1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <sys/eventfd.h>
10 #include <sys/personality.h>
11 #include <sys/prctl.h>
13 #include <sys/types.h>
19 #include <security/pam_appl.h>
23 #include <selinux/selinux.h>
31 #include <sys/apparmor.h>
34 #include "sd-messages.h"
38 #include "alloc-util.h"
40 #include "apparmor-util.h"
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 "random-util.h"
77 #include "rlimit-util.h"
80 #include "seccomp-util.h"
82 #include "securebits-util.h"
83 #include "selinux-util.h"
84 #include "signal-util.h"
85 #include "smack-util.h"
86 #include "socket-util.h"
88 #include "stat-util.h"
89 #include "string-table.h"
90 #include "string-util.h"
92 #include "syslog-util.h"
93 #include "terminal-util.h"
94 #include "tmpfile-util.h"
95 #include "umask-util.h"
97 #include "user-util.h"
98 #include "utmp-wtmp.h"
100 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
101 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
103 #define SNDBUF_SIZE (8*1024*1024)
105 static int shift_fds(int fds
[], size_t n_fds
) {
109 /* Modifies the fds array! (sorts it) */
113 for (int start
= 0;;) {
114 int restart_from
= -1;
116 for (int i
= start
; i
< (int) n_fds
; i
++) {
119 /* Already at right index? */
123 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
130 /* Hmm, the fd we wanted isn't free? Then
131 * let's remember that and try again from here */
132 if (nfd
!= i
+3 && restart_from
< 0)
136 if (restart_from
< 0)
139 start
= restart_from
;
145 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
149 n_fds
= n_socket_fds
+ n_storage_fds
;
155 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
156 * O_NONBLOCK only applies to socket activation though. */
158 for (size_t i
= 0; i
< n_fds
; i
++) {
160 if (i
< n_socket_fds
) {
161 r
= fd_nonblock(fds
[i
], nonblock
);
166 /* We unconditionally drop FD_CLOEXEC from the fds,
167 * since after all we want to pass these fds to our
170 r
= fd_cloexec(fds
[i
], false);
178 static const char *exec_context_tty_path(const ExecContext
*context
) {
181 if (context
->stdio_as_fds
)
184 if (context
->tty_path
)
185 return context
->tty_path
;
187 return "/dev/console";
190 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
195 path
= exec_context_tty_path(context
);
197 if (context
->tty_vhangup
) {
198 if (p
&& p
->stdin_fd
>= 0)
199 (void) terminal_vhangup_fd(p
->stdin_fd
);
201 (void) terminal_vhangup(path
);
204 if (context
->tty_reset
) {
205 if (p
&& p
->stdin_fd
>= 0)
206 (void) reset_terminal_fd(p
->stdin_fd
, true);
208 (void) reset_terminal(path
);
211 if (context
->tty_vt_disallocate
&& path
)
212 (void) vt_disallocate(path
);
215 static bool is_terminal_input(ExecInput i
) {
218 EXEC_INPUT_TTY_FORCE
,
219 EXEC_INPUT_TTY_FAIL
);
222 static bool is_terminal_output(ExecOutput o
) {
225 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
226 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
229 static bool is_kmsg_output(ExecOutput o
) {
232 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
235 static bool exec_context_needs_term(const ExecContext
*c
) {
238 /* Return true if the execution context suggests we should set $TERM to something useful. */
240 if (is_terminal_input(c
->std_input
))
243 if (is_terminal_output(c
->std_output
))
246 if (is_terminal_output(c
->std_error
))
249 return !!c
->tty_path
;
252 static int open_null_as(int flags
, int nfd
) {
257 fd
= open("/dev/null", flags
|O_NOCTTY
);
261 return move_fd(fd
, nfd
, false);
264 static int connect_journal_socket(
266 const char *log_namespace
,
270 union sockaddr_union sa
;
272 uid_t olduid
= UID_INVALID
;
273 gid_t oldgid
= GID_INVALID
;
278 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
279 "/run/systemd/journal/stdout";
280 r
= sockaddr_un_set_path(&sa
.un
, j
);
285 if (gid_is_valid(gid
)) {
288 if (setegid(gid
) < 0)
292 if (uid_is_valid(uid
)) {
295 if (seteuid(uid
) < 0) {
301 r
= connect(fd
, &sa
.sa
, sa_len
) < 0 ? -errno
: 0;
303 /* If we fail to restore the uid or gid, things will likely
304 fail later on. This should only happen if an LSM interferes. */
306 if (uid_is_valid(uid
))
307 (void) seteuid(olduid
);
310 if (gid_is_valid(gid
))
311 (void) setegid(oldgid
);
316 static int connect_logger_as(
318 const ExecContext
*context
,
319 const ExecParameters
*params
,
326 _cleanup_close_
int fd
= -1;
331 assert(output
< _EXEC_OUTPUT_MAX
);
335 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
339 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
343 if (shutdown(fd
, SHUT_RD
) < 0)
346 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
356 context
->syslog_identifier
?: ident
,
357 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
358 context
->syslog_priority
,
359 !!context
->syslog_level_prefix
,
361 is_kmsg_output(output
),
362 is_terminal_output(output
)) < 0)
365 return move_fd(TAKE_FD(fd
), nfd
, false);
368 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
374 fd
= open_terminal(path
, flags
| O_NOCTTY
);
378 return move_fd(fd
, nfd
, false);
381 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
382 union sockaddr_union sa
;
384 _cleanup_close_
int fd
= -1;
389 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
392 fd
= open(path
, flags
|O_NOCTTY
, mode
);
396 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
399 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
401 r
= sockaddr_un_set_path(&sa
.un
, path
);
403 return r
== -EINVAL
? -ENXIO
: r
;
406 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
410 if (connect(fd
, &sa
.sa
, sa_len
) < 0)
411 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
412 * indication that this wasn't an AF_UNIX socket after all */
414 if ((flags
& O_ACCMODE
) == O_RDONLY
)
415 r
= shutdown(fd
, SHUT_WR
);
416 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
417 r
= shutdown(fd
, SHUT_RD
);
426 static int fixup_input(
427 const ExecContext
*context
,
429 bool apply_tty_stdin
) {
435 std_input
= context
->std_input
;
437 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
438 return EXEC_INPUT_NULL
;
440 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
441 return EXEC_INPUT_NULL
;
443 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
444 return EXEC_INPUT_NULL
;
449 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
451 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
452 return EXEC_OUTPUT_INHERIT
;
457 static int setup_input(
458 const ExecContext
*context
,
459 const ExecParameters
*params
,
461 const int named_iofds
[static 3]) {
469 if (params
->stdin_fd
>= 0) {
470 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
473 /* Try to make this the controlling tty, if it is a tty, and reset it */
474 if (isatty(STDIN_FILENO
)) {
475 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
476 (void) reset_terminal_fd(STDIN_FILENO
, true);
482 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
486 case EXEC_INPUT_NULL
:
487 return open_null_as(O_RDONLY
, STDIN_FILENO
);
490 case EXEC_INPUT_TTY_FORCE
:
491 case EXEC_INPUT_TTY_FAIL
: {
494 fd
= acquire_terminal(exec_context_tty_path(context
),
495 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
496 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
497 ACQUIRE_TERMINAL_WAIT
,
502 return move_fd(fd
, STDIN_FILENO
, false);
505 case EXEC_INPUT_SOCKET
:
506 assert(socket_fd
>= 0);
508 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
510 case EXEC_INPUT_NAMED_FD
:
511 assert(named_iofds
[STDIN_FILENO
] >= 0);
513 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
514 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
516 case EXEC_INPUT_DATA
: {
519 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
523 return move_fd(fd
, STDIN_FILENO
, false);
526 case EXEC_INPUT_FILE
: {
530 assert(context
->stdio_file
[STDIN_FILENO
]);
532 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
533 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
535 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
539 return move_fd(fd
, STDIN_FILENO
, false);
543 assert_not_reached("Unknown input type");
547 static bool can_inherit_stderr_from_stdout(
548 const ExecContext
*context
,
554 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
557 if (e
== EXEC_OUTPUT_INHERIT
)
562 if (e
== EXEC_OUTPUT_NAMED_FD
)
563 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
565 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
566 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
571 static int setup_output(
573 const ExecContext
*context
,
574 const ExecParameters
*params
,
577 const int named_iofds
[static 3],
581 dev_t
*journal_stream_dev
,
582 ino_t
*journal_stream_ino
) {
592 assert(journal_stream_dev
);
593 assert(journal_stream_ino
);
595 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
597 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
600 return STDOUT_FILENO
;
603 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
604 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
607 return STDERR_FILENO
;
610 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
611 o
= fixup_output(context
->std_output
, socket_fd
);
613 if (fileno
== STDERR_FILENO
) {
615 e
= fixup_output(context
->std_error
, socket_fd
);
617 /* This expects the input and output are already set up */
619 /* Don't change the stderr file descriptor if we inherit all
620 * the way and are not on a tty */
621 if (e
== EXEC_OUTPUT_INHERIT
&&
622 o
== EXEC_OUTPUT_INHERIT
&&
623 i
== EXEC_INPUT_NULL
&&
624 !is_terminal_input(context
->std_input
) &&
628 /* Duplicate from stdout if possible */
629 if (can_inherit_stderr_from_stdout(context
, o
, e
))
630 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
634 } else if (o
== EXEC_OUTPUT_INHERIT
) {
635 /* If input got downgraded, inherit the original value */
636 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
637 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
639 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
640 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
641 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
643 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
647 /* We need to open /dev/null here anew, to get the right access mode. */
648 return open_null_as(O_WRONLY
, fileno
);
653 case EXEC_OUTPUT_NULL
:
654 return open_null_as(O_WRONLY
, fileno
);
656 case EXEC_OUTPUT_TTY
:
657 if (is_terminal_input(i
))
658 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
660 /* We don't reset the terminal if this is just about output */
661 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
663 case EXEC_OUTPUT_KMSG
:
664 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
665 case EXEC_OUTPUT_JOURNAL
:
666 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
667 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
669 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
670 r
= open_null_as(O_WRONLY
, fileno
);
674 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
675 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
676 * services to detect whether they are connected to the journal or not.
