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 those fds
1251 * 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_prefix(c
->directories
[t
].paths
, ":", pre
);
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 (context
->mount_apivfs
)
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 !(context
->root_directory
|| context
->root_image
)) {
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 (context
->root_directory
|| context
->root_image
))
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
= context
->mount_apivfs
,
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
;
3248 } else if (context
->working_directory
)
3249 wd
= context
->working_directory
;
3253 if (params
->flags
& EXEC_APPLY_CHROOT
)
3256 d
= prefix_roota(context
->root_directory
, wd
);
3258 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3259 *exit_status
= EXIT_CHDIR
;
3266 static int apply_root_directory(
3267 const ExecContext
*context
,
3268 const ExecParameters
*params
,
3269 const bool needs_mount_ns
,
3273 assert(exit_status
);
3275 if (params
->flags
& EXEC_APPLY_CHROOT
)
3276 if (!needs_mount_ns
&& context
->root_directory
)
3277 if (chroot(context
->root_directory
) < 0) {
3278 *exit_status
= EXIT_CHROOT
;
3285 static int setup_keyring(
3287 const ExecContext
*context
,
3288 const ExecParameters
*p
,
3289 uid_t uid
, gid_t gid
) {
3291 key_serial_t keyring
;
3300 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3301 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3302 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3303 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3304 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3305 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3307 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3310 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3311 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3312 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3313 * & group is just as nasty as acquiring a reference to the user keyring. */
3315 saved_uid
= getuid();
3316 saved_gid
= getgid();
3318 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3319 if (setregid(gid
, -1) < 0)
3320 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3323 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3324 if (setreuid(uid
, -1) < 0) {
3325 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3330 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3331 if (keyring
== -1) {
3332 if (errno
== ENOSYS
)
3333 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3334 else if (ERRNO_IS_PRIVILEGE(errno
))
3335 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3336 else if (errno
== EDQUOT
)
3337 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3339 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3344 /* When requested link the user keyring into the session keyring. */
3345 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3347 if (keyctl(KEYCTL_LINK
,
3348 KEY_SPEC_USER_KEYRING
,
3349 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3350 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3355 /* Restore uid/gid back */
3356 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3357 if (setreuid(saved_uid
, -1) < 0) {
3358 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3363 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3364 if (setregid(saved_gid
, -1) < 0)
3365 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3368 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3369 if (!sd_id128_is_null(u
->invocation_id
)) {
3372 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3374 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3376 if (keyctl(KEYCTL_SETPERM
, key
,
3377 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3378 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3379 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3384 /* Revert back uid & gid for the last time, and exit */
3385 /* no extra logging, as only the first already reported error matters */
3386 if (getuid() != saved_uid
)
3387 (void) setreuid(saved_uid
, -1);
3389 if (getgid() != saved_gid
)
3390 (void) setregid(saved_gid
, -1);
3395 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3401 array
[(*n
)++] = pair
[0];
3403 array
[(*n
)++] = pair
[1];
3406 static int close_remaining_fds(
3407 const ExecParameters
*params
,
3408 const ExecRuntime
*runtime
,
3409 const DynamicCreds
*dcreds
,
3413 const int *fds
, size_t n_fds
) {
3415 size_t n_dont_close
= 0;
3416 int dont_close
[n_fds
+ 12];
3420 if (params
->stdin_fd
>= 0)
3421 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3422 if (params
->stdout_fd
>= 0)
3423 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3424 if (params
->stderr_fd
>= 0)
3425 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3428 dont_close
[n_dont_close
++] = socket_fd
;
3430 dont_close
[n_dont_close
++] = exec_fd
;
3432 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3433 n_dont_close
+= n_fds
;
3437 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3441 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3443 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3446 if (user_lookup_fd
>= 0)
3447 dont_close
[n_dont_close
++] = user_lookup_fd
;
3449 return close_all_fds(dont_close
, n_dont_close
);
3452 static int send_user_lookup(
3460 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3461 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3464 if (user_lookup_fd
< 0)
3467 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3470 if (writev(user_lookup_fd
,
3472 IOVEC_INIT(&uid
, sizeof(uid
)),
3473 IOVEC_INIT(&gid
, sizeof(gid
)),
3474 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3480 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3487 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3492 if (!c
->working_directory_home
)
3495 r
= get_home_dir(buf
);
3503 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3504 _cleanup_strv_free_
char ** list
= NULL
;
3511 assert(c
->dynamic_user
);
3513 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3514 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3517 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3520 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3526 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3529 if (exec_directory_is_private(c
, t
))
3530 e
= path_join(p
->prefix
[t
], "private", *i
);
3532 e
= path_join(p
->prefix
[t
], *i
);
3536 r
= strv_consume(&list
, e
);
3542 *ret
= TAKE_PTR(list
);
3547 static char *exec_command_line(char **argv
);
3549 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3550 bool using_subcgroup
;
3556 if (!params
->cgroup_path
)
3559 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3560 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3561 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3562 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3563 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3564 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3565 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3566 * flag, which is only passed for the former statements, not for the latter. */
3568 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3569 if (using_subcgroup
)
3570 p
= path_join(params
->cgroup_path
, ".control");
3572 p
= strdup(params
->cgroup_path
);
3577 return using_subcgroup
;
3580 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3581 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3587 if (!c
->numa_policy
.nodes
.set
) {
3588 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3592 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3598 return cpu_set_add_all(ret
, &s
);
3601 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3604 return c
->cpu_affinity_from_numa
;
3607 static int exec_child(
3609 const ExecCommand
*command
,
3610 const ExecContext
*context
,
3611 const ExecParameters
*params
,
3612 ExecRuntime
*runtime
,
3613 DynamicCreds
*dcreds
,
3615 const int named_iofds
[static 3],
3617 size_t n_socket_fds
,
3618 size_t n_storage_fds
,
3623 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3624 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
3625 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3626 const char *username
= NULL
, *groupname
= NULL
;
3627 _cleanup_free_
char *home_buffer
= NULL
;
3628 const char *home
= NULL
, *shell
= NULL
;
3629 char **final_argv
= NULL
;
3630 dev_t journal_stream_dev
= 0;
3631 ino_t journal_stream_ino
= 0;
3632 bool userns_set_up
= false;
3633 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3634 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3635 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3636 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3638 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3639 bool use_selinux
= false;
3642 bool use_smack
= false;
3645 bool use_apparmor
= false;
3647 uid_t saved_uid
= getuid();
3648 gid_t saved_gid
= getgid();
3649 uid_t uid
= UID_INVALID
;
3650 gid_t gid
= GID_INVALID
;
3653 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3654 int ngids_after_pam
= 0;
3660 assert(exit_status
);
3662 rename_process_from_path(command
->path
);
3664 /* We reset exactly these signals, since they are the
3665 * only ones we set to SIG_IGN in the main daemon. All
3666 * others we leave untouched because we set them to
3667 * SIG_DFL or a valid handler initially, both of which
3668 * will be demoted to SIG_DFL. */
3669 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3670 SIGNALS_IGNORE
, -1);
3672 if (context
->ignore_sigpipe
)
3673 (void) ignore_signals(SIGPIPE
, -1);
3675 r
= reset_signal_mask();
3677 *exit_status
= EXIT_SIGNAL_MASK
;
3678 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3681 if (params
->idle_pipe
)
3682 do_idle_pipe_dance(params
->idle_pipe
);
3684 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3685 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3686 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3687 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3690 log_set_open_when_needed(true);
3692 /* In case anything used libc syslog(), close this here, too */
3695 n_fds
= n_socket_fds
+ n_storage_fds
;
3696 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
3698 *exit_status
= EXIT_FDS
;
3699 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3702 if (!context
->same_pgrp
&&
3704 *exit_status
= EXIT_SETSID
;
3705 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3708 exec_context_tty_reset(context
, params
);
3710 if (unit_shall_confirm_spawn(unit
)) {
3711 const char *vc
= params
->confirm_spawn
;
3712 _cleanup_free_
char *cmdline
= NULL
;
3714 cmdline
= exec_command_line(command
->argv
);
3716 *exit_status
= EXIT_MEMORY
;
3720 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3721 if (r
!= CONFIRM_EXECUTE
) {
3722 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3723 *exit_status
= EXIT_SUCCESS
;
3726 *exit_status
= EXIT_CONFIRM
;
3727 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
