, 0);
2646 /* If the final place already has something mounted, we use that. If the workspace also has
2647 * something mounted we assume it's actually the same mount (but with MS_RDONLY
2649 final_mounted
= true;
2651 if (workspace_mounted
< 0) {
2652 /* If the final place is mounted, but the workspace we isn't, then let's bind mount
2653 * the final version to the workspace, and make it writable, so that we can make
2656 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2660 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2664 workspace_mounted
= true;
2667 final_mounted
= false;
2669 if (workspace_mounted
< 0) {
2670 /* Nothing is mounted on the workspace yet, let's try to mount something now */
2671 for (int try = 0;; try++) {
2674 /* Try "ramfs" first, since it's not swap backed */
2675 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
2677 workspace_mounted
= true;
2681 } else if (try == 1) {
2682 _cleanup_free_
char *opts
= NULL
;
2684 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%lu", CREDENTIALS_BYTES_MAX
) < 0)
2687 /* Fall back to "tmpfs" otherwise */
2688 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
2690 workspace_mounted
= true;
2695 /* 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. */
2696 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2698 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
2701 if (must_mount
) /* If we it's not OK to use the plain directory
2702 * fallback, propagate all errors too */
2705 /* If we lack privileges to bind mount stuff, then let's gracefully
2706 * proceed for compat with container envs, and just use the final dir
2709 workspace_mounted
= false;
2713 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
2714 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2718 workspace_mounted
= true;
2724 assert(!must_mount
|| workspace_mounted
> 0);
2725 where
= workspace_mounted
? workspace
: final
;
2727 r
= acquire_credentials(context
, params
, where
, uid
, workspace_mounted
);
2731 if (workspace_mounted
) {
2732 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
2733 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2737 /* And mount it to the final place, read-only */
2739 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
2741 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
2745 _cleanup_free_
char *parent
= NULL
;
2747 /* If we do not have our own mount put used the plain directory fallback, then we need to
2748 * open access to the top-level credential directory and the per-service directory now */
2750 parent
= dirname_malloc(final
);
2753 if (chmod(parent
, 0755) < 0)
2760 static int setup_credentials(
2761 const ExecContext
*context
,
2762 const ExecParameters
*params
,
2766 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
2773 if (!exec_context_has_credentials(context
))
2776 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
2779 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
2780 * and the subdir we mount over with a read-only file system readable by the service's user */
2781 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
2785 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
2786 if (r
< 0 && r
!= -EEXIST
)
2789 p
= path_join(q
, unit
);
2793 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
2794 if (r
< 0 && r
!= -EEXIST
)
2797 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
2799 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
2801 /* If this is not a privilege or support issue then propagate the error */
2802 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2805 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
2806 * it into place, so that users can't access half-initialized credential stores. */
2807 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
2811 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
2812 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
2813 * after it is fully set up */
2814 u
= path_join(t
, unit
);
2818 FOREACH_STRING(i
, t
, u
) {
2819 r
= mkdir_label(i
, 0700);
2820 if (r
< 0 && r
!= -EEXIST
)
2824 r
= setup_credentials_internal(
2827 p
, /* final mount point */
2828 u
, /* temporary workspace to overmount */
2829 true, /* reuse the workspace if it is already a mount */
2830 false, /* it's OK to fall back to a plain directory if we can't mount anything */
2833 (void) rmdir(u
); /* remove the workspace again if we can. */
2838 } else if (r
== 0) {
2840 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
2841 * we can use the same directory for all cases, after turning off propagation. Question
2842 * though is: where do we turn off propagation exactly, and where do we place the workspace
2843 * directory? We need some place that is guaranteed to be a mount point in the host, and
2844 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
2845 * since we ultimately want to move the resulting file system there, i.e. we need propagation
2846 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
2847 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
2848 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
2849 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
2850 * propagation on the former, and then overmount the latter.
2852 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
2853 * for this purpose, but there are few other candidates that work equally well for us, and
2854 * given that the we do this in a privately namespaced short-lived single-threaded process
2855 * that no one else sees this should be OK to do.*/
2857 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
2861 r
= setup_credentials_internal(
2864 p
, /* final mount point */
2865 "/dev/shm", /* temporary workspace to overmount */
2866 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
2867 true, /* insist that something is mounted, do not allow fallback to plain directory */
2872 _exit(EXIT_SUCCESS
);
2875 _exit(EXIT_FAILURE
);
2882 static int setup_smack(
2883 const ExecContext
*context
,
2884 const char *executable
) {
2890 if (context
->smack_process_label
) {
2891 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2895 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2897 _cleanup_free_
char *exec_label
= NULL
;
2899 r
= mac_smack_read(executable
, SMACK_ATTR_EXEC
, &exec_label
);
2900 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2903 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2913 static int compile_bind_mounts(
2914 const ExecContext
*context
,
2915 const ExecParameters
*params
,
2916 BindMount
**ret_bind_mounts
,
2917 size_t *ret_n_bind_mounts
,
2918 char ***ret_empty_directories
) {
2920 _cleanup_strv_free_
char **empty_directories
= NULL
;
2921 BindMount
*bind_mounts
;
2927 assert(ret_bind_mounts
);
2928 assert(ret_n_bind_mounts
);
2929 assert(ret_empty_directories
);
2931 n
= context
->n_bind_mounts
;
2932 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2933 if (!params
->prefix
[t
])
2936 n
+= strv_length(context
->directories
[t
].paths
);
2940 *ret_bind_mounts
= NULL
;
2941 *ret_n_bind_mounts
= 0;
2942 *ret_empty_directories
= NULL
;
2946 bind_mounts
= new(BindMount
, n
);
2950 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
2951 BindMount
*item
= context
->bind_mounts
+ i
;
2954 s
= strdup(item
->source
);
2960 d
= strdup(item
->destination
);
2967 bind_mounts
[h
++] = (BindMount
) {
2970 .read_only
= item
->read_only
,
2971 .recursive
= item
->recursive
,
2972 .ignore_enoent
= item
->ignore_enoent
,
2976 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2979 if (!params
->prefix
[t
])
2982 if (strv_isempty(context
->directories
[t
].paths
))
2985 if (exec_directory_is_private(context
, t
) &&
2986 !exec_context_with_rootfs(context
)) {
2989 /* So this is for a dynamic user, and we need to make sure the process can access its own
2990 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2991 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2993 private_root
= path_join(params
->prefix
[t
], "private");
2994 if (!private_root
) {
2999 r
= strv_consume(&empty_directories
, private_root
);
3004 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
3007 if (exec_directory_is_private(context
, t
))
3008 s
= path_join(params
->prefix
[t
], "private", *suffix
);
3010 s
= path_join(params
->prefix
[t
], *suffix
);
3016 if (exec_directory_is_private(context
, t
) &&
3017 exec_context_with_rootfs(context
))
3018 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3019 * directory is not created on the root directory. So, let's bind-mount the directory
3020 * on the 'non-private' place. */
3021 d
= path_join(params
->prefix
[t
], *suffix
);
3030 bind_mounts
[h
++] = (BindMount
) {
3034 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3036 .ignore_enoent
= false,
3043 *ret_bind_mounts
= bind_mounts
;
3044 *ret_n_bind_mounts
= n
;
3045 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3050 bind_mount_free_many(bind_mounts
, h
);
3054 static bool insist_on_sandboxing(
3055 const ExecContext
*context
,
3056 const char *root_dir
,
3057 const char *root_image
,
3058 const BindMount
*bind_mounts
,
3059 size_t n_bind_mounts
) {
3062 assert(n_bind_mounts
== 0 || bind_mounts
);
3064 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3065 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3066 * rearrange stuff in a way we cannot ignore gracefully. */
3068 if (context
->n_temporary_filesystems
> 0)
3071 if (root_dir
|| root_image
)
3074 if (context
->n_mount_images
> 0)
3077 if (context
->dynamic_user
)
3080 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3082 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3083 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3086 if (context
->log_namespace
)
3092 static int apply_mount_namespace(
3094 ExecCommandFlags command_flags
,
3095 const ExecContext
*context
,
3096 const ExecParameters
*params
,
3097 const ExecRuntime
*runtime
,
3098 char **error_path
) {
3100 _cleanup_strv_free_
char **empty_directories
= NULL
;
3101 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3102 const char *root_dir
= NULL
, *root_image
= NULL
;
3103 _cleanup_free_
char *creds_path
= NULL
;
3104 NamespaceInfo ns_info
;
3105 bool needs_sandboxing
;
3106 BindMount
*bind_mounts
= NULL
;
3107 size_t n_bind_mounts
= 0;
3112 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3113 root_image
= context
->root_image
;
3116 root_dir
= context
->root_directory
;
3119 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3123 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3124 if (needs_sandboxing
) {
3125 /* The runtime struct only contains the parent of the private /tmp,
3126 * which is non-accessible to world users. Inside of it there's a /tmp
3127 * that is sticky, and that's the one we want to use here.