678 * If both stdout and stderr are connected to a stream then let's make sure to store the data
679 * about STDERR as that's usually the best way to do logging. */
681 if (fstat(fileno
, &st
) >= 0 &&
682 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
683 *journal_stream_dev
= st
.st_dev
;
684 *journal_stream_ino
= st
.st_ino
;
689 case EXEC_OUTPUT_SOCKET
:
690 assert(socket_fd
>= 0);
692 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
694 case EXEC_OUTPUT_NAMED_FD
:
695 assert(named_iofds
[fileno
] >= 0);
697 (void) fd_nonblock(named_iofds
[fileno
], false);
698 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
700 case EXEC_OUTPUT_FILE
:
701 case EXEC_OUTPUT_FILE_APPEND
: {
705 assert(context
->stdio_file
[fileno
]);
707 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
708 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
711 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
714 if (o
== EXEC_OUTPUT_FILE_APPEND
)
717 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
721 return move_fd(fd
, fileno
, 0);
725 assert_not_reached("Unknown error type");
729 static int chown_terminal(int fd
, uid_t uid
) {
734 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
735 if (isatty(fd
) < 1) {
736 if (IN_SET(errno
, EINVAL
, ENOTTY
))
737 return 0; /* not a tty */
742 /* This might fail. What matters are the results. */
743 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, -1);
750 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
751 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
754 assert(_saved_stdin
);
755 assert(_saved_stdout
);
757 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
761 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
762 if (saved_stdout
< 0)
765 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
769 r
= chown_terminal(fd
, getuid());
773 r
= reset_terminal_fd(fd
, true);
777 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
782 *_saved_stdin
= saved_stdin
;
783 *_saved_stdout
= saved_stdout
;
785 saved_stdin
= saved_stdout
= -1;
790 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
793 if (err
== -ETIMEDOUT
)
794 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
797 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
801 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
802 _cleanup_close_
int fd
= -1;
806 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
810 write_confirm_error_fd(err
, fd
, u
);
813 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
817 assert(saved_stdout
);
821 if (*saved_stdin
>= 0)
822 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
825 if (*saved_stdout
>= 0)
826 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
829 *saved_stdin
= safe_close(*saved_stdin
);
830 *saved_stdout
= safe_close(*saved_stdout
);
836 CONFIRM_PRETEND_FAILURE
= -1,
837 CONFIRM_PRETEND_SUCCESS
= 0,
841 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
842 int saved_stdout
= -1, saved_stdin
= -1, r
;
843 _cleanup_free_
char *e
= NULL
;
846 /* For any internal errors, assume a positive response. */
847 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
849 write_confirm_error(r
, vc
, u
);
850 return CONFIRM_EXECUTE
;
853 /* confirm_spawn might have been disabled while we were sleeping. */
854 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
859 e
= ellipsize(cmdline
, 60, 100);
867 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
869 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
876 printf("Resuming normal execution.\n");
877 manager_disable_confirm_spawn();
881 unit_dump(u
, stdout
, " ");
882 continue; /* ask again */
884 printf("Failing execution.\n");
885 r
= CONFIRM_PRETEND_FAILURE
;
888 printf(" c - continue, proceed without asking anymore\n"
889 " D - dump, show the state of the unit\n"
890 " f - fail, don't execute the command and pretend it failed\n"
892 " i - info, show a short summary of the unit\n"
893 " j - jobs, show jobs that are in progress\n"
894 " s - skip, don't execute the command and pretend it succeeded\n"
895 " y - yes, execute the command\n");
896 continue; /* ask again */
898 printf(" Description: %s\n"
901 u
->id
, u
->description
, cmdline
);
902 continue; /* ask again */
904 manager_dump_jobs(u
->manager
, stdout
, " ");
905 continue; /* ask again */
907 /* 'n' was removed in favor of 'f'. */
908 printf("Didn't understand 'n', did you mean 'f'?\n");
909 continue; /* ask again */
911 printf("Skipping execution.\n");
912 r
= CONFIRM_PRETEND_SUCCESS
;
918 assert_not_reached("Unhandled choice");
924 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
928 static int get_fixed_user(const ExecContext
*c
, const char **user
,
929 uid_t
*uid
, gid_t
*gid
,
930 const char **home
, const char **shell
) {
939 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
940 * (i.e. are "/" or "/bin/nologin"). */
943 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
951 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
961 r
= get_group_creds(&name
, gid
, 0);
969 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
970 const char *group
, gid_t gid
,
971 gid_t
**supplementary_gids
, int *ngids
) {
975 bool keep_groups
= false;
976 gid_t
*groups
= NULL
;
977 _cleanup_free_ gid_t
*l_gids
= NULL
;
982 * If user is given, then lookup GID and supplementary groups list.
983 * We avoid NSS lookups for gid=0. Also we have to initialize groups
984 * here and as early as possible so we keep the list of supplementary
985 * groups of the caller.
987 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
988 /* First step, initialize groups from /etc/groups */
989 if (initgroups(user
, gid
) < 0)
995 if (strv_isempty(c
->supplementary_groups
))
999 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1000 * be positive, otherwise fail.
1003 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1004 if (ngroups_max
<= 0)
1005 return errno_or_else(EOPNOTSUPP
);
1007 l_gids
= new(gid_t
, ngroups_max
);
1013 * Lookup the list of groups that the user belongs to, we
1014 * avoid NSS lookups here too for gid=0.
1017 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1022 STRV_FOREACH(i
, c
->supplementary_groups
) {
1025 if (k
>= ngroups_max
)
1029 r
= get_group_creds(&g
, l_gids
+k
, 0);
1037 * Sets ngids to zero to drop all supplementary groups, happens
1038 * when we are under root and SupplementaryGroups= is empty.
1045 /* Otherwise get the final list of supplementary groups */
1046 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1050 *supplementary_gids
= groups
;
1058 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1061 /* Handle SupplementaryGroups= if it is not empty */
1063 r
= maybe_setgroups(ngids
, supplementary_gids
);
1068 if (gid_is_valid(gid
)) {
1069 /* Then set our gids */
1070 if (setresgid(gid
, gid
, gid
) < 0)
1077 static int set_securebits(int bits
, int mask
) {
1078 int current
, applied
;
1079 current
= prctl(PR_GET_SECUREBITS
);
1082 /* Clear all securebits defined in mask and set bits */
1083 applied
= (current
& ~mask
) | bits
;
1084 if (current
== applied
)
1086 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1091 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1095 if (!uid_is_valid(uid
))
1098 /* Sets (but doesn't look up) the uid and make sure we keep the
1099 * capabilities while doing so. For setting secure bits the capability CAP_SETPCAP is
1100 * required, so we also need keep-caps in this case.
1103 if (context
->capability_ambient_set
!= 0 || context
->secure_bits
!= 0) {
1105 /* First step: If we need to keep capabilities but
1106 * drop privileges we need to make sure we keep our
1107 * caps, while we drop privileges. */
1109 /* Add KEEP_CAPS to the securebits */
1110 r
= set_securebits(1<<SECURE_KEEP_CAPS
, 0);
1116 /* Second step: actually set the uids */
1117 if (setresuid(uid
, uid
, uid
) < 0)
1120 /* At this point we should have all necessary capabilities but
1121 are otherwise a normal user. However, the caps might got
1122 corrupted due to the setresuid() so we need clean them up
1123 later. This is done outside of this call. */
1130 static int null_conv(
1132 const struct pam_message
**msg
,
1133 struct pam_response
**resp
,
1134 void *appdata_ptr
) {
1136 /* We don't support conversations */
1138 return PAM_CONV_ERR
;
1143 static int setup_pam(
1150 const int fds
[], size_t n_fds
) {
1154 static const struct pam_conv conv
= {
1159 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1160 pam_handle_t
*handle
= NULL
;
1162 int pam_code
= PAM_SUCCESS
, r
;
1163 char **nv
, **e
= NULL
;
1164 bool close_session
= false;
1165 pid_t pam_pid
= 0, parent_pid
;
1172 /* We set up PAM in the parent process, then fork. The child
1173 * will then stay around until killed via PR_GET_PDEATHSIG or
1174 * systemd via the cgroup logic. It will then remove the PAM
1175 * session again. The parent process will exec() the actual
1176 * daemon. We do things this way to ensure that the main PID
1177 * of the daemon is the one we initially fork()ed. */
1179 r
= barrier_create(&barrier
);
1183 if (log_get_max_level() < LOG_DEBUG
)
1184 flags
|= PAM_SILENT
;
1186 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1187 if (pam_code
!= PAM_SUCCESS
) {
1193 _cleanup_free_
char *q
= NULL
;
1195 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1196 * out if that's the case, and read the TTY off it. */
1198 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1199 tty
= strjoina("/dev/", q
);
1203 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1204 if (pam_code
!= PAM_SUCCESS
)
1208 STRV_FOREACH(nv
, *env
) {
1209 pam_code
= pam_putenv(handle
, *nv
);
1210 if (pam_code
!= PAM_SUCCESS
)
1214 pam_code
= pam_acct_mgmt(handle
, flags
);
1215 if (pam_code
!= PAM_SUCCESS
)
1218 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1219 if (pam_code
!= PAM_SUCCESS
)
1220 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1222 pam_code
= pam_open_session(handle
, flags
);
1223 if (pam_code
!= PAM_SUCCESS
)
1226 close_session
= true;
1228 e
= pam_getenvlist(handle
);
1230 pam_code
= PAM_BUF_ERR
;
1234 /* Block SIGTERM, so that we know that it won't get lost in
1237 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1239 parent_pid
= getpid_cached();
1241 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1245 int sig
, ret
= EXIT_PAM
;
1247 /* The child's job is to reset the PAM session on
1249 barrier_set_role(&barrier
, BARRIER_CHILD
);
1251 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1252 * are open here that have been opened by PAM. */
1253 (void) close_many(fds
, n_fds
);
1255 /* Drop privileges - we don't need any to pam_close_session
1256 * and this will make PR_SET_PDEATHSIG work in most cases.
1257 * If this fails, ignore the error - but expect sd-pam threads
1258 * to fail to exit normally */
1260 r
= maybe_setgroups(0, NULL
);
1262 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1263 if (setresgid(gid
, gid
, gid
) < 0)
1264 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1265 if (setresuid(uid
, uid
, uid
) < 0)
1266 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1268 (void) ignore_signals(SIGPIPE
, -1);
1270 /* Wait until our parent died. This will only work if
1271 * the above setresuid() succeeds, otherwise the kernel
1272 * will not allow unprivileged parents kill their privileged
1273 * children this way. We rely on the control groups kill logic
1274 * to do the rest for us. */
1275 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1278 /* Tell the parent that our setup is done. This is especially
1279 * important regarding dropping privileges. Otherwise, unit
1280 * setup might race against our setresuid(2) call.