3728 "Execution cancelled by the user");
3732 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3733 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3734 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3735 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3736 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3737 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3738 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3739 *exit_status
= EXIT_MEMORY
;
3740 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3743 if (context
->dynamic_user
&& dcreds
) {
3744 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3746 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3747 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
3748 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3749 *exit_status
= EXIT_USER
;
3750 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3753 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3755 *exit_status
= EXIT_MEMORY
;
3759 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3761 *exit_status
= EXIT_USER
;
3763 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3766 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3769 if (!uid_is_valid(uid
)) {
3770 *exit_status
= EXIT_USER
;
3771 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3775 if (!gid_is_valid(gid
)) {
3776 *exit_status
= EXIT_USER
;
3777 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3782 username
= dcreds
->user
->name
;
3785 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3787 *exit_status
= EXIT_USER
;
3788 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3791 r
= get_fixed_group(context
, &groupname
, &gid
);
3793 *exit_status
= EXIT_GROUP
;
3794 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3798 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3799 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3800 &supplementary_gids
, &ngids
);
3802 *exit_status
= EXIT_GROUP
;
3803 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3806 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3808 *exit_status
= EXIT_USER
;
3809 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3812 user_lookup_fd
= safe_close(user_lookup_fd
);
3814 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3816 *exit_status
= EXIT_CHDIR
;
3817 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3820 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3821 * must sure to drop O_NONBLOCK */
3823 (void) fd_nonblock(socket_fd
, false);
3825 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3826 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3827 if (params
->cgroup_path
) {
3828 _cleanup_free_
char *p
= NULL
;
3830 r
= exec_parameters_get_cgroup_path(params
, &p
);
3832 *exit_status
= EXIT_CGROUP
;
3833 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3836 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3838 *exit_status
= EXIT_CGROUP
;
3839 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3843 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3844 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3846 *exit_status
= EXIT_NETWORK
;
3847 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3851 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3853 *exit_status
= EXIT_STDIN
;
3854 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3857 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3859 *exit_status
= EXIT_STDOUT
;
3860 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3863 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3865 *exit_status
= EXIT_STDERR
;
3866 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3869 if (context
->oom_score_adjust_set
) {
3870 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3871 * prohibit write access to this file, and we shouldn't trip up over that. */
3872 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3873 if (ERRNO_IS_PRIVILEGE(r
))
3874 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3876 *exit_status
= EXIT_OOM_ADJUST
;
3877 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3881 if (context
->coredump_filter_set
) {
3882 r
= set_coredump_filter(context
->coredump_filter
);
3883 if (ERRNO_IS_PRIVILEGE(r
))
3884 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
3886 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
3889 if (context
->nice_set
) {
3890 r
= setpriority_closest(context
->nice
);
3892 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3895 if (context
->cpu_sched_set
) {
3896 struct sched_param param
= {
3897 .sched_priority
= context
->cpu_sched_priority
,
3900 r
= sched_setscheduler(0,
3901 context
->cpu_sched_policy
|
3902 (context
->cpu_sched_reset_on_fork
?
3903 SCHED_RESET_ON_FORK
: 0),
3906 *exit_status
= EXIT_SETSCHEDULER
;
3907 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3911 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
3912 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
3913 const CPUSet
*cpu_set
;
3915 if (context
->cpu_affinity_from_numa
) {
3916 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
3918 *exit_status
= EXIT_CPUAFFINITY
;
3919 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
3922 cpu_set
= &converted_cpu_set
;
3924 cpu_set
= &context
->cpu_set
;
3926 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
3927 *exit_status
= EXIT_CPUAFFINITY
;
3928 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3932 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
3933 r
= apply_numa_policy(&context
->numa_policy
);
3934 if (r
== -EOPNOTSUPP
)
3935 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
3937 *exit_status
= EXIT_NUMA_POLICY
;
3938 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
3942 if (context
->ioprio_set
)
3943 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3944 *exit_status
= EXIT_IOPRIO
;
3945 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3948 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3949 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3950 *exit_status
= EXIT_TIMERSLACK
;
3951 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3954 if (context
->personality
!= PERSONALITY_INVALID
) {
3955 r
= safe_personality(context
->personality
);
3957 *exit_status
= EXIT_PERSONALITY
;
3958 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3962 if (context
->utmp_id
)
3963 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3965 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3966 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3970 if (uid_is_valid(uid
)) {
3971 r
= chown_terminal(STDIN_FILENO
, uid
);
3973 *exit_status
= EXIT_STDIN
;
3974 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3978 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3979 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3980 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3981 * touch a single hierarchy too. */
3982 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3983 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3985 *exit_status
= EXIT_CGROUP
;
3986 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3990 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3991 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3993 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3996 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
3997 r
= setup_credentials(context
, params
, unit
->id
, uid
);
3999 *exit_status
= EXIT_CREDENTIALS
;
4000 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4004 r
= build_environment(
4016 *exit_status
= EXIT_MEMORY
;
4020 r
= build_pass_environment(context
, &pass_env
);
4022 *exit_status
= EXIT_MEMORY
;
4026 accum_env
= strv_env_merge(5,
4027 params
->environment
,
4030 context
->environment
,
4033 *exit_status
= EXIT_MEMORY
;
4036 accum_env
= strv_env_clean(accum_env
);
4038 (void) umask(context
->umask
);
4040 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4042 *exit_status
= EXIT_KEYRING
;
4043 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4046 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
4047 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4049 /* 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 */
4050 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4052 /* 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 */
4053 if (needs_ambient_hack
)
4054 needs_setuid
= false;
4056 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4058 if (needs_sandboxing
) {
4059 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
4060 * present. The actual MAC context application will happen later, as late as possible, to avoid
4061 * impacting our own code paths. */
4064 use_selinux
= mac_selinux_use();
4067 use_smack
= mac_smack_use();
4070 use_apparmor
= mac_apparmor_use();
4074 if (needs_sandboxing
) {
4077 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4078 * is set here. (See below.) */
4080 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4082 *exit_status
= EXIT_LIMITS
;
4083 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4087 if (needs_setuid
&& context
->pam_name
&& username
) {
4088 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4089 * wins here. (See above.) */
4091 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4093 *exit_status
= EXIT_PAM
;
4094 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4097 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4098 if (ngids_after_pam
< 0) {
4099 *exit_status
= EXIT_MEMORY
;
4100 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4104 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4105 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4106 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4107 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4109 userns_set_up
= true;
4110 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4112 *exit_status
= EXIT_USER
;
4113 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4117 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4119 if (ns_type_supported(NAMESPACE_NET
)) {
4120 r
= setup_netns(runtime
->netns_storage_socket
);
4122 log_unit_warning_errno(unit
, r
,
4123 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4125 *exit_status
= EXIT_NETWORK
;
4126 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4128 } else if (context
->network_namespace_path
) {
4129 *exit_status
= EXIT_NETWORK
;
4130 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4131 "NetworkNamespacePath= is not supported, refusing.");
4133 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4136 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4137 if (needs_mount_namespace
) {
4138 _cleanup_free_
char *error_path
= NULL
;
4140 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4142 *exit_status
= EXIT_NAMESPACE
;
4143 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4144 error_path
? ": " : "", strempty(error_path
));
4148 if (needs_sandboxing
) {
4149 r
= apply_protect_hostname(unit
, context
, exit_status
);
4154 /* Drop groups as early as possible.