3128 * This does not apply when we are using /run/systemd/empty as fallback. */
3130 if (context
->private_tmp
&& runtime
) {
3131 if (streq_ptr(runtime
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3132 tmp_dir
= runtime
->tmp_dir
;
3133 else if (runtime
->tmp_dir
)
3134 tmp_dir
= strjoina(runtime
->tmp_dir
, "/tmp");
3136 if (streq_ptr(runtime
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3137 var_tmp_dir
= runtime
->var_tmp_dir
;
3138 else if (runtime
->var_tmp_dir
)
3139 var_tmp_dir
= strjoina(runtime
->var_tmp_dir
, "/tmp");
3142 ns_info
= (NamespaceInfo
) {
3143 .ignore_protect_paths
= false,
3144 .private_dev
= context
->private_devices
,
3145 .protect_control_groups
= context
->protect_control_groups
,
3146 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3147 .protect_kernel_modules
= context
->protect_kernel_modules
,
3148 .protect_kernel_logs
= context
->protect_kernel_logs
,
3149 .protect_hostname
= context
->protect_hostname
,
3150 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3151 .private_mounts
= context
->private_mounts
,
3152 .protect_home
= context
->protect_home
,
3153 .protect_system
= context
->protect_system
,
3154 .protect_proc
= context
->protect_proc
,
3155 .proc_subset
= context
->proc_subset
,
3157 } else if (!context
->dynamic_user
&& root_dir
)
3159 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3160 * sandbox info, otherwise enforce it, don't ignore protected paths and
3161 * fail if we are enable to apply the sandbox inside the mount namespace.
3163 ns_info
= (NamespaceInfo
) {
3164 .ignore_protect_paths
= true,
3167 ns_info
= (NamespaceInfo
) {};
3169 if (context
->mount_flags
== MS_SHARED
)
3170 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3172 if (exec_context_has_credentials(context
) && params
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
3173 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3180 r
= setup_namespace(root_dir
, root_image
, context
->root_image_options
,
3181 &ns_info
, context
->read_write_paths
,
3182 needs_sandboxing
? context
->read_only_paths
: NULL
,
3183 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3187 context
->temporary_filesystems
,
3188 context
->n_temporary_filesystems
,
3189 context
->mount_images
,
3190 context
->n_mount_images
,
3194 context
->log_namespace
,
3195 context
->mount_flags
,
3196 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3197 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3198 context
->root_verity
,
3199 DISSECT_IMAGE_DISCARD_ON_LOOP
|DISSECT_IMAGE_RELAX_VAR_CHECK
|DISSECT_IMAGE_FSCK
,
3202 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3203 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3204 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3205 * completely different execution environment. */
3207 if (insist_on_sandboxing(
3209 root_dir
, root_image
,
3212 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3213 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3214 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
3218 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3224 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
3228 static int apply_working_directory(
3229 const ExecContext
*context
,
3230 const ExecParameters
*params
,
3237 assert(exit_status
);
3239 if (context
->working_directory_home
) {
3242 *exit_status
= EXIT_CHDIR
;
3249 wd
= empty_to_root(context
->working_directory
);
3251 if (params
->flags
& EXEC_APPLY_CHROOT
)
3254 d
= prefix_roota(context
->root_directory
, wd
);
3256 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3257 *exit_status
= EXIT_CHDIR
;
3264 static int apply_root_directory(
3265 const ExecContext
*context
,
3266 const ExecParameters
*params
,
3267 const bool needs_mount_ns
,
3271 assert(exit_status
);
3273 if (params
->flags
& EXEC_APPLY_CHROOT
)
3274 if (!needs_mount_ns
&& context
->root_directory
)
3275 if (chroot(context
->root_directory
) < 0) {
3276 *exit_status
= EXIT_CHROOT
;
3283 static int setup_keyring(
3285 const ExecContext
*context
,
3286 const ExecParameters
*p
,
3287 uid_t uid
, gid_t gid
) {
3289 key_serial_t keyring
;
3298 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3299 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3300 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3301 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3302 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3303 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3305 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3308 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3309 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3310 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3311 * & group is just as nasty as acquiring a reference to the user keyring. */
3313 saved_uid
= getuid();
3314 saved_gid
= getgid();
3316 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3317 if (setregid(gid
, -1) < 0)
3318 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3321 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3322 if (setreuid(uid
, -1) < 0) {
3323 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3328 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3329 if (keyring
== -1) {
3330 if (errno
== ENOSYS
)
3331 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3332 else if (ERRNO_IS_PRIVILEGE(errno
))
3333 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3334 else if (errno
== EDQUOT
)
3335 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3337 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3342 /* When requested link the user keyring into the session keyring. */
3343 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3345 if (keyctl(KEYCTL_LINK
,
3346 KEY_SPEC_USER_KEYRING
,
3347 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3348 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3353 /* Restore uid/gid back */
3354 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3355 if (setreuid(saved_uid
, -1) < 0) {
3356 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3361 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3362 if (setregid(saved_gid
, -1) < 0)
3363 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3366 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3367 if (!sd_id128_is_null(u
->invocation_id
)) {
3370 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3372 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3374 if (keyctl(KEYCTL_SETPERM
, key
,
3375 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3376 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3377 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3382 /* Revert back uid & gid for the last time, and exit */
3383 /* no extra logging, as only the first already reported error matters */
3384 if (getuid() != saved_uid
)
3385 (void) setreuid(saved_uid
, -1);
3387 if (getgid() != saved_gid
)
3388 (void) setregid(saved_gid
, -1);
3393 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3399 array
[(*n
)++] = pair
[0];
3401 array
[(*n
)++] = pair
[1];
3404 static int close_remaining_fds(
3405 const ExecParameters
*params
,
3406 const ExecRuntime
*runtime
,
3407 const DynamicCreds
*dcreds
,
3410 const int *fds
, size_t n_fds
) {
3412 size_t n_dont_close
= 0;
3413 int dont_close
[n_fds
+ 12];
3417 if (params
->stdin_fd
>= 0)
3418 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3419 if (params
->stdout_fd
>= 0)
3420 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3421 if (params
->stderr_fd
>= 0)
3422 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3425 dont_close
[n_dont_close
++] = socket_fd
;
3427 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3428 n_dont_close
+= n_fds
;
3432 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3436 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3438 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3441 if (user_lookup_fd
>= 0)
3442 dont_close
[n_dont_close
++] = user_lookup_fd
;
3444 return close_all_fds(dont_close
, n_dont_close
);
3447 static int send_user_lookup(
3455 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3456 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3459 if (user_lookup_fd
< 0)
3462 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3465 if (writev(user_lookup_fd
,
3467 IOVEC_INIT(&uid
, sizeof(uid
)),
3468 IOVEC_INIT(&gid
, sizeof(gid
)),
3469 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3475 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3482 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3487 if (!c
->working_directory_home
)
3490 r
= get_home_dir(buf
);
3498 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3499 _cleanup_strv_free_
char ** list
= NULL
;
3506 assert(c
->dynamic_user
);
3508 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3509 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3512 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3515 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3521 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3524 if (exec_directory_is_private(c
, t
))
3525 e
= path_join(p
->prefix
[t
], "private", *i
);
3527 e
= path_join(p
->prefix
[t
], *i
);
3531 r
= strv_consume(&list
, e
);
3537 *ret
= TAKE_PTR(list
);
3542 static char *exec_command_line(char **argv
);
3544 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3545 bool using_subcgroup
;
3551 if (!params
->cgroup_path
)