1282 * If the parent aborted, we'll detect this below, hence ignore
1283 * return failure here. */
1284 (void) barrier_place(&barrier
);
1286 /* Check if our parent process might already have died? */
1287 if (getppid() == parent_pid
) {
1290 assert_se(sigemptyset(&ss
) >= 0);
1291 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1294 if (sigwait(&ss
, &sig
) < 0) {
1301 assert(sig
== SIGTERM
);
1306 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1307 if (pam_code
!= PAM_SUCCESS
)
1310 /* If our parent died we'll end the session */
1311 if (getppid() != parent_pid
) {
1312 pam_code
= pam_close_session(handle
, flags
);
1313 if (pam_code
!= PAM_SUCCESS
)
1320 pam_end(handle
, pam_code
| flags
);
1324 barrier_set_role(&barrier
, BARRIER_PARENT
);
1326 /* If the child was forked off successfully it will do all the
1327 * cleanups, so forget about the handle here. */
1330 /* Unblock SIGTERM again in the parent */
1331 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1333 /* We close the log explicitly here, since the PAM modules
1334 * might have opened it, but we don't want this fd around. */
1337 /* Synchronously wait for the child to initialize. We don't care for
1338 * errors as we cannot recover. However, warn loudly if it happens. */
1339 if (!barrier_place_and_sync(&barrier
))
1340 log_error("PAM initialization failed");
1342 return strv_free_and_replace(*env
, e
);
1345 if (pam_code
!= PAM_SUCCESS
) {
1346 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1347 r
= -EPERM
; /* PAM errors do not map to errno */
1349 log_error_errno(r
, "PAM failed: %m");
1353 pam_code
= pam_close_session(handle
, flags
);
1355 pam_end(handle
, pam_code
| flags
);
1367 static void rename_process_from_path(const char *path
) {
1368 char process_name
[11];
1372 /* This resulting string must fit in 10 chars (i.e. the length
1373 * of "/sbin/init") to look pretty in /bin/ps */
1377 rename_process("(...)");
1383 /* The end of the process name is usually more
1384 * interesting, since the first bit might just be
1390 process_name
[0] = '(';
1391 memcpy(process_name
+1, p
, l
);
1392 process_name
[1+l
] = ')';
1393 process_name
[1+l
+1] = 0;
1395 rename_process(process_name
);
1398 static bool context_has_address_families(const ExecContext
*c
) {
1401 return c
->address_families_allow_list
||
1402 !set_isempty(c
->address_families
);
1405 static bool context_has_syscall_filters(const ExecContext
*c
) {
1408 return c
->syscall_allow_list
||
1409 !hashmap_isempty(c
->syscall_filter
);
1412 static bool context_has_syscall_logs(const ExecContext
*c
) {
1415 return c
->syscall_log_allow_list
||
1416 !hashmap_isempty(c
->syscall_log
);
1419 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1422 if (c
->no_new_privileges
)
1425 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1428 /* We need NNP if we have any form of seccomp and are unprivileged */
1429 return context_has_address_families(c
) ||
1430 c
->memory_deny_write_execute
||
1431 c
->restrict_realtime
||
1432 c
->restrict_suid_sgid
||
1433 exec_context_restrict_namespaces_set(c
) ||
1435 c
->protect_kernel_tunables
||
1436 c
->protect_kernel_modules
||
1437 c
->protect_kernel_logs
||
1438 c
->private_devices
||
1439 context_has_syscall_filters(c
) ||
1440 context_has_syscall_logs(c
) ||
1441 !set_isempty(c
->syscall_archs
) ||
1442 c
->lock_personality
||
1443 c
->protect_hostname
;
1446 static bool exec_context_has_credentials(const ExecContext
*context
) {
1450 return !hashmap_isempty(context
->set_credentials
) ||
1451 context
->load_credentials
;
1456 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1458 if (is_seccomp_available())
1461 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1465 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1466 uint32_t negative_action
, default_action
, action
;
1472 if (!context_has_syscall_filters(c
))
1475 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1478 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1480 if (c
->syscall_allow_list
) {
1481 default_action
= negative_action
;
1482 action
= SCMP_ACT_ALLOW
;
1484 default_action
= SCMP_ACT_ALLOW
;
1485 action
= negative_action
;
1488 if (needs_ambient_hack
) {
1489 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1494 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1497 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1499 uint32_t default_action
, action
;
1505 if (!context_has_syscall_logs(c
))
1509 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1512 if (c
->syscall_log_allow_list
) {
1513 /* Log nothing but the ones listed */
1514 default_action
= SCMP_ACT_ALLOW
;
1515 action
= SCMP_ACT_LOG
;
1517 /* Log everything but the ones listed */
1518 default_action
= SCMP_ACT_LOG
;
1519 action
= SCMP_ACT_ALLOW
;
1522 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1524 /* old libseccomp */
1525 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1530 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1534 if (set_isempty(c
->syscall_archs
))
1537 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1540 return seccomp_restrict_archs(c
->syscall_archs
);
1543 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1547 if (!context_has_address_families(c
))
1550 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1553 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1556 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1560 if (!c
->memory_deny_write_execute
)
1563 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1566 return seccomp_memory_deny_write_execute();
1569 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1573 if (!c
->restrict_realtime
)
1576 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1579 return seccomp_restrict_realtime();
1582 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1586 if (!c
->restrict_suid_sgid
)
1589 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1592 return seccomp_restrict_suid_sgid();
1595 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1599 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1600 * let's protect even those systems where this is left on in the kernel. */
1602 if (!c
->protect_kernel_tunables
)
1605 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1608 return seccomp_protect_sysctl();
1611 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1615 /* Turn off module syscalls on ProtectKernelModules=yes */
1617 if (!c
->protect_kernel_modules
)
1620 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1623 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1626 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1630 if (!c
->protect_kernel_logs
)
1633 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1636 return seccomp_protect_syslog();
1639 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1643 if (!c
->protect_clock
)
1646 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1649 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1652 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1656 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1658 if (!c
->private_devices
)
1661 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1664 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1667 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1671 if (!exec_context_restrict_namespaces_set(c
))
1674 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1677 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1680 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1681 unsigned long personality
;
1687 if (!c
->lock_personality
)
1690 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1693 personality
= c
->personality
;
1695 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1696 if (personality
== PERSONALITY_INVALID
) {
1698 r
= opinionated_personality(&personality
);
1703 return seccomp_lock_personality(personality
);
1708 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1712 if (!c
->protect_hostname
)
1715 if (ns_type_supported(NAMESPACE_UTS
)) {
1716 if (unshare(CLONE_NEWUTS
) < 0) {
1717 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1718 *ret_exit_status
= EXIT_NAMESPACE
;
1719 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1722 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1725 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1730 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1733 r
= seccomp_protect_hostname();
1735 *ret_exit_status
= EXIT_SECCOMP
;
1736 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1743 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1746 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1747 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1749 if (idle_pipe
[0] >= 0) {
1752 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1754 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1757 /* Signal systemd that we are bored and want to continue. */
1758 n
= write(idle_pipe
[3], "x", 1);
1760 /* Wait for systemd to react to the signal above. */
1761 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1764 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1768 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1771 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1773 static int build_environment(
1775 const ExecContext
*c
,
1776 const ExecParameters
*p
,
1779 const char *username
,
1781 dev_t journal_stream_dev
,
1782 ino_t journal_stream_ino
,
1785 _cleanup_strv_free_
char **our_env
= NULL
;
1794 #define N_ENV_VARS 16
1795 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1800 _cleanup_free_
char *joined
= NULL
;
1802 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1804 our_env
[n_env
++] = x
;
1806 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1808 our_env
[n_env
++] = x
;
1810 joined
= strv_join(p
->fd_names
, ":");
1814 x
= strjoin("LISTEN_FDNAMES=", joined
);
1817 our_env
[n_env
++] = x
;
1820 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1821 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1823 our_env
[n_env
++] = x
;
1825 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1827 our_env
[n_env
++] = x
;
1830 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1831 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1832 * check the database directly. */
1833 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1834 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1837 our_env
[n_env
++] = x
;
1841 x
= strjoin("HOME=", home
);
1845 path_simplify(x
+ 5, true);
1846 our_env
[n_env
++] = x
;
1850 x
= strjoin("LOGNAME=", username
);
1853 our_env
[n_env
++] = x
;
1855 x
= strjoin("USER=", username
);
1858 our_env
[n_env
++] = x
;
1862 x
= strjoin("SHELL=", shell
);
1866 path_simplify(x
+ 6, true);
1867 our_env
[n_env
++] = x
;
1870 if (!sd_id128_is_null(u
->invocation_id
)) {
1871 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1874 our_env
[n_env
++] = x
;
1877 if (exec_context_needs_term(c
)) {
1878 const char *tty_path
, *term
= NULL
;
1880 tty_path
= exec_context_tty_path(c
);
1882 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1883 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1884 * container manager passes to PID 1 ends up all the way in the console login shown. */
1886 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1887 term
= getenv("TERM");
1890 term
= default_term_for_tty(tty_path
);
1892 x
= strjoin("TERM=", term
);
1895 our_env
[n_env
++] = x
;
1898 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1899 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1902 our_env
[n_env
++] = x
;
1905 if (c
->log_namespace
) {
1906 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1910 our_env
[n_env
++] = x
;
1913 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1914 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1920 if (strv_isempty(c
->directories
[t
].paths
))
1923 n
= exec_directory_env_name_to_string(t
);
1927 pre
= strjoin(p
->prefix
[t
], "/");
1931 joined
= strv_join_full(c
->directories
[t
].paths
, ":", pre
, true);
1935 x
= strjoin(n
, "=", joined
);
1939 our_env
[n_env
++] = x
;
1942 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
1943 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
1947 our_env
[n_env
++] = x
;
1950 our_env
[n_env
++] = NULL
;
1951 assert(n_env
<= N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1954 *ret
= TAKE_PTR(our_env
);
1959 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1960 _cleanup_strv_free_
char **pass_env
= NULL
;
1961 size_t n_env
= 0, n_bufsize
= 0;
1964 STRV_FOREACH(i
, c
->pass_environment
) {
1965 _cleanup_free_
char *x
= NULL
;
1971 x
= strjoin(*i
, "=", v
);
1975 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1978 pass_env
[n_env
++] = TAKE_PTR(x
);
1979 pass_env
[n_env
] = NULL
;
1982 *ret
= TAKE_PTR(pass_env
);
1987 static bool exec_needs_mount_namespace(
1988 const ExecContext
*context
,
1989 const ExecParameters
*params
,
1990 const ExecRuntime
*runtime
) {
1995 if (context
->root_image
)
1998 if (!strv_isempty(context
->read_write_paths
) ||
1999 !strv_isempty(context
->read_only_paths
) ||
2000 !strv_isempty(context
->inaccessible_paths
))
2003 if (context
->n_bind_mounts
> 0)
2006 if (context
->n_temporary_filesystems
> 0)
2009 if (context
->n_mount_images
> 0)
2012 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
2015 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
2018 if (context
->private_devices
||
2019 context
->private_mounts
||
2020 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2021 context
->protect_home
!= PROTECT_HOME_NO
||
2022 context
->protect_kernel_tunables
||
2023 context
->protect_kernel_modules
||
2024 context
->protect_kernel_logs
||
2025 context
->protect_control_groups
||
2026 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2027 context
->proc_subset
!= PROC_SUBSET_ALL
)
2030 if (context
->root_directory
) {
2031 if (exec_context_get_effective_mount_apivfs(context
))
2034 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2035 if (!params
->prefix
[t
])
2038 if (!strv_isempty(context
->directories
[t
].paths
))
2043 if (context
->dynamic_user
&&
2044 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
2045 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
2046 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
2049 if (context
->log_namespace
)
2055 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2056 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2057 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
2058 _cleanup_close_
int unshare_ready_fd
= -1;
2059 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2064 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2065 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2066 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2067 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2068 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2069 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2070 * continues execution normally.
2071 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2072 * does not need CAP_SETUID to write the single line mapping to itself. */
2074 /* Can only set up multiple mappings with CAP_SETUID. */
2075 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
2076 r
= asprintf(&uid_map
,
2077 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2078 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2079 ouid
, ouid
, uid
, uid
);
2081 r
= asprintf(&uid_map
,
2082 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2088 /* Can only set up multiple mappings with CAP_SETGID. */
2089 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2090 r
= asprintf(&gid_map
,
2091 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2092 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2093 ogid
, ogid
, gid
, gid
);
2095 r
= asprintf(&gid_map
,
2096 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2102 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2104 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2105 if (unshare_ready_fd
< 0)
2108 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2110 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2113 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2117 _cleanup_close_
int fd
= -1;
2121 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2122 * here, after the parent opened its own user namespace. */
2125 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2127 /* Wait until the parent unshared the user namespace */
2128 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2133 /* Disable the setgroups() system call in the child user namespace, for good. */
2134 a
= procfs_file_alloca(ppid
, "setgroups");
2135 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2137 if (errno
!= ENOENT
) {
2142 /* If the file is missing the kernel is too old, let's continue anyway. */
2144 if (write(fd
, "deny\n", 5) < 0) {
2149 fd
= safe_close(fd
);
2152 /* First write the GID map */
2153 a
= procfs_file_alloca(ppid
, "gid_map");
2154 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2159 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2163 fd
= safe_close(fd
);
2165 /* The write the UID map */
2166 a
= procfs_file_alloca(ppid
, "uid_map");
2167 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2172 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2177 _exit(EXIT_SUCCESS
);
2180 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2181 _exit(EXIT_FAILURE
);
2184 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2186 if (unshare(CLONE_NEWUSER
) < 0)
2189 /* Let the child know that the namespace is ready now */
2190 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2193 /* Try to read an error code from the child */
2194 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2197 if (n
== sizeof(r
)) { /* an error code was sent to us */
2202 if (n
!= 0) /* on success we should have read 0 bytes */
2205 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2209 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2215 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2216 if (!context
->dynamic_user
)
2219 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2222 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2228 static int setup_exec_directory(
2229 const ExecContext
*context
,
2230 const ExecParameters
*params
,
2233 ExecDirectoryType type
,
2236 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2237 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2238 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2239 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2240 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2241 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2248 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2249 assert(exit_status
);
2251 if (!params
->prefix
[type
])
2254 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2255 if (!uid_is_valid(uid
))
2257 if (!gid_is_valid(gid
))
2261 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2262 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2264 p
= path_join(params
->prefix
[type
], *rt
);
2270 r
= mkdir_parents_label(p
, 0755);
2274 if (exec_directory_is_private(context
, type
)) {
2275 _cleanup_free_
char *private_root
= NULL
;
2277 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2278 * case we want to avoid leaving a directory around fully accessible that is owned by
2279 * a dynamic user whose UID is later on reused. To lock this down we use the same
2280 * trick used by container managers to prohibit host users to get access to files of
2281 * the same UID in containers: we place everything inside a directory that has an
2282 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2283 * for unprivileged host code. We then use fs namespacing to make this directory
2284 * permeable for the service itself.