4155 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4156 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4158 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4159 int ngids_to_enforce
= 0;
4161 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4166 if (ngids_to_enforce
< 0) {
4167 *exit_status
= EXIT_MEMORY
;
4168 return log_unit_error_errno(unit
,
4170 "Failed to merge group lists. Group membership might be incorrect: %m");
4173 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4175 *exit_status
= EXIT_GROUP
;
4176 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4180 /* If the user namespace was not set up above, try to do it now.
4181 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4182 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4183 * case of mount namespaces being less privileged when the mount point list is copied from a
4184 * different user namespace). */
4186 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4187 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4189 *exit_status
= EXIT_USER
;
4190 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4194 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4197 _cleanup_free_
char *executable
= NULL
;
4198 r
= find_executable_full(command
->path
, false, &executable
);
4200 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4201 log_struct_errno(LOG_INFO
, r
,
4202 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4204 LOG_UNIT_INVOCATION_ID(unit
),
4205 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4207 "EXECUTABLE=%s", command
->path
);
4211 *exit_status
= EXIT_EXEC
;
4212 return log_struct_errno(LOG_INFO
, r
,
4213 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4215 LOG_UNIT_INVOCATION_ID(unit
),
4216 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4218 "EXECUTABLE=%s", command
->path
);
4222 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
4223 r
= mac_selinux_get_child_mls_label(socket_fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4225 *exit_status
= EXIT_SELINUX_CONTEXT
;
4226 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4231 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4232 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
4233 * however if we have it as we want to keep it open until the final execve(). */
4235 if (params
->exec_fd
>= 0) {
4236 exec_fd
= params
->exec_fd
;
4238 if (exec_fd
< 3 + (int) n_fds
) {
4241 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
4242 * process we are about to execute. */
4244 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
4246 *exit_status
= EXIT_FDS
;
4247 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
4250 CLOSE_AND_REPLACE(exec_fd
, moved_fd
);
4252 /* This fd should be FD_CLOEXEC already, but let's make sure. */
4253 r
= fd_cloexec(exec_fd
, true);
4255 *exit_status
= EXIT_FDS
;
4256 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
4260 fds_with_exec_fd
= newa(int, n_fds
+ 1);
4261 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
4262 fds_with_exec_fd
[n_fds
] = exec_fd
;
4263 n_fds_with_exec_fd
= n_fds
+ 1;
4265 fds_with_exec_fd
= fds
;
4266 n_fds_with_exec_fd
= n_fds
;
4269 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
4271 r
= shift_fds(fds
, n_fds
);
4273 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4275 *exit_status
= EXIT_FDS
;
4276 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4279 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4280 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4281 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4284 secure_bits
= context
->secure_bits
;
4286 if (needs_sandboxing
) {
4289 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4290 * requested. (Note this is placed after the general resource limit initialization, see
4291 * above, in order to take precedence.) */
4292 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4293 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4294 *exit_status
= EXIT_LIMITS
;
4295 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4300 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4301 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4303 r
= setup_smack(context
, executable
);
4305 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4306 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4311 bset
= context
->capability_bounding_set
;
4312 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4313 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4314 * instead of us doing that */
4315 if (needs_ambient_hack
)
4316 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4317 (UINT64_C(1) << CAP_SETUID
) |
4318 (UINT64_C(1) << CAP_SETGID
);
4320 if (!cap_test_all(bset
)) {
4321 r
= capability_bounding_set_drop(bset
, false);
4323 *exit_status
= EXIT_CAPABILITIES
;
4324 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4328 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4330 * To be able to raise the ambient capabilities after setresuid() they have to be
4331 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4332 * After setresuid() the ambient capabilities can be raised as they are present in
4333 * the permitted and inhertiable set. However it is possible that someone wants to
4334 * set ambient capabilities without changing the user, so we also set the ambient
4335 * capabilities here.
4336 * The requested ambient capabilities are raised in the inheritable set if the
4337 * second argument is true. */
4338 if (!needs_ambient_hack
) {
4339 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4341 *exit_status
= EXIT_CAPABILITIES
;
4342 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4347 /* chroot to root directory first, before we lose the ability to chroot */
4348 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4350 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4353 if (uid_is_valid(uid
)) {
4354 r
= enforce_user(context
, uid
);
4356 *exit_status
= EXIT_USER
;
4357 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4360 if (!needs_ambient_hack
&&
4361 context
->capability_ambient_set
!= 0) {
4363 /* Raise the ambient capabilities after user change. */
4364 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4366 *exit_status
= EXIT_CAPABILITIES
;
4367 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4373 /* Apply working directory here, because the working directory might be on NFS and only the user running
4374 * this service might have the correct privilege to change to the working directory */
4375 r
= apply_working_directory(context
, params
, home
, exit_status
);
4377 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4379 if (needs_sandboxing
) {
4380 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4381 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4382 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4383 * are restricted. */
4387 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4390 r
= setexeccon(exec_context
);
4392 *exit_status
= EXIT_SELINUX_CONTEXT
;
4393 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4400 if (use_apparmor
&& context
->apparmor_profile
) {
4401 r
= aa_change_onexec(context
->apparmor_profile
);
4402 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4403 *exit_status
= EXIT_APPARMOR_PROFILE
;
4404 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4409 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4410 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4412 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4413 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4414 * effective set here.