3554 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3555 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3556 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3557 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3558 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3559 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3560 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3561 * flag, which is only passed for the former statements, not for the latter. */
3563 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3564 if (using_subcgroup
)
3565 p
= path_join(params
->cgroup_path
, ".control");
3567 p
= strdup(params
->cgroup_path
);
3572 return using_subcgroup
;
3575 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3576 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3582 if (!c
->numa_policy
.nodes
.set
) {
3583 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3587 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3593 return cpu_set_add_all(ret
, &s
);
3596 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3599 return c
->cpu_affinity_from_numa
;
3602 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
3607 assert(*n_fds
< fds_size
);
3615 if (fd
< 3 + (int) *n_fds
) {
3616 /* Let's move the fd up, so that it's outside of the fd range we will use to store
3617 * the fds we pass to the process (or which are closed only during execve). */
3619 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
3623 CLOSE_AND_REPLACE(fd
, r
);
3626 *ret_fd
= fds
[*n_fds
] = fd
;
3631 static int exec_child(
3633 const ExecCommand
*command
,
3634 const ExecContext
*context
,
3635 const ExecParameters
*params
,
3636 ExecRuntime
*runtime
,
3637 DynamicCreds
*dcreds
,
3639 const int named_iofds
[static 3],
3641 size_t n_socket_fds
,
3642 size_t n_storage_fds
,
3647 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3648 int r
, ngids
= 0, exec_fd
;
3649 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3650 const char *username
= NULL
, *groupname
= NULL
;
3651 _cleanup_free_
char *home_buffer
= NULL
;
3652 const char *home
= NULL
, *shell
= NULL
;
3653 char **final_argv
= NULL
;
3654 dev_t journal_stream_dev
= 0;
3655 ino_t journal_stream_ino
= 0;
3656 bool userns_set_up
= false;
3657 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3658 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3659 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3660 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3662 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3663 bool use_selinux
= false;
3666 bool use_smack
= false;
3669 bool use_apparmor
= false;
3671 uid_t saved_uid
= getuid();
3672 gid_t saved_gid
= getgid();
3673 uid_t uid
= UID_INVALID
;
3674 gid_t gid
= GID_INVALID
;
3675 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
3676 n_keep_fds
; /* total number of fds not to close */
3678 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3679 int ngids_after_pam
= 0;
3685 assert(exit_status
);
3687 rename_process_from_path(command
->path
);
3689 /* We reset exactly these signals, since they are the
3690 * only ones we set to SIG_IGN in the main daemon. All
3691 * others we leave untouched because we set them to
3692 * SIG_DFL or a valid handler initially, both of which
3693 * will be demoted to SIG_DFL. */
3694 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3695 SIGNALS_IGNORE
, -1);
3697 if (context
->ignore_sigpipe
)
3698 (void) ignore_signals(SIGPIPE
, -1);
3700 r
= reset_signal_mask();
3702 *exit_status
= EXIT_SIGNAL_MASK
;
3703 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3706 if (params
->idle_pipe
)
3707 do_idle_pipe_dance(params
->idle_pipe
);
3709 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3710 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3711 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3712 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3715 log_set_open_when_needed(true);
3717 /* In case anything used libc syslog(), close this here, too */
3720 int keep_fds
[n_fds
+ 1];
3721 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
3724 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
3726 *exit_status
= EXIT_FDS
;
3727 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
3730 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
3732 *exit_status
= EXIT_FDS
;
3733 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3736 if (!context
->same_pgrp
&&
3738 *exit_status
= EXIT_SETSID
;
3739 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3742 exec_context_tty_reset(context
, params
);
3744 if (unit_shall_confirm_spawn(unit
)) {
3745 const char *vc
= params
->confirm_spawn
;
3746 _cleanup_free_
char *cmdline
= NULL
;
3748 cmdline
= exec_command_line(command
->argv
);
3750 *exit_status
= EXIT_MEMORY
;
3754 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3755 if (r
!= CONFIRM_EXECUTE
) {
3756 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3757 *exit_status
= EXIT_SUCCESS
;
3760 *exit_status
= EXIT_CONFIRM
;
3761 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
3762 "Execution cancelled by the user");
3766 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3767 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3768 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3769 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3770 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3771 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3772 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3773 *exit_status
= EXIT_MEMORY
;
3774 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3777 if (context
->dynamic_user
&& dcreds
) {
3778 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3780 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3781 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
3782 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3783 *exit_status
= EXIT_USER
;
3784 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3787 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3789 *exit_status
= EXIT_MEMORY
;
3793 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3795 *exit_status
= EXIT_USER
;
3797 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3800 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3803 if (!uid_is_valid(uid
)) {
3804 *exit_status
= EXIT_USER
;
3805 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3809 if (!gid_is_valid(gid
)) {
3810 *exit_status
= EXIT_USER
;
3811 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3816 username
= dcreds
->user
->name
;
3819 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3821 *exit_status
= EXIT_USER
;
3822 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3825 r
= get_fixed_group(context
, &groupname
, &gid
);
3827 *exit_status
= EXIT_GROUP
;
3828 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3832 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3833 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3834 &supplementary_gids
, &ngids
);
3836 *exit_status
= EXIT_GROUP
;
3837 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3840 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3842 *exit_status
= EXIT_USER
;
3843 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3846 user_lookup_fd
= safe_close(user_lookup_fd
);
3848 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3850 *exit_status
= EXIT_CHDIR
;
3851 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3854 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3855 * must sure to drop O_NONBLOCK */
3857 (void) fd_nonblock(socket_fd
, false);
3859 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3860 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3861 if (params
->cgroup_path
) {
3862 _cleanup_free_
char *p
= NULL
;
3864 r
= exec_parameters_get_cgroup_path(params
, &p
);
3866 *exit_status
= EXIT_CGROUP
;
3867 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3870 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3872 *exit_status
= EXIT_CGROUP
;
3873 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3877 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3878 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3880 *exit_status
= EXIT_NETWORK
;
3881 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3885 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3887 *exit_status
= EXIT_STDIN
;
3888 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3891 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3893 *exit_status
= EXIT_STDOUT
;
3894 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3897 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3899 *exit_status
= EXIT_STDERR
;
3900 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3903 if (context
->oom_score_adjust_set
) {
3904 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3905 * prohibit write access to this file, and we shouldn't trip up over that. */
3906 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3907 if (ERRNO_IS_PRIVILEGE(r
))
3908 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3910 *exit_status
= EXIT_OOM_ADJUST
;
3911 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3915 if (context
->coredump_filter_set
) {
3916 r
= set_coredump_filter(context
->coredump_filter
);
3917 if (ERRNO_IS_PRIVILEGE(r
))
3918 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
3920 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
3923 if (context
->nice_set
) {
3924 r
= setpriority_closest(context
->nice
);
3926 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3929 if (context
->cpu_sched_set
) {
3930 struct sched_param param
= {
3931 .sched_priority
= context
->cpu_sched_priority
,
3934 r
= sched_setscheduler(0,
3935 context
->cpu_sched_policy
|
3936 (context
->cpu_sched_reset_on_fork
?