2286 * Specifically: for a service which wants a special directory "foo/" we first create
2287 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2288 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2289 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2290 * unprivileged host users can't look into it. Inside of the namespace of the unit
2291 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2292 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2293 * for the service and making sure it only gets access to the dirs it needs but no
2294 * others. Tricky? Yes, absolutely, but it works!
2296 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2297 * to be owned by the service itself.
2299 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2300 * for sharing files or sockets with other services. */
2302 private_root
= path_join(params
->prefix
[type
], "private");
2303 if (!private_root
) {
2308 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2309 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2313 pp
= path_join(private_root
, *rt
);
2319 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2320 r
= mkdir_parents_label(pp
, 0755);
2324 if (is_dir(p
, false) > 0 &&
2325 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2327 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2328 * it over. Most likely the service has been upgraded from one that didn't use
2329 * DynamicUser=1, to one that does. */
2331 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2332 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2333 exec_directory_type_to_string(type
), p
, pp
);
2335 if (rename(p
, pp
) < 0) {
2340 /* Otherwise, create the actual directory for the service */
2342 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2343 if (r
< 0 && r
!= -EEXIST
)
2347 /* And link it up from the original place */
2348 r
= symlink_idempotent(pp
, p
, true);
2353 _cleanup_free_
char *target
= NULL
;
2355 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2356 readlink_and_make_absolute(p
, &target
) >= 0) {
2357 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2359 /* This already exists and is a symlink? Interesting. Maybe it's one created
2360 * by DynamicUser=1 (see above)?
2362 * We do this for all directory types except for ConfigurationDirectory=,
2363 * since they all support the private/ symlink logic at least in some
2364 * configurations, see above. */
2366 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2370 q
= path_join(params
->prefix
[type
], "private", *rt
);
2376 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2377 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2381 if (path_equal(q_resolved
, target_resolved
)) {
2383 /* Hmm, apparently DynamicUser= was once turned on for this service,
2384 * but is no longer. Let's move the directory back up. */
2386 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2387 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2388 exec_directory_type_to_string(type
), q
, p
);
2390 if (unlink(p
) < 0) {
2395 if (rename(q
, p
) < 0) {
2402 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2407 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2410 /* Don't change the owner/access mode of the configuration directory,
2411 * as in the common case it is not written to by a service, and shall
2412 * not be writable. */
2414 if (stat(p
, &st
) < 0) {
2419 /* Still complain if the access mode doesn't match */
2420 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2421 log_warning("%s \'%s\' already exists but the mode is different. "
2422 "(File system: %o %sMode: %o)",
2423 exec_directory_type_to_string(type
), *rt
,
2424 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2431 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2432 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2433 * current UID/GID ownership.) */
2434 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2438 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2439 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2440 * assignments to exist.*/
2441 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2449 *exit_status
= exit_status_table
[type
];
2453 static int write_credential(
2459 bool ownership_ok
) {
2461 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2462 _cleanup_close_
int fd
= -1;
2465 r
= tempfn_random_child("", "cred", &tmp
);
2469 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2475 r
= loop_write(fd
, data
, size
, /* do_pool = */ false);
2479 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2482 if (uid_is_valid(uid
) && uid
!= getuid()) {
2483 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2485 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2488 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2489 * to express: that the user gets read access and nothing
2490 * else. But if the backing fs can't support that (e.g. ramfs)
2491 * then we can use file ownership instead. But that's only safe if
2492 * we can then re-mount the whole thing read-only, so that the
2493 * user can no longer chmod() the file to gain write access. */
2496 if (fchown(fd
, uid
, (gid_t
) -1) < 0)
2501 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2508 #define CREDENTIALS_BYTES_MAX (1024LU * 1024LU) /* Refuse to pass more than 1M, after all this is unswappable memory */
2510 static int acquire_credentials(
2511 const ExecContext
*context
,
2512 const ExecParameters
*params
,
2516 bool ownership_ok
) {
2518 uint64_t left
= CREDENTIALS_BYTES_MAX
;
2519 _cleanup_close_
int dfd
= -1;
2520 ExecSetCredential
*sc
;
2527 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
2531 /* First we use the literally specified credentials. Note that they might be overridden again below,
2532 * and thus act as a "default" if the same credential is specified multiple times */
2533 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
2536 add
= strlen(sc
->id
) + sc
->size
;
2540 r
= write_credential(dfd
, sc
->id
, sc
->data
, sc
->size
, uid
, ownership_ok
);
2547 /* Then, load credential off disk (or acquire via AF_UNIX socket) */
2548 STRV_FOREACH_PAIR(id
, fn
, context
->load_credentials
) {
2549 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
2550 _cleanup_(erase_and_freep
) char *data
= NULL
;
2551 _cleanup_free_
char *j
= NULL
, *bindname
= NULL
;
2555 if (path_is_absolute(*fn
)) {
2556 /* If this is an absolute path, read the data directly from it, and support AF_UNIX sockets */
2558 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2560 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2561 * via the source socket address in case we read off an AF_UNIX socket. */
2562 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, *id
) < 0)
2565 } else if (params
->received_credentials
) {
2566 /* If this is a relative path, take it relative to the credentials we received
2567 * ourselves. We don't support the AF_UNIX stuff in this mode, since we are operating
2568 * on a credential store, i.e. this is guaranteed to be regular files. */
2569 j
= path_join(params
->received_credentials
, *fn
);
2579 r
= read_full_file_full(AT_FDCWD
, source
, flags
, bindname
, &data
, &size
);
2583 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 */
2588 add
= strlen(*id
) + size
;
2592 r
= write_credential(dfd
, *id
, data
, size
, uid
, ownership_ok
);
2599 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
2602 /* After we created all keys with the right perms, also make sure the credential store as a whole is
2605 if (uid_is_valid(uid
) && uid
!= getuid()) {
2606 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
2608 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2614 if (fchown(dfd
, uid
, (gid_t
) -1) < 0)
2622 static int setup_credentials_internal(
2623 const ExecContext
*context
,
2624 const ExecParameters
*params
,
2626 const char *final
, /* This is where the credential store shall eventually end up at */
2627 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
2628 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
2629 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
2632 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
2633 * if we mounted something; false if we definitely can't mount anything */
2641 if (reuse_workspace
) {
2642 r
= path_is_mount_point(workspace
, NULL
, 0);
2646 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
2648 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
2650 workspace_mounted
= -1; /* ditto */
2652 r
= path_is_mount_point(final
, NULL
, 0);
2656 /* If the final place already has something mounted, we use that. If the workspace also has
2657 * something mounted we assume it's actually the same mount (but with MS_RDONLY
2659 final_mounted
= true;
2661 if (workspace_mounted
< 0) {
2662 /* If the final place is mounted, but the workspace we isn't, then let's bind mount
2663 * the final version to the workspace, and make it writable, so that we can make
2666 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2670 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2674 workspace_mounted
= true;
2677 final_mounted
= false;
2679 if (workspace_mounted
< 0) {
2680 /* Nothing is mounted on the workspace yet, let's try to mount something now */
2681 for (int try = 0;; try++) {
2684 /* Try "ramfs" first, since it's not swap backed */
2685 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
2687 workspace_mounted
= true;
2691 } else if (try == 1) {
2692 _cleanup_free_
char *opts
= NULL
;
2694 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%lu", CREDENTIALS_BYTES_MAX
) < 0)
2697 /* Fall back to "tmpfs" otherwise */
2698 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
2700 workspace_mounted
= true;
2705 /* 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. */
2706 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2708 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
2711 if (must_mount
) /* If we it's not OK to use the plain directory
2712 * fallback, propagate all errors too */
2715 /* If we lack privileges to bind mount stuff, then let's gracefully
2716 * proceed for compat with container envs, and just use the final dir
2719 workspace_mounted
= false;
2723 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
2724 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2728 workspace_mounted
= true;
2734 assert(!must_mount
|| workspace_mounted
> 0);
2735 where
= workspace_mounted
? workspace
: final
;
2737 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
2741 if (workspace_mounted
) {
2742 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
2743 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2747 /* And mount it to the final place, read-only */
2749 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
2751 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
2755 _cleanup_free_
char *parent
= NULL
;
2757 /* If we do not have our own mount put used the plain directory fallback, then we need to
2758 * open access to the top-level credential directory and the per-service directory now */
2760 parent
= dirname_malloc(final
);
2763 if (chmod(parent
, 0755) < 0)
2770 static int setup_credentials(
2771 const ExecContext
*context
,
2772 const ExecParameters
*params
,
2776 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
2783 if (!exec_context_has_credentials(context
))
2786 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
2789 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
2790 * and the subdir we mount over with a read-only file system readable by the service's user */
2791 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
2795 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
2796 if (r
< 0 && r
!= -EEXIST
)
2799 p
= path_join(q
, unit
);
2803 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
2804 if (r
< 0 && r
!= -EEXIST
)
2807 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
2809 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
2811 /* If this is not a privilege or support issue then propagate the error */
2812 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2815 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
2816 * it into place, so that users can't access half-initialized credential stores. */
2817 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
2821 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
2822 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
2823 * after it is fully set up */
2824 u
= path_join(t
, unit
);
2828 FOREACH_STRING(i
, t
, u
) {
2829 r
= mkdir_label(i
, 0700);
2830 if (r
< 0 && r
!= -EEXIST
)
2834 r
= setup_credentials_internal(
2838 p
, /* final mount point */
2839 u
, /* temporary workspace to overmount */
2840 true, /* reuse the workspace if it is already a mount */
2841 false, /* it's OK to fall back to a plain directory if we can't mount anything */
2844 (void) rmdir(u
); /* remove the workspace again if we can. */
2849 } else if (r
== 0) {
2851 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
2852 * we can use the same directory for all cases, after turning off propagation. Question
2853 * though is: where do we turn off propagation exactly, and where do we place the workspace
2854 * directory? We need some place that is guaranteed to be a mount point in the host, and
2855 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
2856 * since we ultimately want to move the resulting file system there, i.e. we need propagation
2857 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
2858 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
2859 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
2860 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
2861 * propagation on the former, and then overmount the latter.