4415 * The effective set is overwritten during execve with the following values:
4416 * - ambient set (for non-root processes)
4417 * - (inheritable | bounding) set for root processes)
4419 * Hence there is no security impact to raise it in the effective set before execve
4421 r
= capability_gain_cap_setpcap(NULL
);
4423 *exit_status
= EXIT_CAPABILITIES
;
4424 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
4426 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
4427 *exit_status
= EXIT_SECUREBITS
;
4428 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
4432 if (context_has_no_new_privileges(context
))
4433 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
4434 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
4435 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
4439 r
= apply_address_families(unit
, context
);
4441 *exit_status
= EXIT_ADDRESS_FAMILIES
;
4442 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
4445 r
= apply_memory_deny_write_execute(unit
, context
);
4447 *exit_status
= EXIT_SECCOMP
;
4448 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
4451 r
= apply_restrict_realtime(unit
, context
);
4453 *exit_status
= EXIT_SECCOMP
;
4454 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
4457 r
= apply_restrict_suid_sgid(unit
, context
);
4459 *exit_status
= EXIT_SECCOMP
;
4460 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
4463 r
= apply_restrict_namespaces(unit
, context
);
4465 *exit_status
= EXIT_SECCOMP
;
4466 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
4469 r
= apply_protect_sysctl(unit
, context
);
4471 *exit_status
= EXIT_SECCOMP
;
4472 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
4475 r
= apply_protect_kernel_modules(unit
, context
);
4477 *exit_status
= EXIT_SECCOMP
;
4478 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
4481 r
= apply_protect_kernel_logs(unit
, context
);
4483 *exit_status
= EXIT_SECCOMP
;
4484 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
4487 r
= apply_protect_clock(unit
, context
);
4489 *exit_status
= EXIT_SECCOMP
;
4490 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
4493 r
= apply_private_devices(unit
, context
);
4495 *exit_status
= EXIT_SECCOMP
;
4496 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
4499 r
= apply_syscall_archs(unit
, context
);
4501 *exit_status
= EXIT_SECCOMP
;
4502 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
4505 r
= apply_lock_personality(unit
, context
);
4507 *exit_status
= EXIT_SECCOMP
;
4508 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
4511 r
= apply_syscall_log(unit
, context
);
4513 *exit_status
= EXIT_SECCOMP
;
4514 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
4517 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
4518 * by the filter as little as possible. */
4519 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
4521 *exit_status
= EXIT_SECCOMP
;
4522 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
4527 if (!strv_isempty(context
->unset_environment
)) {
4530 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
4532 *exit_status
= EXIT_MEMORY
;
4536 strv_free_and_replace(accum_env
, ee
);
4539 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
4540 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
4541 if (!replaced_argv
) {
4542 *exit_status
= EXIT_MEMORY
;
4545 final_argv
= replaced_argv
;
4547 final_argv
= command
->argv
;
4549 if (DEBUG_LOGGING
) {
4550 _cleanup_free_
char *line
;
4552 line
= exec_command_line(final_argv
);
4554 log_struct(LOG_DEBUG
,
4555 "EXECUTABLE=%s", executable
,
4556 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
4558 LOG_UNIT_INVOCATION_ID(unit
));
4564 /* We have finished with all our initializations. Let's now let the manager know that. From this point
4565 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
4567 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4568 *exit_status
= EXIT_EXEC
;
4569 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
4573 execve(executable
, final_argv
, accum_env
);
4579 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4580 * that POLLHUP on it no longer means execve() succeeded. */
4582 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4583 *exit_status
= EXIT_EXEC
;
4584 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4588 *exit_status
= EXIT_EXEC
;
4589 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
4592 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4593 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4595 int exec_spawn(Unit
*unit
,
4596 ExecCommand
*command
,
4597 const ExecContext
*context
,
4598 const ExecParameters
*params
,
4599 ExecRuntime
*runtime
,
4600 DynamicCreds
*dcreds
,
4603 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4604 _cleanup_free_
char *subcgroup_path
= NULL
;
4605 _cleanup_strv_free_
char **files_env
= NULL
;
4606 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4607 _cleanup_free_
char *line
= NULL
;
4615 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4617 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4618 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4619 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4621 if (params
->n_socket_fds
> 1) {
4622 log_unit_error(unit
, "Got more than one socket.");
4626 if (params
->n_socket_fds
== 0) {
4627 log_unit_error(unit
, "Got no socket.");
4631 socket_fd
= params
->fds
[0];
4635 n_socket_fds
= params
->n_socket_fds
;
4636 n_storage_fds
= params
->n_storage_fds
;
4639 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4641 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4643 r
= exec_context_load_environment(unit
, context
, &files_env
);
4645 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4647 line
= exec_command_line(command
->argv
);
4651 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
4652 and, until the next SELinux policy changes, we save further reloads in future children. */
4653 mac_selinux_maybe_reload();
4655 log_struct(LOG_DEBUG
,
4656 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
4657 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
4658 the mount namespace in the child, but we want to log
4659 from the parent, so we need to use the (possibly
4660 inaccurate) path here. */
4662 LOG_UNIT_INVOCATION_ID(unit
));
4664 if (params
->cgroup_path
) {
4665 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4667 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4668 if (r
> 0) { /* We are using a child cgroup */
4669 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4671 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4677 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4680 int exit_status
= EXIT_SUCCESS
;
4682 r
= exec_child(unit
,
4694 unit
->manager
->user_lookup_fds
[1],
4698 const char *status
=
4699 exit_status_to_string(exit_status
,
4700 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4702 log_struct_errno(LOG_ERR
, r
,
4703 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4705 LOG_UNIT_INVOCATION_ID(unit
),
4706 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4707 status
, command
->path
),
4708 "EXECUTABLE=%s", command
->path
);
4714 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4716 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4717 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4718 * process will be killed too). */
4720 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4722 exec_status_start(&command
->exec_status
, pid
);
4728 void exec_context_init(ExecContext
*c
) {
4732 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4733 c
->cpu_sched_policy
= SCHED_OTHER
;
4734 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4735 c
->syslog_level_prefix
= true;
4736 c
->ignore_sigpipe
= true;
4737 c
->timer_slack_nsec
= NSEC_INFINITY
;
4738 c
->personality
= PERSONALITY_INVALID
;
4739 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4740 c
->directories
[t
].mode
= 0755;
4741 c
->timeout_clean_usec
= USEC_INFINITY
;
4742 c
->capability_bounding_set
= CAP_ALL
;
4743 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4744 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4745 c
->log_level_max
= -1;
4747 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
4749 numa_policy_reset(&c
->numa_policy
);
4752 void exec_context_done(ExecContext
*c
) {
4755 c
->environment
= strv_free(c
->environment
);
4756 c
->environment_files
= strv_free(c
->environment_files
);
4757 c
->pass_environment
= strv_free(c
->pass_environment
);
4758 c
->unset_environment
= strv_free(c
->unset_environment
);
4760 rlimit_free_all(c
->rlimit
);
4762 for (size_t l
= 0; l
< 3; l
++) {
4763 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4764 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4767 c
->working_directory
= mfree(c
->working_directory
);
4768 c
->root_directory
= mfree(c
->root_directory
);
4769 c
->root_image
= mfree(c
->root_image
);
4770 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
4771 c
->root_hash
= mfree(c
->root_hash
);
4772 c
->root_hash_size
= 0;
4773 c
->root_hash_path
= mfree(c
->root_hash_path
);
4774 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
4775 c
->root_hash_sig_size
= 0;
4776 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
4777 c
->root_verity
= mfree(c
->root_verity
);
4778 c
->tty_path
= mfree(c
->tty_path
);
4779 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4780 c