3937 SCHED_RESET_ON_FORK
: 0),
3940 *exit_status
= EXIT_SETSCHEDULER
;
3941 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3945 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
3946 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
3947 const CPUSet
*cpu_set
;
3949 if (context
->cpu_affinity_from_numa
) {
3950 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
3952 *exit_status
= EXIT_CPUAFFINITY
;
3953 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
3956 cpu_set
= &converted_cpu_set
;
3958 cpu_set
= &context
->cpu_set
;
3960 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
3961 *exit_status
= EXIT_CPUAFFINITY
;
3962 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3966 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
3967 r
= apply_numa_policy(&context
->numa_policy
);
3968 if (r
== -EOPNOTSUPP
)
3969 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
3971 *exit_status
= EXIT_NUMA_POLICY
;
3972 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
3976 if (context
->ioprio_set
)
3977 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3978 *exit_status
= EXIT_IOPRIO
;
3979 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3982 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3983 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3984 *exit_status
= EXIT_TIMERSLACK
;
3985 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3988 if (context
->personality
!= PERSONALITY_INVALID
) {
3989 r
= safe_personality(context
->personality
);
3991 *exit_status
= EXIT_PERSONALITY
;
3992 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3996 if (context
->utmp_id
)
3997 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3999 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
4000 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
4004 if (uid_is_valid(uid
)) {
4005 r
= chown_terminal(STDIN_FILENO
, uid
);
4007 *exit_status
= EXIT_STDIN
;
4008 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
4012 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4013 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4014 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4015 * touch a single hierarchy too. */
4016 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
4017 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
4019 *exit_status
= EXIT_CGROUP
;
4020 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4024 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4025 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
4027 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4030 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4031 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4033 *exit_status
= EXIT_CREDENTIALS
;
4034 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4038 r
= build_environment(
4050 *exit_status
= EXIT_MEMORY
;
4054 r
= build_pass_environment(context
, &pass_env
);
4056 *exit_status
= EXIT_MEMORY
;
4060 accum_env
= strv_env_merge(5,
4061 params
->environment
,
4064 context
->environment
,
4067 *exit_status
= EXIT_MEMORY
;
4070 accum_env
= strv_env_clean(accum_env
);
4072 (void) umask(context
->umask
);
4074 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4076 *exit_status
= EXIT_KEYRING
;
4077 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4080 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
4081 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4083 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked for it, and the kernel doesn't actually support ambient caps */
4084 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4086 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not desired */
4087 if (needs_ambient_hack
)
4088 needs_setuid
= false;
4090 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4092 if (needs_sandboxing
) {
4093 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
4094 * present. The actual MAC context application will happen later, as late as possible, to avoid
4095 * impacting our own code paths. */
4098 use_selinux
= mac_selinux_use();
4101 use_smack
= mac_smack_use();
4104 use_apparmor
= mac_apparmor_use();
4108 if (needs_sandboxing
) {
4111 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4112 * is set here. (See below.) */
4114 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4116 *exit_status
= EXIT_LIMITS
;
4117 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4121 if (needs_setuid
&& context
->pam_name
&& username
) {
4122 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4123 * wins here. (See above.) */
4125 /* All fds passed in the fds array will be closed in the pam child process. */
4126 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4128 *exit_status
= EXIT_PAM
;
4129 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4132 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4133 if (ngids_after_pam
< 0) {
4134 *exit_status
= EXIT_MEMORY
;
4135 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4139 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4140 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4141 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4142 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4144 userns_set_up
= true;
4145 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4147 *exit_status
= EXIT_USER
;
4148 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4152 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4154 if (ns_type_supported(NAMESPACE_NET
)) {
4155 r
= setup_netns(runtime
->netns_storage_socket
);
4157 log_unit_warning_errno(unit
, r
,
4158 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4160 *exit_status
= EXIT_NETWORK
;
4161 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4163 } else if (context
->network_namespace_path
) {
4164 *exit_status
= EXIT_NETWORK
;
4165 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4166 "NetworkNamespacePath= is not supported, refusing.");
4168 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4171 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4172 if (needs_mount_namespace
) {
4173 _cleanup_free_
char *error_path
= NULL
;
4175 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4177 *exit_status
= EXIT_NAMESPACE
;
4178 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4179 error_path
? ": " : "", strempty(error_path
));
4183 if (needs_sandboxing
) {
4184 r
= apply_protect_hostname(unit
, context
, exit_status
);
4189 /* Drop groups as early as possible.
4190 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4191 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4193 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4194 int ngids_to_enforce
= 0;
4196 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4201 if (ngids_to_enforce
< 0) {
4202 *exit_status
= EXIT_MEMORY
;
4203 return log_unit_error_errno(unit
,
4205 "Failed to merge group lists. Group membership might be incorrect: %m");
4208 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4210 *exit_status
= EXIT_GROUP
;
4211 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4215 /* If the user namespace was not set up above, try to do it now.
4216 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4217 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4218 * case of mount namespaces being less privileged when the mount point list is copied from a
4219 * different user namespace). */
4221 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4222 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4224 *exit_status
= EXIT_USER
;
4225 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4229 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4232 _cleanup_free_
char *executable
= NULL
;
4233 r
= find_executable_full(command
->path
, false, &executable
, NULL
);
4235 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4236 log_struct_errno(LOG_INFO
, r
,
4237 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4239 LOG_UNIT_INVOCATION_ID(unit
),
4240 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4242 "EXECUTABLE=%s", command
->path
);
4246 *exit_status
= EXIT_EXEC
;
4247 return log_struct_errno(LOG_INFO
, r
,
4248 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4250 LOG_UNIT_INVOCATION_ID(unit
),
4251 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4253 "EXECUTABLE=%s", command
->path
);
4257 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
4258 r
= mac_selinux_get_child_mls_label(socket_fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4260 *exit_status
= EXIT_SELINUX_CONTEXT
;
4261 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4266 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4267 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
4268 * however if we have it as we want to keep it open until the final execve(). */
4270 r
= close_all_fds(keep_fds
, n_keep_fds
);
4272 r
= shift_fds(fds
, n_fds
);
4274 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4276 *exit_status
= EXIT_FDS
;
4277 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4280 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4281 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4282 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4285 secure_bits
= context
->secure_bits
;
4287 if (needs_sandboxing
) {
4290 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4291 * requested. (Note this is placed after the general resource limit initialization, see
4292 * above, in order to take precedence.) */
4293 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4294 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4295 *exit_status
= EXIT_LIMITS
;
4296 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4301 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4302 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4304 r
= setup_smack(context
, executable
);
4306 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4307 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4312 bset
= context
->capability_bounding_set
;
4313 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4314 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4315 * instead of us doing that */
4316 if (needs_ambient_hack
)
4317 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4318 (UINT64_C(1) << CAP_SETUID
) |
4319 (UINT64_C(1) << CAP_SETGID
);
4321 if (!cap_test_all(bset
)) {
4322 r
= capability_bounding_set_drop(bset
, false);
4324 *exit_status
= EXIT_CAPABILITIES
;
4325 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4329 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4331 * To be able to raise the ambient capabilities after setresuid() they have to be
4332 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4333 * After setresuid() the ambient capabilities can be raised as they are present in
4334 * the permitted and inhertiable set. However it is possible that someone wants to
4335 * set ambient capabilities without changing the user, so we also set the ambient
4336 * capabilities here.
4337 * The requested ambient capabilities are raised in the inheritable set if the
4338 * second argument is true. */
4339 if (!needs_ambient_hack
) {
4340 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4342 *exit_status
= EXIT_CAPABILITIES
;
4343 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4348 /* chroot to root directory first, before we lose the ability to chroot */
4349 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4351 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4354 if (uid_is_valid(uid
)) {
4355 r
= enforce_user(context
, uid
);
4357 *exit_status
= EXIT_USER
;
4358 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4361 if (!needs_ambient_hack
&&
4362 context
->capability_ambient_set
!= 0) {
4364 /* Raise the ambient capabilities after user change. */
4365 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4367 *exit_status
= EXIT_CAPABILITIES
;
4368 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4374 /* Apply working directory here, because the working directory might be on NFS and only the user running
4375 * this service might have the correct privilege to change to the working directory */
4376 r
= apply_working_directory(context
, params
, home
, exit_status
);
4378 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4380 if (needs_sandboxing
) {
4381 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4382 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4383 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4384 * are restricted. */
4388 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4391 r
= setexeccon(exec_context
);
4393 *exit_status
= EXIT_SELINUX_CONTEXT
;
4394 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4401 if (use_apparmor
&& context
->apparmor_profile
) {
4402 r
= aa_change_onexec(context
->apparmor_profile
);
4403 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4404 *exit_status
= EXIT_APPARMOR_PROFILE
;
4405 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4410 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4411 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4413 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4414 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4415 * effective set here.