2863 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
2864 * for this purpose, but there are few other candidates that work equally well for us, and
2865 * given that the we do this in a privately namespaced short-lived single-threaded process
2866 * that no one else sees this should be OK to do.*/
2868 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
2872 r
= setup_credentials_internal(
2876 p
, /* final mount point */
2877 "/dev/shm", /* temporary workspace to overmount */
2878 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
2879 true, /* insist that something is mounted, do not allow fallback to plain directory */
2884 _exit(EXIT_SUCCESS
);
2887 _exit(EXIT_FAILURE
);
2894 static int setup_smack(
2895 const ExecContext
*context
,
2896 const char *executable
) {
2902 if (context
->smack_process_label
) {
2903 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2907 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2909 _cleanup_free_
char *exec_label
= NULL
;
2911 r
= mac_smack_read(executable
, SMACK_ATTR_EXEC
, &exec_label
);
2912 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2915 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2925 static int compile_bind_mounts(
2926 const ExecContext
*context
,
2927 const ExecParameters
*params
,
2928 BindMount
**ret_bind_mounts
,
2929 size_t *ret_n_bind_mounts
,
2930 char ***ret_empty_directories
) {
2932 _cleanup_strv_free_
char **empty_directories
= NULL
;
2933 BindMount
*bind_mounts
;
2939 assert(ret_bind_mounts
);
2940 assert(ret_n_bind_mounts
);
2941 assert(ret_empty_directories
);
2943 n
= context
->n_bind_mounts
;
2944 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2945 if (!params
->prefix
[t
])
2948 n
+= strv_length(context
->directories
[t
].paths
);
2952 *ret_bind_mounts
= NULL
;
2953 *ret_n_bind_mounts
= 0;
2954 *ret_empty_directories
= NULL
;
2958 bind_mounts
= new(BindMount
, n
);
2962 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
2963 BindMount
*item
= context
->bind_mounts
+ i
;
2966 s
= strdup(item
->source
);
2972 d
= strdup(item
->destination
);
2979 bind_mounts
[h
++] = (BindMount
) {
2982 .read_only
= item
->read_only
,
2983 .recursive
= item
->recursive
,
2984 .ignore_enoent
= item
->ignore_enoent
,
2988 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2991 if (!params
->prefix
[t
])
2994 if (strv_isempty(context
->directories
[t
].paths
))
2997 if (exec_directory_is_private(context
, t
) &&
2998 !exec_context_with_rootfs(context
)) {
3001 /* So this is for a dynamic user, and we need to make sure the process can access its own
3002 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3003 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3005 private_root
= path_join(params
->prefix
[t
], "private");
3006 if (!private_root
) {
3011 r
= strv_consume(&empty_directories
, private_root
);
3016 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
3019 if (exec_directory_is_private(context
, t
))
3020 s
= path_join(params
->prefix
[t
], "private", *suffix
);
3022 s
= path_join(params
->prefix
[t
], *suffix
);
3028 if (exec_directory_is_private(context
, t
) &&
3029 exec_context_with_rootfs(context
))
3030 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3031 * directory is not created on the root directory. So, let's bind-mount the directory
3032 * on the 'non-private' place. */
3033 d
= path_join(params
->prefix
[t
], *suffix
);
3042 bind_mounts
[h
++] = (BindMount
) {
3046 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3048 .ignore_enoent
= false,
3055 *ret_bind_mounts
= bind_mounts
;
3056 *ret_n_bind_mounts
= n
;
3057 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3062 bind_mount_free_many(bind_mounts
, h
);
3066 static bool insist_on_sandboxing(
3067 const ExecContext
*context
,
3068 const char *root_dir
,
3069 const char *root_image
,
3070 const BindMount
*bind_mounts
,
3071 size_t n_bind_mounts
) {
3074 assert(n_bind_mounts
== 0 || bind_mounts
);
3076 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3077 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3078 * rearrange stuff in a way we cannot ignore gracefully. */
3080 if (context
->n_temporary_filesystems
> 0)
3083 if (root_dir
|| root_image
)
3086 if (context
->n_mount_images
> 0)
3089 if (context
->dynamic_user
)
3092 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3094 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3095 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3098 if (context
->log_namespace
)
3104 static int apply_mount_namespace(
3106 ExecCommandFlags command_flags
,
3107 const ExecContext
*context
,
3108 const ExecParameters
*params
,
3109 const ExecRuntime
*runtime
,
3110 char **error_path
) {
3112 _cleanup_strv_free_
char **empty_directories
= NULL
;
3113 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3114 const char *root_dir
= NULL
, *root_image
= NULL
;
3115 _cleanup_free_
char *creds_path
= NULL
;
3116 NamespaceInfo ns_info
;
3117 bool needs_sandboxing
;
3118 BindMount
*bind_mounts
= NULL
;
3119 size_t n_bind_mounts
= 0;
3124 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3125 root_image
= context
->root_image
;
3128 root_dir
= context
->root_directory
;
3131 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3135 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3136 if (needs_sandboxing
) {
3137 /* The runtime struct only contains the parent of the private /tmp,
3138 * which is non-accessible to world users. Inside of it there's a /tmp
3139 * that is sticky, and that's the one we want to use here.
3140 * This does not apply when we are using /run/systemd/empty as fallback. */
3142 if (context
->private_tmp
&& runtime
) {
3143 if (streq_ptr(runtime
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3144 tmp_dir
= runtime
->tmp_dir
;
3145 else if (runtime
->tmp_dir
)
3146 tmp_dir
= strjoina(runtime
->tmp_dir
, "/tmp");
3148 if (streq_ptr(runtime
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3149 var_tmp_dir
= runtime
->var_tmp_dir
;
3150 else if (runtime
->var_tmp_dir
)
3151 var_tmp_dir
= strjoina(runtime
->var_tmp_dir
, "/tmp");
3154 ns_info
= (NamespaceInfo
) {
3155 .ignore_protect_paths
= false,
3156 .private_dev
= context
->private_devices
,
3157 .protect_control_groups
= context
->protect_control_groups
,
3158 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3159 .protect_kernel_modules
= context
->protect_kernel_modules
,
3160 .protect_kernel_logs
= context
->protect_kernel_logs
,
3161 .protect_hostname
= context
->protect_hostname
,
3162 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3163 .private_mounts
= context
->private_mounts
,
3164 .protect_home
= context
->protect_home
,
3165 .protect_system
= context
->protect_system
,
3166 .protect_proc
= context
->protect_proc
,
3167 .proc_subset
= context
->proc_subset
,
3169 } else if (!context
->dynamic_user
&& root_dir
)
3171 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3172 * sandbox info, otherwise enforce it, don't ignore protected paths and
3173 * fail if we are enable to apply the sandbox inside the mount namespace.
3175 ns_info
= (NamespaceInfo
) {
3176 .ignore_protect_paths
= true,
3179 ns_info
= (NamespaceInfo
) {};
3181 if (context
->mount_flags
== MS_SHARED
)
3182 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3184 if (exec_context_has_credentials(context
) && params
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
3185 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3192 r
= setup_namespace(root_dir
, root_image
, context
->root_image_options
,
3193 &ns_info
, context
->read_write_paths
,
3194 needs_sandboxing
? context
->read_only_paths
: NULL
,
3195 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3199 context
->temporary_filesystems
,
3200 context
->n_temporary_filesystems
,
3201 context
->mount_images
,
3202 context
->n_mount_images
,
3206 context
->log_namespace
,
3207 context
->mount_flags
,
3208 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3209 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3210 context
->root_verity
,
3211 DISSECT_IMAGE_DISCARD_ON_LOOP
|DISSECT_IMAGE_RELAX_VAR_CHECK
|DISSECT_IMAGE_FSCK
,
3214 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3215 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3216 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3217 * completely different execution environment. */
3219 if (insist_on_sandboxing(
3221 root_dir
, root_image
,
3224 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3225 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3226 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
3230 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3236 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
3240 static int apply_working_directory(
3241 const ExecContext
*context
,
3242 const ExecParameters
*params
,
3249 assert(exit_status
);
3251 if (context
->working_directory_home
) {
3254 *exit_status
= EXIT_CHDIR
;
3261 wd
= empty_to_root(context
->working_directory
);
3263 if (params
->flags
& EXEC_APPLY_CHROOT
)
3266 d
= prefix_roota(context
->root_directory
, wd
);
3268 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3269 *exit_status
= EXIT_CHDIR
;
3276 static int apply_root_directory(
3277 const ExecContext
*context
,
3278 const ExecParameters
*params
,
3279 const bool needs_mount_ns
,
3283 assert(exit_status
);
3285 if (params
->flags
& EXEC_APPLY_CHROOT
)
3286 if (!needs_mount_ns
&& context
->root_directory
)
3287 if (chroot(context
->root_directory
) < 0) {
3288 *exit_status
= EXIT_CHROOT
;
3295 static int setup_keyring(
3297 const ExecContext
*context
,
3298 const ExecParameters
*p
,
3299 uid_t uid
, gid_t gid
) {
3301 key_serial_t keyring
;
3310 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3311 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3312 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3313 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3314 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3315 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3317 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3320 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3321 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3322 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3323 * & group is just as nasty as acquiring a reference to the user keyring. */
3325 saved_uid
= getuid();
3326 saved_gid
= getgid();
3328 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3329 if (setregid(gid
, -1) < 0)
3330 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3333 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3334 if (setreuid(uid
, -1) < 0) {
3335 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3340 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3341 if (keyring
== -1) {
3342 if (errno
== ENOSYS
)
3343 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3344 else if (ERRNO_IS_PRIVILEGE(errno
))
3345 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3346 else if (errno
== EDQUOT
)
3347 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3349 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3354 /* When requested link the user keyring into the session keyring. */
3355 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3357 if (keyctl(KEYCTL_LINK
,
3358 KEY_SPEC_USER_KEYRING
,
3359 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3360 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3365 /* Restore uid/gid back */
3366 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3367 if (setreuid(saved_uid
, -1) < 0) {
3368 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3373 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3374 if (setregid(saved_gid
, -1) < 0)
3375 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3378 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3379 if (!sd_id128_is_null(u
->invocation_id
)) {
3382 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3384 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3386 if (keyctl(KEYCTL_SETPERM
, key
,
3387 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3388 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3389 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3394 /* Revert back uid & gid for the last time, and exit */
3395 /* no extra logging, as only the first already reported error matters */
3396 if (getuid() != saved_uid
)
3397 (void) setreuid(saved_uid
, -1);
3399 if (getgid() != saved_gid
)
3400 (void) setregid(saved_gid
, -1);
3405 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3411 array
[(*n
)++] = pair
[0];
3413 array
[(*n
)++] = pair
[1];
3416 static int close_remaining_fds(
3417 const ExecParameters
*params
,
3418 const ExecRuntime
*runtime
,
3419 const DynamicCreds
*dcreds
,
3423 const int *fds
, size_t n_fds
) {
3425 size_t n_dont_close
= 0;
3426 int dont_close
[n_fds
+ 12];
3430 if (params
->stdin_fd
>= 0)
3431 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3432 if (params
->stdout_fd
>= 0)
3433 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3434 if (params
->stderr_fd
>= 0)
3435 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3438 dont_close
[n_dont_close
++] = socket_fd
;
3440 dont_close
[n_dont_close
++] = exec_fd
;
3442 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3443 n_dont_close
+= n_fds
;
3447 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3451 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3453 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3456 if (user_lookup_fd
>= 0)
3457 dont_close
[n_dont_close
++] = user_lookup_fd
;
3459 return close_all_fds(dont_close
, n_dont_close
);
3462 static int send_user_lookup(
3470 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3471 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3474 if (user_lookup_fd
< 0)
3477 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3480 if (writev(user_lookup_fd
,
3482 IOVEC_INIT(&uid
, sizeof(uid
)),
3483 IOVEC_INIT(&gid
, sizeof(gid
)),
3484 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3490 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3497 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3502 if (!c
->working_directory_home
)
3505 r
= get_home_dir(buf
);
3513 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3514 _cleanup_strv_free_
char ** list
= NULL
;
3521 assert(c
->dynamic_user
);
3523 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3524 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3527 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3530 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3536 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3539 if (exec_directory_is_private(c
, t
))
3540 e
= path_join(p
->prefix
[t
], "private", *i
);
3542 e
= path_join(p
->prefix
[t
], *i
);
3546 r
= strv_consume(&list
, e
);
3552 *ret
= TAKE_PTR(list
);
3557 static char *exec_command_line(char **argv
);
3559 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3560 bool using_subcgroup
;
3566 if (!params
->cgroup_path
)
3569 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3570 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3571 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3572 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3573 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3574 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3575 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3576 * flag, which is only passed for the former statements, not for the latter. */