->user
= mfree(c
->user
);
4781 c
->group
= mfree(c
->group
);
4783 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4785 c
->pam_name
= mfree(c
->pam_name
);
4787 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4788 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4789 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4791 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4792 c
->bind_mounts
= NULL
;
4793 c
->n_bind_mounts
= 0;
4794 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4795 c
->temporary_filesystems
= NULL
;
4796 c
->n_temporary_filesystems
= 0;
4797 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
4799 cpu_set_reset(&c
->cpu_set
);
4800 numa_policy_reset(&c
->numa_policy
);
4802 c
->utmp_id
= mfree(c
->utmp_id
);
4803 c
->selinux_context
= mfree(c
->selinux_context
);
4804 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4805 c
->smack_process_label
= mfree(c
->smack_process_label
);
4807 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4808 c
->syscall_archs
= set_free(c
->syscall_archs
);
4809 c
->address_families
= set_free(c
->address_families
);
4811 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4812 c
->directories
[t
].paths
= strv_free(c
->directories
[t
].paths
);
4814 c
->log_level_max
= -1;
4816 exec_context_free_log_extra_fields(c
);
4818 c
->log_ratelimit_interval_usec
= 0;
4819 c
->log_ratelimit_burst
= 0;
4821 c
->stdin_data
= mfree(c
->stdin_data
);
4822 c
->stdin_data_size
= 0;
4824 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4826 c
->log_namespace
= mfree(c
->log_namespace
);
4828 c
->load_credentials
= strv_free(c
->load_credentials
);
4829 c
->set_credentials
= hashmap_free(c
->set_credentials
);
4832 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4837 if (!runtime_prefix
)
4840 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4841 _cleanup_free_
char *p
;
4843 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4844 p
= path_join(runtime_prefix
, "private", *i
);
4846 p
= path_join(runtime_prefix
, *i
);
4850 /* We execute this synchronously, since we need to be sure this is gone when we start the
4852 (void) rm_rf(p
, REMOVE_ROOT
);
4858 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
4859 _cleanup_free_
char *p
= NULL
;
4863 if (!runtime_prefix
|| !unit
)
4866 p
= path_join(runtime_prefix
, "credentials", unit
);
4870 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
4871 * unmount it, and afterwards remove the mount point */
4872 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
4873 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
4878 static void exec_command_done(ExecCommand
*c
) {
4881 c
->path
= mfree(c
->path
);
4882 c
->argv
= strv_free(c
->argv
);
4885 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4888 for (i
= 0; i
< n
; i
++)
4889 exec_command_done(c
+i
);
4892 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4896 LIST_REMOVE(command
, c
, i
);
4897 exec_command_done(i
);
4904 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4905 for (size_t i
= 0; i
< n
; i
++)
4906 c
[i
] = exec_command_free_list(c
[i
]);
4909 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4910 for (size_t i
= 0; i
< n
; i
++)
4911 exec_status_reset(&c
[i
].exec_status
);
4914 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
4915 for (size_t i
= 0; i
< n
; i
++) {
4918 LIST_FOREACH(command
, z
, c
[i
])
4919 exec_status_reset(&z
->exec_status
);
4923 typedef struct InvalidEnvInfo
{
4928 static void invalid_env(const char *p
, void *userdata
) {
4929 InvalidEnvInfo
*info
= userdata
;
4931 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4934 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4940 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4943 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4946 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4949 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4952 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4955 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4962 static int exec_context_named_iofds(
4963 const ExecContext
*c
,
4964 const ExecParameters
*p
,
4965 int named_iofds
[static 3]) {
4968 const char* stdio_fdname
[3];
4973 assert(named_iofds
);
4975 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4976 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4977 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4979 for (size_t i
= 0; i
< 3; i
++)
4980 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4982 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4984 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
4985 if (named_iofds
[STDIN_FILENO
] < 0 &&
4986 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4987 stdio_fdname
[STDIN_FILENO
] &&
4988 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4990 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4993 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4994 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4995 stdio_fdname
[STDOUT_FILENO
] &&
4996 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4998 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5001 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5002 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5003 stdio_fdname
[STDERR_FILENO
] &&
5004 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5006 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5010 return targets
== 0 ? 0 : -ENOENT
;
5013 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
5014 char **i
, **r
= NULL
;
5019 STRV_FOREACH(i
, c
->environment_files
) {
5022 bool ignore
= false;
5024 _cleanup_globfree_ glob_t pglob
= {};
5033 if (!path_is_absolute(fn
)) {
5041 /* Filename supports globbing, take all matching files */
5042 k
= safe_glob(fn
, 0, &pglob
);
5051 /* When we don't match anything, -ENOENT should be returned */
5052 assert(pglob
.gl_pathc
> 0);
5054 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5055 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5063 /* Log invalid environment variables with filename */
5065 InvalidEnvInfo info
= {
5067 .path
= pglob
.gl_pathv
[n
]
5070 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5078 m
= strv_env_merge(2, r
, p
);
5094 static bool tty_may_match_dev_console(const char *tty
) {
5095 _cleanup_free_
char *resolved
= NULL
;
5100 tty
= skip_dev_prefix(tty
);
5102 /* trivial identity? */
5103 if (streq(tty
, "console"))
5106 if (resolve_dev_console(&resolved
) < 0)
5107 return true; /* if we could not resolve, assume it may */
5109 /* "tty0" means the active VC, so it may be the same sometimes */
5110 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5113 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5116 return ec
->tty_reset
||
5118 ec
->tty_vt_disallocate
||
5119 is_terminal_input(ec
->std_input
) ||
5120 is_terminal_output(ec
->std_output
) ||
5121 is_terminal_output(ec
->std_error
);
5124 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5126 return exec_context_may_touch_tty(ec
) &&
5127 tty_may_match_dev_console(exec_context_tty_path(ec
));
5130 static void strv_fprintf(FILE *f
, char **l
) {
5136 fprintf(f
, " %s", *g
);
5139 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5140 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
5146 prefix
= strempty(prefix
);
5150 "%sWorkingDirectory: %s\n"
5151 "%sRootDirectory: %s\n"
5152 "%sNonBlocking: %s\n"
5153 "%sPrivateTmp: %s\n"
5154 "%sPrivateDevices: %s\n"
5155 "%sProtectKernelTunables: %s\n"
5156 "%sProtectKernelModules: %s\n"
5157 "%sProtectKernelLogs: %s\n"
5158 "%sProtectClock: %s\n"
5159 "%sProtectControlGroups: %s\n"
5160 "%sPrivateNetwork: %s\n"
5161 "%sPrivateUsers: %s\n"
5162 "%sProtectHome: %s\n"
5163 "%sProtectSystem: %s\n"
5164 "%sMountAPIVFS: %s\n"
5165 "%sIgnoreSIGPIPE: %s\n"
5166 "%sMemoryDenyWriteExecute: %s\n"
5167 "%sRestrictRealtime: %s\n"
5168 "%sRestrictSUIDSGID: %s\n"
5169 "%sKeyringMode: %s\n"
5170 "%sProtectHostname: %s\n"
5171 "%sProtectProc: %s\n"
5172 "%sProcSubset: %s\n",
5174 prefix
, c
->working_directory
? c
->working_directory
: "/",
5175 prefix
, c
->root_directory
? c
->root_directory
: "/",
5176 prefix
, yes_no(c
->non_blocking
),
5177 prefix
, yes_no(c
->private_tmp
),
5178 prefix
, yes_no(c
->private_devices
),
5179 prefix
, yes_no(c
->protect_kernel_tunables
),
5180 prefix
, yes_no(c
->protect_kernel_modules
),
5181 prefix
, yes_no(c
->protect_kernel_logs
),
5182 prefix
, yes_no(c
->protect_clock
),
5183 prefix
, yes_no(c
->protect_control_groups
),
5184 prefix
, yes_no(c
->private_network
),
5185 prefix
, yes_no(c
->private_users
),
5186 prefix
, protect_home_to_string(c
->protect_home
),
5187 prefix
, protect_system_to_string(c
->protect_system
),
5188 prefix
, yes_no(c
->mount_apivfs
),
5189 prefix
, yes_no(c
->ignore_sigpipe
),
5190 prefix
, yes_no(c
->memory_deny_write_execute
),
5191 prefix
, yes_no(c
->restrict_realtime
),
5192 prefix
, yes_no(c
->restrict_suid_sgid
),
5193 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5194 prefix
, yes_no(c
->protect_hostname
),
5195 prefix
, protect_proc_to_string(c
->protect_proc
),
5196 prefix
, proc_subset_to_string(c
->proc_subset
));
5199 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5201 if (c
->root_image_options
) {
5204 fprintf(f
, "%sRootImageOptions:", prefix
);
5205 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5206 if (!isempty(o
->options
))