4416 * The effective set is overwritten during execve with the following values:
4417 * - ambient set (for non-root processes)
4418 * - (inheritable | bounding) set for root processes)
4420 * Hence there is no security impact to raise it in the effective set before execve
4422 r
= capability_gain_cap_setpcap(NULL
);
4424 *exit_status
= EXIT_CAPABILITIES
;
4425 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
4427 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
4428 *exit_status
= EXIT_SECUREBITS
;
4429 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
4433 if (context_has_no_new_privileges(context
))
4434 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
4435 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
4436 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
4440 r
= apply_address_families(unit
, context
);
4442 *exit_status
= EXIT_ADDRESS_FAMILIES
;
4443 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
4446 r
= apply_memory_deny_write_execute(unit
, context
);
4448 *exit_status
= EXIT_SECCOMP
;
4449 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
4452 r
= apply_restrict_realtime(unit
, context
);
4454 *exit_status
= EXIT_SECCOMP
;
4455 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
4458 r
= apply_restrict_suid_sgid(unit
, context
);
4460 *exit_status
= EXIT_SECCOMP
;
4461 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
4464 r
= apply_restrict_namespaces(unit
, context
);
4466 *exit_status
= EXIT_SECCOMP
;
4467 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
4470 r
= apply_protect_sysctl(unit
, context
);
4472 *exit_status
= EXIT_SECCOMP
;
4473 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
4476 r
= apply_protect_kernel_modules(unit
, context
);
4478 *exit_status
= EXIT_SECCOMP
;
4479 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
4482 r
= apply_protect_kernel_logs(unit
, context
);
4484 *exit_status
= EXIT_SECCOMP
;
4485 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
4488 r
= apply_protect_clock(unit
, context
);
4490 *exit_status
= EXIT_SECCOMP
;
4491 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
4494 r
= apply_private_devices(unit
, context
);
4496 *exit_status
= EXIT_SECCOMP
;
4497 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
4500 r
= apply_syscall_archs(unit
, context
);
4502 *exit_status
= EXIT_SECCOMP
;
4503 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
4506 r
= apply_lock_personality(unit
, context
);
4508 *exit_status
= EXIT_SECCOMP
;
4509 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
4512 r
= apply_syscall_log(unit
, context
);
4514 *exit_status
= EXIT_SECCOMP
;
4515 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
4518 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
4519 * by the filter as little as possible. */
4520 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
4522 *exit_status
= EXIT_SECCOMP
;
4523 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
4528 if (!strv_isempty(context
->unset_environment
)) {
4531 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
4533 *exit_status
= EXIT_MEMORY
;
4537 strv_free_and_replace(accum_env
, ee
);
4540 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
4541 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
4542 if (!replaced_argv
) {
4543 *exit_status
= EXIT_MEMORY
;
4546 final_argv
= replaced_argv
;
4548 final_argv
= command
->argv
;
4550 if (DEBUG_LOGGING
) {
4551 _cleanup_free_
char *line
;
4553 line
= exec_command_line(final_argv
);
4555 log_struct(LOG_DEBUG
,
4556 "EXECUTABLE=%s", executable
,
4557 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
4559 LOG_UNIT_INVOCATION_ID(unit
));
4565 /* We have finished with all our initializations. Let's now let the manager know that. From this point
4566 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
4568 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4569 *exit_status
= EXIT_EXEC
;
4570 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
4574 execve(executable
, final_argv
, accum_env
);
4580 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4581 * that POLLHUP on it no longer means execve() succeeded. */
4583 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4584 *exit_status
= EXIT_EXEC
;
4585 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4589 *exit_status
= EXIT_EXEC
;
4590 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
4593 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4594 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4596 int exec_spawn(Unit
*unit
,
4597 ExecCommand
*command
,
4598 const ExecContext
*context
,
4599 const ExecParameters
*params
,
4600 ExecRuntime
*runtime
,
4601 DynamicCreds
*dcreds
,
4604 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4605 _cleanup_free_
char *subcgroup_path
= NULL
;
4606 _cleanup_strv_free_
char **files_env
= NULL
;
4607 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4608 _cleanup_free_
char *line
= NULL
;
4616 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4618 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4619 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4620 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4622 if (params
->n_socket_fds
> 1) {
4623 log_unit_error(unit
, "Got more than one socket.");
4627 if (params
->n_socket_fds
== 0) {
4628 log_unit_error(unit
, "Got no socket.");
4632 socket_fd
= params
->fds
[0];
4636 n_socket_fds
= params
->n_socket_fds
;
4637 n_storage_fds
= params
->n_storage_fds
;
4640 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4642 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4644 r
= exec_context_load_environment(unit
, context
, &files_env
);
4646 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4648 line
= exec_command_line(command
->argv
);
4652 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
4653 and, until the next SELinux policy changes, we save further reloads in future children. */
4654 mac_selinux_maybe_reload();
4656 log_struct(LOG_DEBUG
,
4657 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
4658 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
4659 the mount namespace in the child, but we want to log
4660 from the parent, so we need to use the (possibly
4661 inaccurate) path here. */
4663 LOG_UNIT_INVOCATION_ID(unit
));
4665 if (params
->cgroup_path
) {
4666 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4668 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4669 if (r
> 0) { /* We are using a child cgroup */
4670 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4672 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4678 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4681 int exit_status
= EXIT_SUCCESS
;
4683 r
= exec_child(unit
,
4695 unit
->manager
->user_lookup_fds
[1],
4699 const char *status
=
4700 exit_status_to_string(exit_status
,
4701 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4703 log_struct_errno(LOG_ERR
, r
,
4704 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4706 LOG_UNIT_INVOCATION_ID(unit
),
4707 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4708 status
, command
->path
),
4709 "EXECUTABLE=%s", command
->path
);
4715 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4717 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4718 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4719 * process will be killed too). */
4721 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4723 exec_status_start(&command
->exec_status
, pid
);
4729 void exec_context_init(ExecContext
*c
) {
4733 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4734 c
->cpu_sched_policy
= SCHED_OTHER
;
4735 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4736 c
->syslog_level_prefix
= true;
4737 c
->ignore_sigpipe
= true;
4738 c
->timer_slack_nsec
= NSEC_INFINITY
;
4739 c
->personality
= PERSONALITY_INVALID
;
4740 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4741 c
->directories
[t
].mode
= 0755;
4742 c
->timeout_clean_usec
= USEC_INFINITY
;
4743 c
->capability_bounding_set
= CAP_ALL
;
4744 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4745 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4746 c
->log_level_max
= -1;
4748 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
4750 numa_policy_reset(&c
->numa_policy
);
4753 void exec_context_done(ExecContext
*c
) {
4756 c
->environment
= strv_free(c
->environment
);
4757 c
->environment_files
= strv_free(c
->environment_files
);
4758 c
->pass_environment
= strv_free(c
->pass_environment
);
4759 c
->unset_environment
= strv_free(c
->unset_environment
);
4761 rlimit_free_all(c
->rlimit
);
4763 for (size_t l
= 0; l
< 3; l
++) {
4764 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4765 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4768 c
->working_directory
= mfree(c
->working_directory
);
4769 c
->root_directory
= mfree(c
->root_directory
);
4770 c
->root_image
= mfree(c
->root_image
);
4771 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
4772 c
->root_hash
= mfree(c
->root_hash
);
4773 c
->root_hash_size
= 0;
4774 c
->root_hash_path
= mfree(c
->root_hash_path
);
4775 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
4776 c
->root_hash_sig_size
= 0;
4777 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
4778 c
->root_verity
= mfree(c
->root_verity
);
4779 c
->tty_path
= mfree(c
->tty_path
);
4780 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4781 c
->user
= mfree(c
->user
);
4782 c
->group
= mfree(c
->group
);
4784 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4786 c
->pam_name
= mfree(c
->pam_name
);
4788 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4789 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4790 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4792 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4793 c
->bind_mounts
= NULL
;
4794 c
->n_bind_mounts
= 0;
4795 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4796 c
->temporary_filesystems
= NULL
;
4797 c
->n_temporary_filesystems
= 0;
4798 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
4800 cpu_set_reset(&c
->cpu_set
);
4801 numa_policy_reset(&c
->numa_policy
);
4803 c
->utmp_id
= mfree(c
->utmp_id
);
4804 c
->selinux_context
= mfree(c
->selinux_context
);
4805 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4806 c
->smack_process_label
= mfree(c
->smack_process_label
);
4808 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4809 c
->syscall_archs
= set_free(c
->syscall_archs
);
4810 c
->address_families
= set_free(c
->address_families
);
4812 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4813 c
->directories
[t
].paths
= strv_free(c
->directories
[t
].paths
);
4815 c
->log_level_max
= -1;
4817 exec_context_free_log_extra_fields(c
);
4819 c
->log_ratelimit_interval_usec
= 0;
4820 c
->log_ratelimit_burst
= 0;
4822 c
->stdin_data
= mfree(c
->stdin_data
);
4823 c
->stdin_data_size
= 0;
4825 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4827 c
->log_namespace
= mfree(c
->log_namespace
);
4829 c
->load_credentials
= strv_free(c
->load_credentials
);
4830 c
->set_credentials
= hashmap_free(c
->set_credentials
);
4833 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4838 if (!runtime_prefix