3578 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3579 if (using_subcgroup
)
3580 p
= path_join(params
->cgroup_path
, ".control");
3582 p
= strdup(params
->cgroup_path
);
3587 return using_subcgroup
;
3590 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3591 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3597 if (!c
->numa_policy
.nodes
.set
) {
3598 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3602 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3608 return cpu_set_add_all(ret
, &s
);
3611 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3614 return c
->cpu_affinity_from_numa
;
3617 static int exec_child(
3619 const ExecCommand
*command
,
3620 const ExecContext
*context
,
3621 const ExecParameters
*params
,
3622 ExecRuntime
*runtime
,
3623 DynamicCreds
*dcreds
,
3625 const int named_iofds
[static 3],
3627 size_t n_socket_fds
,
3628 size_t n_storage_fds
,
3633 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3634 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
3635 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3636 const char *username
= NULL
, *groupname
= NULL
;
3637 _cleanup_free_
char *home_buffer
= NULL
;
3638 const char *home
= NULL
, *shell
= NULL
;
3639 char **final_argv
= NULL
;
3640 dev_t journal_stream_dev
= 0;
3641 ino_t journal_stream_ino
= 0;
3642 bool userns_set_up
= false;
3643 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3644 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3645 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3646 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3648 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3649 bool use_selinux
= false;
3652 bool use_smack
= false;
3655 bool use_apparmor
= false;
3657 uid_t saved_uid
= getuid();
3658 gid_t saved_gid
= getgid();
3659 uid_t uid
= UID_INVALID
;
3660 gid_t gid
= GID_INVALID
;
3663 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3664 int ngids_after_pam
= 0;
3670 assert(exit_status
);
3672 rename_process_from_path(command
->path
);
3674 /* We reset exactly these signals, since they are the
3675 * only ones we set to SIG_IGN in the main daemon. All
3676 * others we leave untouched because we set them to
3677 * SIG_DFL or a valid handler initially, both of which
3678 * will be demoted to SIG_DFL. */
3679 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3680 SIGNALS_IGNORE
, -1);
3682 if (context
->ignore_sigpipe
)
3683 (void) ignore_signals(SIGPIPE
, -1);
3685 r
= reset_signal_mask();
3687 *exit_status
= EXIT_SIGNAL_MASK
;
3688 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3691 if (params
->idle_pipe
)
3692 do_idle_pipe_dance(params
->idle_pipe
);
3694 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3695 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3696 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3697 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3700 log_set_open_when_needed(true);
3702 /* In case anything used libc syslog(), close this here, too */
3705 n_fds
= n_socket_fds
+ n_storage_fds
;
3706 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
3708 *exit_status
= EXIT_FDS
;
3709 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3712 if (!context
->same_pgrp
&&
3714 *exit_status
= EXIT_SETSID
;
3715 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3718 exec_context_tty_reset(context
, params
);
3720 if (unit_shall_confirm_spawn(unit
)) {
3721 const char *vc
= params
->confirm_spawn
;
3722 _cleanup_free_
char *cmdline
= NULL
;
3724 cmdline
= exec_command_line(command
->argv
);
3726 *exit_status
= EXIT_MEMORY
;
3730 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3731 if (r
!= CONFIRM_EXECUTE
) {
3732 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3733 *exit_status
= EXIT_SUCCESS
;
3736 *exit_status
= EXIT_CONFIRM
;
3737 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
3738 "Execution cancelled by the user");
3742 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3743 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3744 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3745 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3746 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3747 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3748 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3749 *exit_status
= EXIT_MEMORY
;
3750 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3753 if (context
->dynamic_user
&& dcreds
) {
3754 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3756 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3757 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
3758 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3759 *exit_status
= EXIT_USER
;
3760 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3763 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3765 *exit_status
= EXIT_MEMORY
;
3769 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3771 *exit_status
= EXIT_USER
;
3773 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3776 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3779 if (!uid_is_valid(uid
)) {
3780 *exit_status
= EXIT_USER
;
3781 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3785 if (!gid_is_valid(gid
)) {
3786 *exit_status
= EXIT_USER
;
3787 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3792 username
= dcreds
->user
->name
;
3795 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3797 *exit_status
= EXIT_USER
;
3798 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3801 r
= get_fixed_group(context
, &groupname
, &gid
);
3803 *exit_status
= EXIT_GROUP
;
3804 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3808 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3809 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3810 &supplementary_gids
, &ngids
);
3812 *exit_status
= EXIT_GROUP
;
3813 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3816 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3818 *exit_status
= EXIT_USER
;
3819 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3822 user_lookup_fd
= safe_close(user_lookup_fd
);
3824 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3826 *exit_status
= EXIT_CHDIR
;
3827 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3830 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3831 * must sure to drop O_NONBLOCK */
3833 (void) fd_nonblock(socket_fd
, false);
3835 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3836 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3837 if (params
->cgroup_path
) {
3838 _cleanup_free_
char *p
= NULL
;
3840 r
= exec_parameters_get_cgroup_path(params
, &p
);
3842 *exit_status
= EXIT_CGROUP
;
3843 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3846 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3848 *exit_status
= EXIT_CGROUP
;
3849 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3853 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3854 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3856 *exit_status
= EXIT_NETWORK
;
3857 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3861 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3863 *exit_status
= EXIT_STDIN
;
3864 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3867 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3869 *exit_status
= EXIT_STDOUT
;
3870 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3873 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3875 *exit_status
= EXIT_STDERR
;
3876 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3879 if (context
->oom_score_adjust_set
) {
3880 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3881 * prohibit write access to this file, and we shouldn't trip up over that. */
3882 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3883 if (ERRNO_IS_PRIVILEGE(r
))
3884 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3886 *exit_status
= EXIT_OOM_ADJUST
;
3887 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3891 if (context
->coredump_filter_set
) {
3892 r
= set_coredump_filter(context
->coredump_filter
);
3893 if (ERRNO_IS_PRIVILEGE(r
))
3894 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
3896 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
3899 if (context
->nice_set
) {
3900 r
= setpriority_closest(context
->nice
);
3902 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3905 if (context
->cpu_sched_set
) {
3906 struct sched_param param
= {
3907 .sched_priority
= context
->cpu_sched_priority
,
3910 r
= sched_setscheduler(0,
3911 context
->cpu_sched_policy
|
3912 (context
->cpu_sched_reset_on_fork
?
3913 SCHED_RESET_ON_FORK
: 0),
3916 *exit_status
= EXIT_SETSCHEDULER
;
3917 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3921 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
3922 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
3923 const CPUSet
*cpu_set
;
3925 if (context
->cpu_affinity_from_numa
) {
3926 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
3928 *exit_status
= EXIT_CPUAFFINITY
;
3929 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
3932 cpu_set
= &converted_cpu_set
;
3934 cpu_set
= &context
->cpu_set
;
3936 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
3937 *exit_status
= EXIT_CPUAFFINITY
;
3938 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3942 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
3943 r
= apply_numa_policy(&context
->numa_policy
);
3944 if (r
== -EOPNOTSUPP
)
3945 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
3947 *exit_status
= EXIT_NUMA_POLICY
;
3948 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
3952 if (context
->ioprio_set
)
3953 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3954 *exit_status
= EXIT_IOPRIO
;
3955 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3958 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3959 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3960 *exit_status
= EXIT_TIMERSLACK
;
3961 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3964 if (context
->personality
!= PERSONALITY_INVALID
) {
3965 r
= safe_personality(context
->personality
);
3967 *exit_status
= EXIT_PERSONALITY
;
3968 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3972 if (context
->utmp_id
)
3973 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3975 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3976 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3980 if (uid_is_valid(uid
)) {
3981 r
= chown_terminal(STDIN_FILENO
, uid
);
3983 *exit_status
= EXIT_STDIN
;
3984 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3988 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3989 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3990 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3991 * touch a single hierarchy too. */
3992 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3993 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3995 *exit_status
= EXIT_CGROUP
;
3996 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4000 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4001 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
4003 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4006 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4007 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4009 *exit_status
= EXIT_CREDENTIALS
;
4010 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4014 r
= build_environment(
4026 *exit_status
= EXIT_MEMORY
;
4030 r
= build_pass_environment(context
, &pass_env
);
4032 *exit_status
= EXIT_MEMORY
;
4036 accum_env
= strv_env_merge(5,
4037 params
->environment
,
4040 context
->environment
,
4043 *exit_status
= EXIT_MEMORY
;
4046 accum_env
= strv_env_clean(accum_env
);
4048 (void) umask(context
->umask
);
4050 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4052 *exit_status
= EXIT_KEYRING
;
4053 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4056 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
4057 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4059 /* 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 */
4060 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4062 /* 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 */
4063 if (needs_ambient_hack
)
4064 needs_setuid
= false;
4066 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4068 if (needs_sandboxing
) {
4069 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
4070 * present. The actual MAC context application will happen later, as late as possible, to avoid
4071 * impacting our own code paths. */
4074 use_selinux
= mac_selinux_use();
4077 use_smack
= mac_smack_use();
4080 use_apparmor
= mac_apparmor_use();
4084 if (needs_sandboxing
) {
4087 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4088 * is set here. (See below.) */
4090 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4092 *exit_status
= EXIT_LIMITS
;
4093 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4097 if (needs_setuid
&& context
->pam_name
&& username
) {
4098 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4099 * wins here. (See above.) */
4101 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4103 *exit_status
= EXIT_PAM
;
4104 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4107 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4108 if (ngids_after_pam
< 0) {
4109 *exit_status
= EXIT_MEMORY
;
4110 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4114 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4115 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4116 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4117 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4119 userns_set_up
= true;
4120 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4122 *exit_status
= EXIT_USER
;
4123 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4127 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4129 if (ns_type_supported(NAMESPACE_NET
)) {
4130 r
= setup_netns(runtime
->netns_storage_socket
);
4132 log_unit_warning_errno(unit
, r
,
4133 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4135 *exit_status
= EXIT_NETWORK
;
4136 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4138 } else if (context
->network_namespace_path
) {
4139 *exit_status
= EXIT_NETWORK
;
4140 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4141 "NetworkNamespacePath= is not supported, refusing.");
4143 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4146 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4147 if (needs_mount_namespace
) {
4148 _cleanup_free_
char *error_path
= NULL
;
4150 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4152 *exit_status
= EXIT_NAMESPACE
;
4153 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4154 error_path
? ": " : "", strempty(error_path
));
4158 if (needs_sandboxing
) {
4159 r
= apply_protect_hostname(unit
, context
, exit_status
);
4164 /* Drop groups as early as possible.
4165 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4166 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4168 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4169 int ngids_to_enforce
= 0;
4171 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4176 if (ngids_to_enforce
< 0) {
4177 *exit_status
= EXIT_MEMORY
;
4178 return log_unit_error_errno(unit
,
4180 "Failed to merge group lists. Group membership might be incorrect: %m");
4183 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4185 *exit_status
= EXIT_GROUP
;
4186 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4190 /* If the user namespace was not set up above, try to do it now.