5207 fprintf(f
, " %s:%s",
5208 partition_designator_to_string(o
->partition_designator
),
5214 _cleanup_free_
char *encoded
= NULL
;
5215 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5217 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5220 if (c
->root_hash_path
)
5221 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5223 if (c
->root_hash_sig
) {
5224 _cleanup_free_
char *encoded
= NULL
;
5226 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5228 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5231 if (c
->root_hash_sig_path
)
5232 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5235 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5237 STRV_FOREACH(e
, c
->environment
)
5238 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5240 STRV_FOREACH(e
, c
->environment_files
)
5241 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5243 STRV_FOREACH(e
, c
->pass_environment
)
5244 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5246 STRV_FOREACH(e
, c
->unset_environment
)
5247 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5249 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5251 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5252 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5254 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
5255 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
5259 "%sTimeoutCleanSec: %s\n",
5260 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
5267 if (c
->oom_score_adjust_set
)
5269 "%sOOMScoreAdjust: %i\n",
5270 prefix
, c
->oom_score_adjust
);
5272 if (c
->coredump_filter_set
)
5274 "%sCoredumpFilter: 0x%"PRIx64
"\n",
5275 prefix
, c
->coredump_filter
);
5277 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5279 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5280 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5281 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5282 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5285 if (c
->ioprio_set
) {
5286 _cleanup_free_
char *class_str
= NULL
;
5288 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
5290 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5292 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
5295 if (c
->cpu_sched_set
) {
5296 _cleanup_free_
char *policy_str
= NULL
;
5298 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5300 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5303 "%sCPUSchedulingPriority: %i\n"
5304 "%sCPUSchedulingResetOnFork: %s\n",
5305 prefix
, c
->cpu_sched_priority
,
5306 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5309 if (c
->cpu_set
.set
) {
5310 _cleanup_free_
char *affinity
= NULL
;
5312 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5313 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5316 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5317 _cleanup_free_
char *nodes
= NULL
;
5319 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5320 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5321 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5324 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5325 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5328 "%sStandardInput: %s\n"
5329 "%sStandardOutput: %s\n"
5330 "%sStandardError: %s\n",
5331 prefix
, exec_input_to_string(c
->std_input
),
5332 prefix
, exec_output_to_string(c
->std_output
),
5333 prefix
, exec_output_to_string(c
->std_error
));
5335 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5336 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5337 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5338 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5339 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5340 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5342 if (c
->std_input
== EXEC_INPUT_FILE
)
5343 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5344 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5345 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5346 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5347 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5348 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5349 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5350 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5351 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5357 "%sTTYVHangup: %s\n"
5358 "%sTTYVTDisallocate: %s\n",
5359 prefix
, c
->tty_path
,
5360 prefix
, yes_no(c
->tty_reset
),
5361 prefix
, yes_no(c
->tty_vhangup
),
5362 prefix
, yes_no(c
->tty_vt_disallocate
));
5364 if (IN_SET(c
->std_output
,
5366 EXEC_OUTPUT_JOURNAL
,
5367 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5368 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5369 IN_SET(c
->std_error
,
5371 EXEC_OUTPUT_JOURNAL
,
5372 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5373 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5375 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5377 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5379 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5381 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5383 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5386 if (c
->log_level_max
>= 0) {
5387 _cleanup_free_
char *t
= NULL
;
5389 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5391 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5394 if (c
->log_ratelimit_interval_usec
> 0) {
5395 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
5398 "%sLogRateLimitIntervalSec: %s\n",
5399 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
5402 if (c
->log_ratelimit_burst
> 0)
5403 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
5405 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
5406 fprintf(f
, "%sLogExtraFields: ", prefix
);
5407 fwrite(c
->log_extra_fields
[j
].iov_base
,
5408 1, c
->log_extra_fields
[j
].iov_len
,
5413 if (c
->log_namespace
)
5414 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
5416 if (c
->secure_bits
) {
5417 _cleanup_free_
char *str
= NULL
;
5419 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
5421 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
5424 if (c
->capability_bounding_set
!= CAP_ALL
) {
5425 _cleanup_free_
char *str
= NULL
;
5427 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
5429 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
5432 if (c
->capability_ambient_set
!= 0) {
5433 _cleanup_free_
char *str
= NULL
;
5435 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
5437 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
5441 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
5443 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
5445 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
5447 if (!strv_isempty(c
->supplementary_groups
)) {
5448 fprintf(f
, "%sSupplementaryGroups:", prefix
);
5449 strv_fprintf(f
, c
->supplementary_groups
);
5454 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
5456 if (!strv_isempty(c
->read_write_paths
)) {
5457 fprintf(f
, "%sReadWritePaths:", prefix
);
5458 strv_fprintf(f
, c
->read_write_paths
);
5462 if (!strv_isempty(c
->read_only_paths
)) {
5463 fprintf(f
, "%sReadOnlyPaths:", prefix
);
5464 strv_fprintf(f
, c
->read_only_paths
);
5468 if (!strv_isempty(c
->inaccessible_paths
)) {
5469 fprintf(f
, "%sInaccessiblePaths:", prefix
);
5470 strv_fprintf(f
, c
->inaccessible_paths
);
5474 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
5475 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
5476 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
5477 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
5478 c
->bind_mounts
[i
].source
,
5479 c
->bind_mounts
[i
].destination
,
5480 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
5482 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
5483 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
5485 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
5487 isempty(t
->options
) ? "" : ":",
5488 strempty(t
->options
));
5493 "%sUtmpIdentifier: %s\n",
5494 prefix
, c
->utmp_id
);
5496 if (c
->selinux_context
)
5498 "%sSELinuxContext: %s%s\n",
5499 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
5501 if (c
->apparmor_profile
)
5503 "%sAppArmorProfile: %s%s\n",
5504 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
5506 if (c
->smack_process_label
)
5508 "%sSmackProcessLabel: %s%s\n",
5509 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
5511 if (c
->personality
!= PERSONALITY_INVALID
)
5513 "%sPersonality: %s\n",
5514 prefix
, strna(personality_to_string(c
->personality
)));
5517 "%sLockPersonality: %s\n",
5518 prefix
, yes_no(c
->lock_personality
));
5520 if (c
->syscall_filter
) {
5527 "%sSystemCallFilter: ",
5530 if (!c
->syscall_allow_list
)
5534 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
5535 _cleanup_free_
char *name
= NULL
;
5536 const char *errno_name
= NULL
;
5537 int num
= PTR_TO_INT(val
);
5544 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
5545 fputs(strna(name
), f
);
5548 errno_name
= seccomp_errno_or_action_to_string(num
);
5550 fprintf(f
, ":%s", errno_name
);
5552 fprintf(f
, ":%d", num
);
5560 if (c
->syscall_archs
) {
5566 "%sSystemCallArchitectures:",
5570 SET_FOREACH(id
, c
->syscall_archs
)
5571 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
5576 if (exec_context_restrict_namespaces_set(c
)) {
5577 _cleanup_free_
char *s
= NULL
;