)
4841 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4842 _cleanup_free_
char *p
;
4844 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4845 p
= path_join(runtime_prefix
, "private", *i
);
4847 p
= path_join(runtime_prefix
, *i
);
4851 /* We execute this synchronously, since we need to be sure this is gone when we start the
4853 (void) rm_rf(p
, REMOVE_ROOT
);
4859 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
4860 _cleanup_free_
char *p
= NULL
;
4864 if (!runtime_prefix
|| !unit
)
4867 p
= path_join(runtime_prefix
, "credentials", unit
);
4871 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
4872 * unmount it, and afterwards remove the mount point */
4873 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
4874 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
4879 static void exec_command_done(ExecCommand
*c
) {
4882 c
->path
= mfree(c
->path
);
4883 c
->argv
= strv_free(c
->argv
);
4886 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4889 for (i
= 0; i
< n
; i
++)
4890 exec_command_done(c
+i
);
4893 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4897 LIST_REMOVE(command
, c
, i
);
4898 exec_command_done(i
);
4905 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4906 for (size_t i
= 0; i
< n
; i
++)
4907 c
[i
] = exec_command_free_list(c
[i
]);
4910 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4911 for (size_t i
= 0; i
< n
; i
++)
4912 exec_status_reset(&c
[i
].exec_status
);
4915 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
4916 for (size_t i
= 0; i
< n
; i
++) {
4919 LIST_FOREACH(command
, z
, c
[i
])
4920 exec_status_reset(&z
->exec_status
);
4924 typedef struct InvalidEnvInfo
{
4929 static void invalid_env(const char *p
, void *userdata
) {
4930 InvalidEnvInfo
*info
= userdata
;
4932 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4935 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4941 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4944 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4947 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4950 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4953 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4956 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4963 static int exec_context_named_iofds(
4964 const ExecContext
*c
,
4965 const ExecParameters
*p
,
4966 int named_iofds
[static 3]) {
4969 const char* stdio_fdname
[3];
4974 assert(named_iofds
);
4976 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4977 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4978 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4980 for (size_t i
= 0; i
< 3; i
++)
4981 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4983 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4985 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
4986 if (named_iofds
[STDIN_FILENO
] < 0 &&
4987 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4988 stdio_fdname
[STDIN_FILENO
] &&
4989 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4991 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4994 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4995 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4996 stdio_fdname
[STDOUT_FILENO
] &&
4997 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4999 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5002 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5003 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5004 stdio_fdname
[STDERR_FILENO
] &&
5005 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5007 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5011 return targets
== 0 ? 0 : -ENOENT
;
5014 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
5015 char **i
, **r
= NULL
;
5020 STRV_FOREACH(i
, c
->environment_files
) {
5023 bool ignore
= false;
5025 _cleanup_globfree_ glob_t pglob
= {};
5034 if (!path_is_absolute(fn
)) {
5042 /* Filename supports globbing, take all matching files */
5043 k
= safe_glob(fn
, 0, &pglob
);
5052 /* When we don't match anything, -ENOENT should be returned */
5053 assert(pglob
.gl_pathc
> 0);
5055 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5056 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5064 /* Log invalid environment variables with filename */
5066 InvalidEnvInfo info
= {
5068 .path
= pglob
.gl_pathv
[n
]
5071 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5079 m
= strv_env_merge(2, r
, p
);
5095 static bool tty_may_match_dev_console(const char *tty
) {
5096 _cleanup_free_
char *resolved
= NULL
;
5101 tty
= skip_dev_prefix(tty
);
5103 /* trivial identity? */
5104 if (streq(tty
, "console"))
5107 if (resolve_dev_console(&resolved
) < 0)
5108 return true; /* if we could not resolve, assume it may */
5110 /* "tty0" means the active VC, so it may be the same sometimes */
5111 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5114 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5117 return ec
->tty_reset
||
5119 ec
->tty_vt_disallocate
||
5120 is_terminal_input(ec
->std_input
) ||
5121 is_terminal_output(ec
->std_output
) ||
5122 is_terminal_output(ec
->std_error
);
5125 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5127 return exec_context_may_touch_tty(ec
) &&
5128 tty_may_match_dev_console(exec_context_tty_path(ec
));
5131 static void strv_fprintf(FILE *f
, char **l
) {
5137 fprintf(f
, " %s", *g
);
5140 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5141 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
5147 prefix
= strempty(prefix
);
5151 "%sWorkingDirectory: %s\n"
5152 "%sRootDirectory: %s\n"
5153 "%sNonBlocking: %s\n"
5154 "%sPrivateTmp: %s\n"
5155 "%sPrivateDevices: %s\n"
5156 "%sProtectKernelTunables: %s\n"
5157 "%sProtectKernelModules: %s\n"
5158 "%sProtectKernelLogs: %s\n"
5159 "%sProtectClock: %s\n"
5160 "%sProtectControlGroups: %s\n"
5161 "%sPrivateNetwork: %s\n"
5162 "%sPrivateUsers: %s\n"
5163 "%sProtectHome: %s\n"
5164 "%sProtectSystem: %s\n"
5165 "%sMountAPIVFS: %s\n"
5166 "%sIgnoreSIGPIPE: %s\n"
5167 "%sMemoryDenyWriteExecute: %s\n"
5168 "%sRestrictRealtime: %s\n"
5169 "%sRestrictSUIDSGID: %s\n"
5170 "%sKeyringMode: %s\n"
5171 "%sProtectHostname: %s\n"
5172 "%sProtectProc: %s\n"
5173 "%sProcSubset: %s\n",
5175 prefix
, empty_to_root(c
->working_directory
),
5176 prefix
, empty_to_root(c
->root_directory
),
5177 prefix
, yes_no(c
->non_blocking
),
5178 prefix
, yes_no(c
->private_tmp
),
5179 prefix
, yes_no(c
->private_devices
),
5180 prefix
, yes_no(c
->protect_kernel_tunables
),
5181 prefix
, yes_no(c
->protect_kernel_modules
),
5182 prefix
, yes_no(c
->protect_kernel_logs
),
5183 prefix
, yes_no(c
->protect_clock
),
5184 prefix
, yes_no(c
->protect_control_groups
),
5185 prefix
, yes_no(c
->private_network
),
5186 prefix
, yes_no(c
->private_users
),
5187 prefix
, protect_home_to_string(c
->protect_home
),
5188 prefix
, protect_system_to_string(c
->protect_system
),
5189 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
5190 prefix
, yes_no(c
->ignore_sigpipe
),
5191 prefix
, yes_no(c
->memory_deny_write_execute
),
5192 prefix
, yes_no(c
->restrict_realtime
),
5193 prefix
, yes_no(c
->restrict_suid_sgid
),
5194 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5195 prefix
, yes_no(c
->protect_hostname
),
5196 prefix
, protect_proc_to_string(c
->protect_proc
),
5197 prefix
, proc_subset_to_string(c
->proc_subset
));
5200 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5202 if (c
->root_image_options
) {
5205 fprintf(f
, "%sRootImageOptions:", prefix
);
5206 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5207 if (!isempty(o
->options
))
5208 fprintf(f
, " %s:%s",
5209 partition_designator_to_string(o
->partition_designator
),
5215 _cleanup_free_
char *encoded
= NULL
;
5216 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5218 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5221 if (c
->root_hash_path
)
5222 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5224 if (c
->root_hash_sig
) {
5225 _cleanup_free_
char *encoded
= NULL
;
5227 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5229 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5232 if (c
->root_hash_sig_path
)
5233 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5236 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5238 STRV_FOREACH(e
, c
->environment
)
5239 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5241 STRV_FOREACH(e
, c
->environment_files
)
5242 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5244 STRV_FOREACH(e
, c
->pass_environment
)
5245 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5247 STRV_FOREACH(e
, c
->unset_environment
)
5248 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5250 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5252 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5253 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5255 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
5256 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
5260 "%sTimeoutCleanSec: %s\n",
5261 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
5268 if (c
->oom_score_adjust_set
)
5270 "%sOOMScoreAdjust: %i\n",
5271 prefix
, c
->oom_score_adjust
);
5273 if (c
->coredump_filter_set
)
5275 "%sCoredumpFilter: 0x%"PRIx64
"\n",
5276 prefix
, c
->coredump_filter
);
5278 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5280 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5281 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5282 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5283 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5286 if (c
->ioprio_set
) {
5287 _cleanup_free_
char *class_str
= NULL
;
5289 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
5291 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5293 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
5296 if (c
->cpu_sched_set
) {
5297 _cleanup_free_
char *policy_str
= NULL
;
5299 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5301 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5304 "%sCPUSchedulingPriority: %i\n"
5305 "%sCPUSchedulingResetOnFork: %s\n",
5306 prefix
, c
->cpu_sched_priority
,
5307 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5310 if (c
->cpu_set
.set
) {
5311 _cleanup_free_
char *affinity
= NULL
;
5313 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5314 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5317 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5318 _cleanup_free_
char *nodes
= NULL
;
5320 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5321 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5322 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5325 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5326 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5329 "%sStandardInput: %s\n"
5330 "%sStandardOutput: %s\n"
5331 "%sStandardError: %s\n",
5332 prefix
, exec_input_to_string(c
->std_input
),
5333 prefix
, exec_output_to_string(c
->std_output
),
5334 prefix
, exec_output_to_string(c
->std_error
));