4191 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4192 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4193 * case of mount namespaces being less privileged when the mount point list is copied from a
4194 * different user namespace). */
4196 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4197 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4199 *exit_status
= EXIT_USER
;
4200 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4204 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4207 _cleanup_free_
char *executable
= NULL
;
4208 r
= find_executable_full(command
->path
, false, &executable
);
4210 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4211 log_struct_errno(LOG_INFO
, r
,
4212 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4214 LOG_UNIT_INVOCATION_ID(unit
),
4215 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4217 "EXECUTABLE=%s", command
->path
);
4221 *exit_status
= EXIT_EXEC
;
4222 return log_struct_errno(LOG_INFO
, r
,
4223 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4225 LOG_UNIT_INVOCATION_ID(unit
),
4226 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4228 "EXECUTABLE=%s", command
->path
);
4232 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
4233 r
= mac_selinux_get_child_mls_label(socket_fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4235 *exit_status
= EXIT_SELINUX_CONTEXT
;
4236 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4241 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4242 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
4243 * however if we have it as we want to keep it open until the final execve(). */
4245 if (params
->exec_fd
>= 0) {
4246 exec_fd
= params
->exec_fd
;
4248 if (exec_fd
< 3 + (int) n_fds
) {
4251 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
4252 * process we are about to execute. */
4254 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
4256 *exit_status
= EXIT_FDS
;
4257 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
4260 CLOSE_AND_REPLACE(exec_fd
, moved_fd
);
4262 /* This fd should be FD_CLOEXEC already, but let's make sure. */
4263 r
= fd_cloexec(exec_fd
, true);
4265 *exit_status
= EXIT_FDS
;
4266 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
4270 fds_with_exec_fd
= newa(int, n_fds
+ 1);
4271 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
4272 fds_with_exec_fd
[n_fds
] = exec_fd
;
4273 n_fds_with_exec_fd
= n_fds
+ 1;
4275 fds_with_exec_fd
= fds
;
4276 n_fds_with_exec_fd
= n_fds
;
4279 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
4281 r
= shift_fds(fds
, n_fds
);
4283 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4285 *exit_status
= EXIT_FDS
;
4286 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4289 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4290 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4291 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4294 secure_bits
= context
->secure_bits
;
4296 if (needs_sandboxing
) {
4299 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4300 * requested. (Note this is placed after the general resource limit initialization, see
4301 * above, in order to take precedence.) */
4302 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4303 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4304 *exit_status
= EXIT_LIMITS
;
4305 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4310 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4311 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4313 r
= setup_smack(context
, executable
);
4315 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4316 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4321 bset
= context
->capability_bounding_set
;
4322 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4323 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4324 * instead of us doing that */
4325 if (needs_ambient_hack
)
4326 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4327 (UINT64_C(1) << CAP_SETUID
) |
4328 (UINT64_C(1) << CAP_SETGID
);
4330 if (!cap_test_all(bset
)) {
4331 r
= capability_bounding_set_drop(bset
, false);
4333 *exit_status
= EXIT_CAPABILITIES
;
4334 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4338 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4340 * To be able to raise the ambient capabilities after setresuid() they have to be
4341 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4342 * After setresuid() the ambient capabilities can be raised as they are present in
4343 * the permitted and inhertiable set. However it is possible that someone wants to
4344 * set ambient capabilities without changing the user, so we also set the ambient
4345 * capabilities here.
4346 * The requested ambient capabilities are raised in the inheritable set if the
4347 * second argument is true. */
4348 if (!needs_ambient_hack
) {
4349 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4351 *exit_status
= EXIT_CAPABILITIES
;
4352 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4357 /* chroot to root directory first, before we lose the ability to chroot */
4358 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4360 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4363 if (uid_is_valid(uid
)) {
4364 r
= enforce_user(context
, uid
);
4366 *exit_status
= EXIT_USER
;
4367 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4370 if (!needs_ambient_hack
&&
4371 context
->capability_ambient_set
!= 0) {
4373 /* Raise the ambient capabilities after user change. */
4374 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4376 *exit_status
= EXIT_CAPABILITIES
;
4377 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4383 /* Apply working directory here, because the working directory might be on NFS and only the user running
4384 * this service might have the correct privilege to change to the working directory */
4385 r
= apply_working_directory(context
, params
, home
, exit_status
);
4387 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4389 if (needs_sandboxing
) {
4390 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4391 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4392 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4393 * are restricted. */
4397 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4400 r
= setexeccon(exec_context
);
4402 *exit_status
= EXIT_SELINUX_CONTEXT
;
4403 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4410 if (use_apparmor
&& context
->apparmor_profile
) {
4411 r
= aa_change_onexec(context
->apparmor_profile
);
4412 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4413 *exit_status
= EXIT_APPARMOR_PROFILE
;
4414 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4419 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4420 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4422 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4423 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4424 * effective set here.
4425 * The effective set is overwritten during execve with the following values:
4426 * - ambient set (for non-root processes)
4427 * - (inheritable | bounding) set for root processes)
4429 * Hence there is no security impact to raise it in the effective set before execve
4431 r
= capability_gain_cap_setpcap(NULL
);
4433 *exit_status
= EXIT_CAPABILITIES
;
4434 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
4436 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
4437 *exit_status
= EXIT_SECUREBITS
;
4438 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
4442 if (context_has_no_new_privileges(context
))
4443 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
4444 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
4445 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
4449 r
= apply_address_families(unit
, context
);
4451 *exit_status
= EXIT_ADDRESS_FAMILIES
;
4452 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
4455 r
= apply_memory_deny_write_execute(unit
, context
);
4457 *exit_status
= EXIT_SECCOMP
;
4458 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
4461 r
= apply_restrict_realtime(unit
, context
);
4463 *exit_status
= EXIT_SECCOMP
;
4464 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
4467 r
= apply_restrict_suid_sgid(unit
, context
);
4469 *exit_status
= EXIT_SECCOMP
;
4470 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
4473 r
= apply_restrict_namespaces(unit
, context
);
4475 *exit_status
= EXIT_SECCOMP
;
4476 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
4479 r
= apply_protect_sysctl(unit
, context
);
4481 *exit_status
= EXIT_SECCOMP
;
4482 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
4485 r
= apply_protect_kernel_modules(unit
, context
);
4487 *exit_status
= EXIT_SECCOMP
;
4488 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
4491 r
= apply_protect_kernel_logs(unit
, context
);
4493 *exit_status
= EXIT_SECCOMP
;
4494 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
4497 r
= apply_protect_clock(unit
, context
);
4499 *exit_status
= EXIT_SECCOMP
;
4500 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
4503 r
= apply_private_devices(unit
, context
);
4505 *exit_status
= EXIT_SECCOMP
;
4506 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
4509 r
= apply_syscall_archs(unit
, context
);
4511 *exit_status
= EXIT_SECCOMP
;
4512 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
4515 r
= apply_lock_personality(unit
, context
);
4517 *exit_status
= EXIT_SECCOMP
;
4518 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
4521 r
= apply_syscall_log(unit
, context
);
4523 *exit_status
= EXIT_SECCOMP
;
4524 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
4527 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
4528 * by the filter as little as possible. */
4529 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
4531 *exit_status
= EXIT_SECCOMP
;
4532 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
4537 if (!strv_isempty(context
->unset_environment
)) {
4540 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
4542 *exit_status
= EXIT_MEMORY
;
4546 strv_free_and_replace(accum_env
, ee
);
4549 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
4550 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
4551 if (!replaced_argv
) {
4552 *exit_status
= EXIT_MEMORY
;
4555 final_argv
= replaced_argv
;
4557 final_argv
= command
->argv
;
4559 if (DEBUG_LOGGING
) {
4560 _cleanup_free_
char *line
;
4562 line
= exec_command_line(final_argv
);
4564 log_struct(LOG_DEBUG
,
4565 "EXECUTABLE=%s", executable
,
4566 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
4568 LOG_UNIT_INVOCATION_ID(unit
));
4574 /* We have finished with all our initializations. Let's now let the manager know that. From this point
4575 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
4577 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4578 *exit_status
= EXIT_EXEC
;
4579 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
4583 execve(executable
, final_argv
, accum_env
);
4589 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4590 * that POLLHUP on it no longer means execve() succeeded. */
4592 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4593 *exit_status
= EXIT_EXEC
;
4594 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4598 *exit_status
= EXIT_EXEC
;
4599 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
4602 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4603 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4605 int exec_spawn(Unit
*unit
,
4606 ExecCommand
*command
,
4607 const ExecContext
*context
,
4608 const ExecParameters
*params
,
4609 ExecRuntime
*runtime
,
4610 DynamicCreds
*dcreds
,
4613 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4614 _cleanup_free_
char *subcgroup_path
= NULL
;
4615 _cleanup_strv_free_
char **files_env
= NULL
;
4616 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4617 _cleanup_free_
char *line
= NULL
;
4625 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4627 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4628 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4629 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4631 if (params
->n_socket_fds
> 1) {
4632 log_unit_error(unit
, "Got more than one socket.");
4636 if (params
->n_socket_fds
== 0) {
4637 log_unit_error(unit
, "Got no socket.");
4641 socket_fd
= params
->fds
[0];
4645 n_socket_fds
= params
->n_socket_fds
;
4646 n_storage_fds
= params
->n_storage_fds
;
4649 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4651 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4653 r
= exec_context_load_environment(unit
, context
, &files_env
);
4655 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4657 line
= exec_command_line(command
->argv
);
4661 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
4662 and, until the next SELinux policy changes, we save further reloads in future children. */
4663 mac_selinux_maybe_reload();
4665 log_struct(LOG_DEBUG
,
4666 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
4667 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
4668 the mount namespace in the child, but we want to log
4669 from the parent, so we need to use the (possibly
4670 inaccurate) path here. */
4672 LOG_UNIT_INVOCATION_ID(unit
));
4674 if (params
->cgroup_path
) {
4675 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4677 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4678 if (r
> 0) { /* We are using a child cgroup */
4679 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4681 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4687 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4690 int exit_status
= EXIT_SUCCESS
;
4692 r
= exec_child(unit
,
4704 unit
->manager
->user_lookup_fds
[1],
4708 const char *status
=
4709 exit_status_to_string(exit_status
,
4710 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4712 log_struct_errno(LOG_ERR
, r
,
4713 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4715 LOG_UNIT_INVOCATION_ID(unit
),
4716 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4717 status
, command
->path
),
4718 "EXECUTABLE=%s", command
->path
);
4724 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4726 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4727 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4728 * process will be killed too). */
4730 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4732 exec_status_start(&command
->exec_status
, pid
);
4738 void exec_context_init(ExecContext
*c
) {
4742 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4743 c
->cpu_sched_policy
= SCHED_OTHER
;
4744 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4745 c
->syslog_level_prefix
= true;
4746 c
->ignore_sigpipe
= true;
4747 c
->timer_slack_nsec
= NSEC_INFINITY
;
4748 c
->personality
= PERSONALITY_INVALID
;
4749 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4750 c
->directories
[t
].mode
= 0755;
4751 c
->timeout_clean_usec
= USEC_INFINITY
;
4752 c
->capability_bounding_set
= CAP_ALL
;
4753 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4754 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4755 c
->log_level_max
= -1;
4757 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
4759 numa_policy_reset(&c
->numa_policy
);
4762 void exec_context_done(ExecContext
*c
) {
4765 c
->environment
= strv_free(c
->environment
);
4766 c
->environment_files
= strv_free(c
->environment_files
);
4767 c
->pass_environment
= strv_free(c
->pass_environment
);
4768 c
->unset_environment
= strv_free(c
->unset_environment
);
4770 rlimit_free_all(c
->rlimit
);
4772 for (size_t l
= 0; l
< 3; l
++) {
4773 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4774 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4777 c
->working_directory
= mfree(c
->working_directory
);
4778 c
->root_directory