5579 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
5581 fprintf(f
, "%sRestrictNamespaces: %s\n",
5585 if (c
->network_namespace_path
)
5587 "%sNetworkNamespacePath: %s\n",
5588 prefix
, c
->network_namespace_path
);
5590 if (c
->syscall_errno
> 0) {
5592 const char *errno_name
;
5595 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
5598 errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
5600 fputs(errno_name
, f
);
5602 fprintf(f
, "%d", c
->syscall_errno
);
5607 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
5610 fprintf(f
, "%sMountImages: %s%s:%s%s", prefix
,
5611 c
->mount_images
[i
].ignore_enoent
? "-": "",
5612 c
->mount_images
[i
].source
,
5613 c
->mount_images
[i
].destination
,
5614 LIST_IS_EMPTY(c
->mount_images
[i
].mount_options
) ? "": ":");
5615 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
5617 partition_designator_to_string(o
->partition_designator
),
5623 bool exec_context_maintains_privileges(const ExecContext
*c
) {
5626 /* Returns true if the process forked off would run under
5627 * an unchanged UID or as root. */
5632 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
5638 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
5646 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5648 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5653 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5656 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
5657 free(c
->log_extra_fields
[l
].iov_base
);
5658 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5659 c
->n_log_extra_fields
= 0;
5662 void exec_context_revert_tty(ExecContext
*c
) {
5667 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5668 exec_context_tty_reset(c
, NULL
);
5670 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5671 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5672 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5674 if (exec_context_may_touch_tty(c
)) {
5677 path
= exec_context_tty_path(c
);
5679 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
5680 if (r
< 0 && r
!= -ENOENT
)
5681 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5686 int exec_context_get_clean_directories(
5692 _cleanup_strv_free_
char **l
= NULL
;
5699 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5702 if (!FLAGS_SET(mask
, 1U << t
))
5708 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5711 j
= path_join(prefix
[t
], *i
);
5715 r
= strv_consume(&l
, j
);
5719 /* Also remove private directories unconditionally. */
5720 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5721 j
= path_join(prefix
[t
], "private", *i
);
5725 r
= strv_consume(&l
, j
);
5736 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5737 ExecCleanMask mask
= 0;
5742 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5743 if (!strv_isempty(c
->directories
[t
].paths
))
5750 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5757 dual_timestamp_get(&s
->start_timestamp
);
5760 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5763 if (s
->pid
!= pid
) {
5769 dual_timestamp_get(&s
->exit_timestamp
);
5774 if (context
&& context
->utmp_id
)
5775 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5778 void exec_status_reset(ExecStatus
*s
) {
5781 *s
= (ExecStatus
) {};
5784 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5785 char buf
[FORMAT_TIMESTAMP_MAX
];
5793 prefix
= strempty(prefix
);
5796 "%sPID: "PID_FMT
"\n",
5799 if (dual_timestamp_is_set(&s
->start_timestamp
))
5801 "%sStart Timestamp: %s\n",
5802 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5804 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5806 "%sExit Timestamp: %s\n"
5808 "%sExit Status: %i\n",
5809 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5810 prefix
, sigchld_code_to_string(s
->code
),
5814 static char *exec_command_line(char **argv
) {
5822 STRV_FOREACH(a
, argv
)
5830 STRV_FOREACH(a
, argv
) {
5837 if (strpbrk(*a
, WHITESPACE
)) {
5848 /* FIXME: this doesn't really handle arguments that have
5849 * spaces and ticks in them */
5854 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5855 _cleanup_free_
char *cmd
= NULL
;
5856 const char *prefix2
;
5861 prefix
= strempty(prefix
);
5862 prefix2
= strjoina(prefix
, "\t");
5864 cmd
= exec_command_line(c
->argv
);
5866 "%sCommand Line: %s\n",
5867 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
5869 exec_status_dump(&c
->exec_status
, f
, prefix2
);
5872 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5875 prefix
= strempty(prefix
);
5877 LIST_FOREACH(command
, c
, c
)
5878 exec_command_dump(c
, f
, prefix
);
5881 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
5888 /* It's kind of important, that we keep the order here */
5889 LIST_FIND_TAIL(command
, *l
, end
);
5890 LIST_INSERT_AFTER(command
, *l
, end
, e
);
5895 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
5903 l
= strv_new_ap(path
, ap
);
5915 free_and_replace(c
->path
, p
);
5917 return strv_free_and_replace(c
->argv
, l
);
5920 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
5921 _cleanup_strv_free_
char **l
= NULL
;
5929 l
= strv_new_ap(path
, ap
);
5935 r
= strv_extend_strv(&c
->argv
, l
, false);
5942 static void *remove_tmpdir_thread(void *p
) {
5943 _cleanup_free_
char *path
= p
;
5945 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
5949 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
5956 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
5958 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
5960 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
5961 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
5963 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
5965 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
5970 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
5971 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
5973 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
5975 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
5977 rt
->var_tmp_dir
= NULL
;
5980 rt
->id
= mfree(rt
->id
);
5981 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
5982 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
5983 safe_close_pair(rt
->netns_storage_socket
);
5987 static void exec_runtime_freep(ExecRuntime
**rt
) {
5988 (void) exec_runtime_free(*rt
, false);
5991 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
5992 _cleanup_free_
char *id_copy
= NULL
;
5997 id_copy
= strdup(id
);
6001 n
= new(ExecRuntime
, 1);
6005 *n
= (ExecRuntime
) {
6006 .id
= TAKE_PTR(id_copy
),
6007 .netns_storage_socket
= { -1, -1 },
6014 static int exec_runtime_add(
6019 int netns_storage_socket
[2],
6020 ExecRuntime
**ret
) {
6022 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6028 /* tmp_dir, var_tmp_dir, netns_storage_socket fds are donated on success */
6030 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
6034 r
= exec_runtime_allocate(&rt
, id
);
6038 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
6042 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6043 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6044 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6046 if (netns_storage_socket
) {
6047 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6048 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6055 /* do not remove created ExecRuntime object when the operation succeeds. */
6060 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
6061 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6062 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
6069 /* It is not necessary to create ExecRuntime object. */
6070 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
6073 if (c
->private_tmp
&&
6074 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6075 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6076 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6077 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6082 if (c
->private_network
|| c
->network_namespace_path
) {
6083 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6087 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ret
);
6094 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6102 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6104 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
6110 /* If not found, then create a new object. */
6111 r
= exec_runtime_make(m
, c
, id
, &rt
);
6113 /* When r == 0, it is not necessary to create ExecRuntime object. */
6117 /* increment reference counter. */
6123 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6127 assert(rt
->n_ref
> 0);
6133 return exec_runtime_free(rt
, destroy
);
6136 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6143 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6144 fprintf(f
, "exec-runtime=%s", rt
->id
);
6147 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6149 if (rt
->var_tmp_dir
)
6150 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6152 if (rt
->netns_storage_socket
[0] >= 0) {
6155 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6159 fprintf(f
, " netns-socket-0=%i", copy
);
6162 if (rt
->netns_storage_socket
[1] >= 0) {
6165 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6169 fprintf(f
, " netns-socket-1=%i", copy
);
6178 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6179 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6183 /* This is for the migration from old (v237 or earlier) deserialization text.