5336 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5337 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5338 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5339 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5340 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5341 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5343 if (c
->std_input
== EXEC_INPUT_FILE
)
5344 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5345 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5346 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5347 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5348 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5349 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5350 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5351 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5352 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5358 "%sTTYVHangup: %s\n"
5359 "%sTTYVTDisallocate: %s\n",
5360 prefix
, c
->tty_path
,
5361 prefix
, yes_no(c
->tty_reset
),
5362 prefix
, yes_no(c
->tty_vhangup
),
5363 prefix
, yes_no(c
->tty_vt_disallocate
));
5365 if (IN_SET(c
->std_output
,
5367 EXEC_OUTPUT_JOURNAL
,
5368 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5369 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5370 IN_SET(c
->std_error
,
5372 EXEC_OUTPUT_JOURNAL
,
5373 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5374 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5376 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5378 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5380 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5382 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5384 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5387 if (c
->log_level_max
>= 0) {
5388 _cleanup_free_
char *t
= NULL
;
5390 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5392 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5395 if (c
->log_ratelimit_interval_usec
> 0) {
5396 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
5399 "%sLogRateLimitIntervalSec: %s\n",
5400 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
5403 if (c
->log_ratelimit_burst
> 0)
5404 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
5406 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
5407 fprintf(f
, "%sLogExtraFields: ", prefix
);
5408 fwrite(c
->log_extra_fields
[j
].iov_base
,
5409 1, c
->log_extra_fields
[j
].iov_len
,
5414 if (c
->log_namespace
)
5415 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
5417 if (c
->secure_bits
) {
5418 _cleanup_free_
char *str
= NULL
;
5420 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
5422 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
5425 if (c
->capability_bounding_set
!= CAP_ALL
) {
5426 _cleanup_free_
char *str
= NULL
;
5428 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
5430 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
5433 if (c
->capability_ambient_set
!= 0) {
5434 _cleanup_free_
char *str
= NULL
;
5436 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
5438 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
5442 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
5444 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
5446 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
5448 if (!strv_isempty(c
->supplementary_groups
)) {
5449 fprintf(f
, "%sSupplementaryGroups:", prefix
);
5450 strv_fprintf(f
, c
->supplementary_groups
);
5455 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
5457 if (!strv_isempty(c
->read_write_paths
)) {
5458 fprintf(f
, "%sReadWritePaths:", prefix
);
5459 strv_fprintf(f
, c
->read_write_paths
);
5463 if (!strv_isempty(c
->read_only_paths
)) {
5464 fprintf(f
, "%sReadOnlyPaths:", prefix
);
5465 strv_fprintf(f
, c
->read_only_paths
);
5469 if (!strv_isempty(c
->inaccessible_paths
)) {
5470 fprintf(f
, "%sInaccessiblePaths:", prefix
);
5471 strv_fprintf(f
, c
->inaccessible_paths
);
5475 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
5476 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
5477 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
5478 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
5479 c
->bind_mounts
[i
].source
,
5480 c
->bind_mounts
[i
].destination
,
5481 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
5483 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
5484 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
5486 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
5488 isempty(t
->options
) ? "" : ":",
5489 strempty(t
->options
));
5494 "%sUtmpIdentifier: %s\n",
5495 prefix
, c
->utmp_id
);
5497 if (c
->selinux_context
)
5499 "%sSELinuxContext: %s%s\n",
5500 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
5502 if (c
->apparmor_profile
)
5504 "%sAppArmorProfile: %s%s\n",
5505 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
5507 if (c
->smack_process_label
)
5509 "%sSmackProcessLabel: %s%s\n",
5510 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
5512 if (c
->personality
!= PERSONALITY_INVALID
)
5514 "%sPersonality: %s\n",
5515 prefix
, strna(personality_to_string(c
->personality
)));
5518 "%sLockPersonality: %s\n",
5519 prefix
, yes_no(c
->lock_personality
));
5521 if (c
->syscall_filter
) {
5528 "%sSystemCallFilter: ",
5531 if (!c
->syscall_allow_list
)
5535 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
5536 _cleanup_free_
char *name
= NULL
;
5537 const char *errno_name
= NULL
;
5538 int num
= PTR_TO_INT(val
);
5545 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
5546 fputs(strna(name
), f
);
5549 errno_name
= seccomp_errno_or_action_to_string(num
);
5551 fprintf(f
, ":%s", errno_name
);
5553 fprintf(f
, ":%d", num
);
5561 if (c
->syscall_archs
) {
5567 "%sSystemCallArchitectures:",
5571 SET_FOREACH(id
, c
->syscall_archs
)
5572 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
5577 if (exec_context_restrict_namespaces_set(c
)) {
5578 _cleanup_free_
char *s
= NULL
;
5580 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
5582 fprintf(f
, "%sRestrictNamespaces: %s\n",
5586 if (c
->network_namespace_path
)
5588 "%sNetworkNamespacePath: %s\n",
5589 prefix
, c
->network_namespace_path
);
5591 if (c
->syscall_errno
> 0) {
5593 const char *errno_name
;
5596 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
5599 errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
5601 fputs(errno_name
, f
);
5603 fprintf(f
, "%d", c
->syscall_errno
);
5608 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
5611 fprintf(f
, "%sMountImages: %s%s:%s%s", prefix
,
5612 c
->mount_images
[i
].ignore_enoent
? "-": "",
5613 c
->mount_images
[i
].source
,
5614 c
->mount_images
[i
].destination
,
5615 LIST_IS_EMPTY(c
->mount_images
[i
].mount_options
) ? "": ":");
5616 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
5618 partition_designator_to_string(o
->partition_designator
),
5624 bool exec_context_maintains_privileges(const ExecContext
*c
) {
5627 /* Returns true if the process forked off would run under
5628 * an unchanged UID or as root. */
5633 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
5639 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
5647 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5649 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5654 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
5657 /* Explicit setting wins */
5658 if (c
->mount_apivfs_set
)
5659 return c
->mount_apivfs
;
5661 /* Default to "yes" if root directory or image are specified */
5662 if (exec_context_with_rootfs(c
))
5668 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5671 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
5672 free(c
->log_extra_fields
[l
].iov_base
);
5673 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5674 c
->n_log_extra_fields
= 0;
5677 void exec_context_revert_tty(ExecContext
*c
) {
5682 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5683 exec_context_tty_reset(c
, NULL
);
5685 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5686 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5687 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5689 if (exec_context_may_touch_tty(c
)) {
5692 path
= exec_context_tty_path(c
);
5694 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
5695 if (r
< 0 && r
!= -ENOENT
)
5696 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5701 int exec_context_get_clean_directories(
5707 _cleanup_strv_free_
char **l
= NULL
;
5714 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5717 if (!FLAGS_SET(mask
, 1U << t
))
5723 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5726 j
= path_join(prefix
[t
], *i
);
5730 r
= strv_consume(&l
, j
);
5734 /* Also remove private directories unconditionally. */
5735 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5736 j
= path_join(prefix
[t
], "private", *i
);
5740 r
= strv_consume(&l
, j
);
5751 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5752 ExecCleanMask mask
= 0;
5757 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5758 if (!strv_isempty(c
->directories
[t
].paths
))
5765 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5772 dual_timestamp_get(&s
->start_timestamp
);
5775 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5783 dual_timestamp_get(&s
->exit_timestamp
);
5788 if (context
&& context
->utmp_id
)
5789 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5792 void exec_status_reset(ExecStatus
*s
) {
5795 *s
= (ExecStatus
) {};
5798 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5799 char buf
[FORMAT_TIMESTAMP_MAX
];
5807 prefix
= strempty(prefix
);
5810 "%sPID: "PID_FMT
"\n",
5813 if (dual_timestamp_is_set(&s
->start_timestamp
))
5815 "%sStart Timestamp: %s\n",
5816 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5818 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5820 "%sExit Timestamp: %s\n"
5822 "%sExit Status: %i\n",
5823 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5824 prefix
, sigchld_code_to_string(s
->code
),
5828 static char *exec_command_line(char **argv
) {
5836 STRV_FOREACH(a
, argv
)
5844 STRV_FOREACH(a
, argv
) {
5851 if (strpbrk(*a
, WHITESPACE
)) {
5862 /* FIXME: this doesn't really handle arguments that have
5863 * spaces and ticks in them */
5868 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5869 _cleanup_free_
char *cmd
= NULL
;
5870 const char *prefix2
;
5875 prefix
= strempty(prefix
);
5876 prefix2
= strjoina(prefix
, "\t");
5878 cmd
= exec_command_line(c
->argv
);
5880 "%sCommand Line: %s\n",
5881 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
5883 exec_status_dump(&c
->exec_status
, f
, prefix2
);
5886 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5889 prefix
= strempty(prefix
);
5891 LIST_FOREACH(command
, c
, c
)
5892 exec_command_dump(c
, f
, prefix
);
5895 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
5902 /* It's kind of important, that we keep the order here */
5903 LIST_FIND_TAIL(command
, *l
, end
);
5904 LIST_INSERT_AFTER(command
, *l
, end
, e
);
5909 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
5917 l
= strv_new_ap(path
, ap
);
5929 free_and_replace(c
->path
, p
);
5931 return strv_free_and_replace(c
->argv
, l
);
5934 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
5935 _cleanup_strv_free_
char **l
= NULL
;
5943 l
= strv_new_ap(path
, ap
);
5949 r
= strv_extend_strv(&c
->argv
, l
, false);
5956 static void *remove_tmpdir_thread(void *p
) {
5957 _cleanup_free_
char *path
= p
;
5959 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
5963 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
5970 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
5972 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
5974 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
5975 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
5977 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
5979 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
5984 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
5985 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
5987 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
5989 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
5991 rt
->var_tmp_dir
= NULL
;
5994 rt
->id
= mfree(rt
->id
);
5995 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
5996 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
5997 safe_close_pair(rt
->netns_storage_socket
);
6001 static void exec_runtime_freep(ExecRuntime
**rt
) {
6002 (void) exec_runtime_free(*rt
, false);
6005 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
6006 _cleanup_free_
char *id_copy
= NULL
;
6011 id_copy
= strdup(id
);
6015 n
= new(ExecRuntime
, 1);
6019 *n
= (ExecRuntime
) {
6020 .id
= TAKE_PTR(id_copy
),
6021 .netns_storage_socket
= { -1, -1 },
6028 static int exec_runtime_add(
6033 int netns_storage_socket
[2],
6034 ExecRuntime
**ret
) {
6036 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6042 /* tmp_dir, var_tmp_dir, netns_storage_socket fds are donated on success */
6044 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
6048 r
= exec_runtime_allocate(&rt
, id
);
6052 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
6056 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6057 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6058 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6060 if (netns_storage_socket
) {
6061 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6062 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6069 /* do not remove created ExecRuntime object when the operation succeeds. */
6074 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
6075 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6076 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
6083 /* It is not necessary to create ExecRuntime object. */
6084 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
6087 if (c
->private_tmp
&&
6088 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6089 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6090 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6091 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6096 if (c
->private_network
|| c
->network_namespace_path
) {
6097 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6101 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ret
);
6108 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6116 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6118 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
6124 /* If not found, then create a new object. */
6125 r
= exec_runtime_make(m
, c
, id
, &rt
);
6127 /* When r == 0, it is not necessary to create ExecRuntime object. */
6131 /* increment reference counter. */
6137 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6141 assert(rt
->n_ref
> 0);
6147 return exec_runtime_free(rt
, destroy
);
6150 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6157 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6158 fprintf(f
, "exec-runtime=%s", rt
->id
);
6161 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6163 if (rt
->var_tmp_dir
)
6164 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6166 if (rt
->netns_storage_socket
[0] >= 0) {
6169 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6173 fprintf(f
, " netns-socket-0=%i", copy
);
6176 if (rt
->netns_storage_socket
[1] >= 0) {
6179 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6183 fprintf(f
, " netns-socket-1=%i", copy
);
6192 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6193 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6197 /* This is for the migration from old (v237 or earlier) deserialization text.
6198 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6199 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6200 * so or not from the serialized text, then we always creates a new object owned by this. */
6206 /* Manager manages ExecRuntime objects by the unit id.
6207 * So, we omit the serialized text when the unit does not have id (yet?)... */
6208 if (isempty(u
->id
)) {
6209 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6213 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
6215 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
6219 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6221 r
= exec_runtime_allocate(&rt_create
, u
->id
);
6228 if (streq(key
, "tmp-dir")) {
6231 copy
= strdup(value
);
6235 free_and_replace(rt
->tmp_dir
, copy
);
6237 } else if (streq(key
, "var-tmp-dir")) {
6240 copy
= strdup(value
);
6244 free_and_replace(rt
->var_tmp_dir
, copy
);
6246 } else if (streq(key
, "netns-socket-0")) {
6249 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6250 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6254 safe_close(rt
->netns_storage_socket
[0]);
6255 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6257 } else if (streq(key
, "netns-socket-1")) {
6260 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6261 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6265 safe_close(rt
->netns_storage_socket
[1]);
6266 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6270 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6272 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6274 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6278 rt_create
->manager
= u
->manager
;
6281 TAKE_PTR(rt_create
);
6287 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6288 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6290 int r
, fdpair
[] = {-1, -1};
6291 const char *p
, *v
= value
;
6298 n
= strcspn(v
, " ");
6299 id
= strndupa(v
, n
);
6304 v
= startswith(p
, "tmp-dir=");
6306 n
= strcspn(v
, " ");
6307 tmp_dir
= strndup(v
, n
);
6315 v
= startswith(p
, "var-tmp-dir=");
6317 n
= strcspn(v
, " ");
6318 var_tmp_dir
= strndup(v
, n
);
6326 v
= startswith(p
, "netns-socket-0=");
6330 n
= strcspn(v
, " ");
6331 buf
= strndupa(v
, n
);
6333 r
= safe_atoi(buf
, &fdpair
[0]);
6335 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6336 if (!fdset_contains(fds
, fdpair
[0]))
6337 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6338 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", fdpair
[0]);
6339 fdpair
[0] = fdset_remove(fds
, fdpair
[0]);
6345 v
= startswith(p
, "netns-socket-1=");
6349 n
= strcspn(v
, " ");
6350 buf
= strndupa(v
, n
);
6351 r
= safe_atoi(buf
, &fdpair
[1]);
6353 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6354 if (!fdset_contains(fds
, fdpair
[0]))
6355 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6356 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", fdpair
[1]);
6357 fdpair
[1] = fdset_remove(fds
, fdpair
[1]);
6361 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, fdpair
, NULL
);
6363 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
6367 void exec_runtime_vacuum(Manager
*m
) {
6372 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
6374 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6378 (void) exec_runtime_free(rt
, false);
6382 void exec_params_clear(ExecParameters
*p
) {
6386 p
->environment
= strv_free(p
->environment
);
6387 p
->fd_names
= strv_free(p
->fd_names
);
6388 p
->fds
= mfree(p
->fds
);
6389 p
->exec_fd
= safe_close(p
->exec_fd
);
6392 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
6401 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
6403 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
6404 [EXEC_INPUT_NULL
] = "null",
6405 [EXEC_INPUT_TTY
] = "tty",
6406 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
6407 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
6408 [EXEC_INPUT_SOCKET
] = "socket",
6409 [EXEC_INPUT_NAMED_FD
] = "fd",
6410 [EXEC_INPUT_DATA
] = "data",
6411 [EXEC_INPUT_FILE
] = "file",
6414 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
6416 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
6417 [EXEC_OUTPUT_INHERIT
] = "inherit",
6418 [EXEC_OUTPUT_NULL
] = "null",
6419 [EXEC_OUTPUT_TTY
] = "tty",
6420 [EXEC_OUTPUT_KMSG
] = "kmsg",
6421 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
6422 [EXEC_OUTPUT_JOURNAL
] = "journal",
6423 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
6424 [EXEC_OUTPUT_SOCKET
] = "socket",
6425 [EXEC_OUTPUT_NAMED_FD
] = "fd",
6426 [EXEC_OUTPUT_FILE
] = "file",
6427 [EXEC_OUTPUT_FILE_APPEND
] = "append",
6430 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
6432 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
6433 [EXEC_UTMP_INIT
] = "init",
6434 [EXEC_UTMP_LOGIN
] = "login",
6435 [EXEC_UTMP_USER
] = "user",
6438 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
6440 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
6441 [EXEC_PRESERVE_NO
] = "no",
6442 [EXEC_PRESERVE_YES
] = "yes",
6443 [EXEC_PRESERVE_RESTART
] = "restart",
6446 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
6448 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
6449 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6450 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
6451 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
6452 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
6453 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
6454 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
6457 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
6459 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
6460 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
6461 * directories, specifically .timer units with their timestamp touch file. */
6462 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6463 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
6464 [EXEC_DIRECTORY_STATE
] = "state",
6465 [EXEC_DIRECTORY_CACHE
] = "cache",
6466 [EXEC_DIRECTORY_LOGS
] = "logs",
6467 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
6470 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
6472 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
6473 * the service payload in. */
6474 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6475 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
6476 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
6477 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
6478 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
6479 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
6482 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
6484 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
6485 [EXEC_KEYRING_INHERIT
] = "inherit",
6486 [EXEC_KEYRING_PRIVATE
] = "private",
6487 [EXEC_KEYRING_SHARED
] = "shared",
6490 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);