6184 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6185 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6186 * so or not from the serialized text, then we always creates a new object owned by this. */
6192 /* Manager manages ExecRuntime objects by the unit id.
6193 * So, we omit the serialized text when the unit does not have id (yet?)... */
6194 if (isempty(u
->id
)) {
6195 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6199 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
6201 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
6205 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6207 r
= exec_runtime_allocate(&rt_create
, u
->id
);
6214 if (streq(key
, "tmp-dir")) {
6217 copy
= strdup(value
);
6221 free_and_replace(rt
->tmp_dir
, copy
);
6223 } else if (streq(key
, "var-tmp-dir")) {
6226 copy
= strdup(value
);
6230 free_and_replace(rt
->var_tmp_dir
, copy
);
6232 } else if (streq(key
, "netns-socket-0")) {
6235 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6236 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6240 safe_close(rt
->netns_storage_socket
[0]);
6241 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6243 } else if (streq(key
, "netns-socket-1")) {
6246 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6247 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6251 safe_close(rt
->netns_storage_socket
[1]);
6252 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6256 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6258 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6260 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6264 rt_create
->manager
= u
->manager
;
6267 TAKE_PTR(rt_create
);
6273 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6274 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6276 int r
, fdpair
[] = {-1, -1};
6277 const char *p
, *v
= value
;
6284 n
= strcspn(v
, " ");
6285 id
= strndupa(v
, n
);
6290 v
= startswith(p
, "tmp-dir=");
6292 n
= strcspn(v
, " ");
6293 tmp_dir
= strndup(v
, n
);
6301 v
= startswith(p
, "var-tmp-dir=");
6303 n
= strcspn(v
, " ");
6304 var_tmp_dir
= strndup(v
, n
);
6312 v
= startswith(p
, "netns-socket-0=");
6316 n
= strcspn(v
, " ");
6317 buf
= strndupa(v
, n
);
6319 r
= safe_atoi(buf
, &fdpair
[0]);
6321 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6322 if (!fdset_contains(fds
, fdpair
[0]))
6323 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6324 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", fdpair
[0]);
6325 fdpair
[0] = fdset_remove(fds
, fdpair
[0]);
6331 v
= startswith(p
, "netns-socket-1=");
6335 n
= strcspn(v
, " ");
6336 buf
= strndupa(v
, n
);
6337 r
= safe_atoi(buf
, &fdpair
[1]);
6339 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6340 if (!fdset_contains(fds
, fdpair
[0]))
6341 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6342 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", fdpair
[1]);
6343 fdpair
[1] = fdset_remove(fds
, fdpair
[1]);
6347 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, fdpair
, NULL
);
6349 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
6353 void exec_runtime_vacuum(Manager
*m
) {
6358 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
6360 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6364 (void) exec_runtime_free(rt
, false);
6368 void exec_params_clear(ExecParameters
*p
) {
6372 p
->environment
= strv_free(p
->environment
);
6373 p
->fd_names
= strv_free(p
->fd_names
);
6374 p
->fds
= mfree(p
->fds
);
6375 p
->exec_fd
= safe_close(p
->exec_fd
);
6378 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
6387 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
6389 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
6390 [EXEC_INPUT_NULL
] = "null",
6391 [EXEC_INPUT_TTY
] = "tty",
6392 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
6393 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
6394 [EXEC_INPUT_SOCKET
] = "socket",
6395 [EXEC_INPUT_NAMED_FD
] = "fd",
6396 [EXEC_INPUT_DATA
] = "data",
6397 [EXEC_INPUT_FILE
] = "file",
6400 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
6402 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
6403 [EXEC_OUTPUT_INHERIT
] = "inherit",
6404 [EXEC_OUTPUT_NULL
] = "null",
6405 [EXEC_OUTPUT_TTY
] = "tty",
6406 [EXEC_OUTPUT_KMSG
] = "kmsg",
6407 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
6408 [EXEC_OUTPUT_JOURNAL
] = "journal",
6409 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
6410 [EXEC_OUTPUT_SOCKET
] = "socket",
6411 [EXEC_OUTPUT_NAMED_FD
] = "fd",
6412 [EXEC_OUTPUT_FILE
] = "file",
6413 [EXEC_OUTPUT_FILE_APPEND
] = "append",
6416 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
6418 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
6419 [EXEC_UTMP_INIT
] = "init",
6420 [EXEC_UTMP_LOGIN
] = "login",
6421 [EXEC_UTMP_USER
] = "user",
6424 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
6426 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
6427 [EXEC_PRESERVE_NO
] = "no",
6428 [EXEC_PRESERVE_YES
] = "yes",
6429 [EXEC_PRESERVE_RESTART
] = "restart",
6432 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
6434 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
6435 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6436 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
6437 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
6438 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
6439 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
6440 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
6443 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
6445 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
6446 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
6447 * directories, specifically .timer units with their timestamp touch file. */
6448 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6449 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
6450 [EXEC_DIRECTORY_STATE
] = "state",
6451 [EXEC_DIRECTORY_CACHE
] = "cache",
6452 [EXEC_DIRECTORY_LOGS
] = "logs",
6453 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
6456 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
6458 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
6459 * the service payload in. */
6460 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6461 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
6462 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
6463 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
6464 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
6465 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
6468 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
6470 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
6471 [EXEC_KEYRING_INHERIT
] = "inherit",
6472 [EXEC_KEYRING_PRIVATE
] = "private",
6473 [EXEC_KEYRING_SHARED
] = "shared",
6476 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);