= mfree(c
->root_directory
);
4779 c
->root_image
= mfree(c
->root_image
);
4780 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
4781 c
->root_hash
= mfree(c
->root_hash
);
4782 c
->root_hash_size
= 0;
4783 c
->root_hash_path
= mfree(c
->root_hash_path
);
4784 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
4785 c
->root_hash_sig_size
= 0;
4786 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
4787 c
->root_verity
= mfree(c
->root_verity
);
4788 c
->tty_path
= mfree(c
->tty_path
);
4789 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4790 c
->user
= mfree(c
->user
);
4791 c
->group
= mfree(c
->group
);
4793 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4795 c
->pam_name
= mfree(c
->pam_name
);
4797 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4798 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4799 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4801 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4802 c
->bind_mounts
= NULL
;
4803 c
->n_bind_mounts
= 0;
4804 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4805 c
->temporary_filesystems
= NULL
;
4806 c
->n_temporary_filesystems
= 0;
4807 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
4809 cpu_set_reset(&c
->cpu_set
);
4810 numa_policy_reset(&c
->numa_policy
);
4812 c
->utmp_id
= mfree(c
->utmp_id
);
4813 c
->selinux_context
= mfree(c
->selinux_context
);
4814 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4815 c
->smack_process_label
= mfree(c
->smack_process_label
);
4817 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4818 c
->syscall_archs
= set_free(c
->syscall_archs
);
4819 c
->address_families
= set_free(c
->address_families
);
4821 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4822 c
->directories
[t
].paths
= strv_free(c
->directories
[t
].paths
);
4824 c
->log_level_max
= -1;
4826 exec_context_free_log_extra_fields(c
);
4828 c
->log_ratelimit_interval_usec
= 0;
4829 c
->log_ratelimit_burst
= 0;
4831 c
->stdin_data
= mfree(c
->stdin_data
);
4832 c
->stdin_data_size
= 0;
4834 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4836 c
->log_namespace
= mfree(c
->log_namespace
);
4838 c
->load_credentials
= strv_free(c
->load_credentials
);
4839 c
->set_credentials
= hashmap_free(c
->set_credentials
);
4842 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4847 if (!runtime_prefix
)
4850 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4851 _cleanup_free_
char *p
;
4853 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4854 p
= path_join(runtime_prefix
, "private", *i
);
4856 p
= path_join(runtime_prefix
, *i
);
4860 /* We execute this synchronously, since we need to be sure this is gone when we start the
4862 (void) rm_rf(p
, REMOVE_ROOT
);
4868 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
4869 _cleanup_free_
char *p
= NULL
;
4873 if (!runtime_prefix
|| !unit
)
4876 p
= path_join(runtime_prefix
, "credentials", unit
);
4880 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
4881 * unmount it, and afterwards remove the mount point */
4882 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
4883 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
4888 static void exec_command_done(ExecCommand
*c
) {
4891 c
->path
= mfree(c
->path
);
4892 c
->argv
= strv_free(c
->argv
);
4895 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4898 for (i
= 0; i
< n
; i
++)
4899 exec_command_done(c
+i
);
4902 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4906 LIST_REMOVE(command
, c
, i
);
4907 exec_command_done(i
);
4914 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4915 for (size_t i
= 0; i
< n
; i
++)
4916 c
[i
] = exec_command_free_list(c
[i
]);
4919 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4920 for (size_t i
= 0; i
< n
; i
++)
4921 exec_status_reset(&c
[i
].exec_status
);
4924 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
4925 for (size_t i
= 0; i
< n
; i
++) {
4928 LIST_FOREACH(command
, z
, c
[i
])
4929 exec_status_reset(&z
->exec_status
);
4933 typedef struct InvalidEnvInfo
{
4938 static void invalid_env(const char *p
, void *userdata
) {
4939 InvalidEnvInfo
*info
= userdata
;
4941 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4944 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4950 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4953 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4956 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4959 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4962 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4965 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4972 static int exec_context_named_iofds(
4973 const ExecContext
*c
,
4974 const ExecParameters
*p
,
4975 int named_iofds
[static 3]) {
4978 const char* stdio_fdname
[3];
4983 assert(named_iofds
);
4985 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4986 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4987 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4989 for (size_t i
= 0; i
< 3; i
++)
4990 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4992 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4994 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
4995 if (named_iofds
[STDIN_FILENO
] < 0 &&
4996 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4997 stdio_fdname
[STDIN_FILENO
] &&
4998 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
5000 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
5003 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
5004 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
5005 stdio_fdname
[STDOUT_FILENO
] &&
5006 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
5008 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5011 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5012 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5013 stdio_fdname
[STDERR_FILENO
] &&
5014 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5016 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5020 return targets
== 0 ? 0 : -ENOENT
;
5023 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
5024 char **i
, **r
= NULL
;
5029 STRV_FOREACH(i
, c
->environment_files
) {
5032 bool ignore
= false;
5034 _cleanup_globfree_ glob_t pglob
= {};
5043 if (!path_is_absolute(fn
)) {
5051 /* Filename supports globbing, take all matching files */
5052 k
= safe_glob(fn
, 0, &pglob
);
5061 /* When we don't match anything, -ENOENT should be returned */
5062 assert(pglob
.gl_pathc
> 0);
5064 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5065 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5073 /* Log invalid environment variables with filename */
5075 InvalidEnvInfo info
= {
5077 .path
= pglob
.gl_pathv
[n
]
5080 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5088 m
= strv_env_merge(2, r
, p
);
5104 static bool tty_may_match_dev_console(const char *tty
) {
5105 _cleanup_free_
char *resolved
= NULL
;
5110 tty
= skip_dev_prefix(tty
);
5112 /* trivial identity? */
5113 if (streq(tty
, "console"))
5116 if (resolve_dev_console(&resolved
) < 0)
5117 return true; /* if we could not resolve, assume it may */
5119 /* "tty0" means the active VC, so it may be the same sometimes */
5120 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5123 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5126 return ec
->tty_reset
||
5128 ec
->tty_vt_disallocate
||
5129 is_terminal_input(ec
->std_input
) ||
5130 is_terminal_output(ec
->std_output
) ||
5131 is_terminal_output(ec
->std_error
);
5134 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5136 return exec_context_may_touch_tty(ec
) &&
5137 tty_may_match_dev_console(exec_context_tty_path(ec
));
5140 static void strv_fprintf(FILE *f
, char **l
) {
5146 fprintf(f
, " %s", *g
);
5149 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5150 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
5156 prefix
= strempty(prefix
);
5160 "%sWorkingDirectory: %s\n"
5161 "%sRootDirectory: %s\n"
5162 "%sNonBlocking: %s\n"
5163 "%sPrivateTmp: %s\n"
5164 "%sPrivateDevices: %s\n"
5165 "%sProtectKernelTunables: %s\n"
5166 "%sProtectKernelModules: %s\n"
5167 "%sProtectKernelLogs: %s\n"
5168 "%sProtectClock: %s\n"
5169 "%sProtectControlGroups: %s\n"
5170 "%sPrivateNetwork: %s\n"
5171 "%sPrivateUsers: %s\n"
5172 "%sProtectHome: %s\n"
5173 "%sProtectSystem: %s\n"
5174 "%sMountAPIVFS: %s\n"
5175 "%sIgnoreSIGPIPE: %s\n"
5176 "%sMemoryDenyWriteExecute: %s\n"
5177 "%sRestrictRealtime: %s\n"
5178 "%sRestrictSUIDSGID: %s\n"
5179 "%sKeyringMode: %s\n"
5180 "%sProtectHostname: %s\n"
5181 "%sProtectProc: %s\n"
5182 "%sProcSubset: %s\n",
5184 prefix
, empty_to_root(c
->working_directory
),
5185 prefix
, empty_to_root(c
->root_directory
),
5186 prefix
, yes_no(c
->non_blocking
),
5187 prefix
, yes_no(c
->private_tmp
),
5188 prefix
, yes_no(c
->private_devices
),
5189 prefix
, yes_no(c
->protect_kernel_tunables
),
5190 prefix
, yes_no(c
->protect_kernel_modules
),
5191 prefix
, yes_no(c
->protect_kernel_logs
),
5192 prefix
, yes_no(c
->protect_clock
),
5193 prefix
, yes_no(c
->protect_control_groups
),
5194 prefix
, yes_no(c
->private_network
),
5195 prefix
, yes_no(c
->private_users
),
5196 prefix
, protect_home_to_string(c
->protect_home
),
5197 prefix
, protect_system_to_string(c
->protect_system
),
5198 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
5199 prefix
, yes_no(c
->ignore_sigpipe
),
5200 prefix
, yes_no(c
->memory_deny_write_execute
),
5201 prefix
, yes_no(c
->restrict_realtime
),
5202 prefix
, yes_no(c
->restrict_suid_sgid
),
5203 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5204 prefix
, yes_no(c
->protect_hostname
),
5205 prefix
, protect_proc_to_string(c
->protect_proc
),
5206 prefix
, proc_subset_to_string(c
->proc_subset
));
5209 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5211 if (c
->root_image_options
) {
5214 fprintf(f
, "%sRootImageOptions:", prefix
);
5215 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5216 if (!isempty(o
->options
))
5217 fprintf(f
, " %s:%s",
5218 partition_designator_to_string(o
->partition_designator
),
5224 _cleanup_free_
char *encoded
= NULL
;
5225 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5227 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5230 if (c
->root_hash_path
)
5231 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5233 if (c
->root_hash_sig
) {
5234 _cleanup_free_
char *encoded
= NULL
;
5236 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5238 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5241 if (c
->root_hash_sig_path
)
5242 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5245 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5247 STRV_FOREACH(e
, c
->environment
)
5248 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5250 STRV_FOREACH(e
, c
->environment_files
)
5251 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5253 STRV_FOREACH(e
, c
->pass_environment
)
5254 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5256 STRV_FOREACH(e
, c
->unset_environment
)
5257 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5259 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5261 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5262 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5264 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
5265 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
5269 "%sTimeoutCleanSec: %s\n",
5270 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
5277 if (c
->oom_score_adjust_set
)
5279 "%sOOMScoreAdjust: %i\n",
5280 prefix
, c
->oom_score_adjust
);
5282 if (c
->coredump_filter_set
)
5284 "%sCoredumpFilter: 0x%"PRIx64
"\n",
5285 prefix
, c
->coredump_filter
);
5287 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5289 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5290 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5291 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5292 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5295 if (c
->ioprio_set
) {
5296 _cleanup_free_
char *class_str
= NULL
;
5298 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
5300 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5302 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
5305 if (c
->cpu_sched_set
) {
5306 _cleanup_free_
char *policy_str
= NULL
;
5308 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5310 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5313 "%sCPUSchedulingPriority: %i\n"
5314 "%sCPUSchedulingResetOnFork: %s\n",
5315 prefix
, c
->cpu_sched_priority
,
5316 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5319 if (c
->cpu_set
.set
) {
5320 _cleanup_free_
char *affinity
= NULL
;
5322 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5323 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5326 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5327 _cleanup_free_
char *nodes
= NULL
;
5329 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5330 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5331 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5334 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5335 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5338 "%sStandardInput: %s\n"
5339 "%sStandardOutput: %s\n"
5340 "%sStandardError: %s\n",
5341 prefix
, exec_input_to_string(c
->std_input
),
5342 prefix
, exec_output_to_string(c
->std_output
),
5343 prefix
, exec_output_to_string(c
->std_error
));
5345 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5346 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5347 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5348 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5349 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5350 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5352 if (c
->std_input
== EXEC_INPUT_FILE
)
5353 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5354 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5355 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5356 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5357 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5358 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5359 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5360 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5361 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5367 "%sTTYVHangup: %s\n"
5368 "%sTTYVTDisallocate: %s\n",
5369 prefix
, c
->tty_path
,
5370 prefix
, yes_no(c
->tty_reset
),
5371 prefix
, yes_no(c
->tty_vhangup
),
5372 prefix
, yes_no(c
->tty_vt_disallocate
));
5374 if (IN_SET(c
->std_output
,
5376 EXEC_OUTPUT_JOURNAL
,
5377 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5378 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5379 IN_SET(c
->std_error
,
5381 EXEC_OUTPUT_JOURNAL
,
5382 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5383 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5385 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5387 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5389 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5391 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5393 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5396 if (c
->log_level_max
>= 0) {
5397 _cleanup_free_
char *t
= NULL
;
5399 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5401 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5404 if (c
->log_ratelimit_interval_usec
> 0) {
5405 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
5408 "%sLogRateLimitIntervalSec: %s\n",
5409 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
5412 if (c
->log_ratelimit_burst
> 0)
5413 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
5415 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
5416 fprintf(f
, "%sLogExtraFields: ", prefix
);
5417 fwrite(c
->log_extra_fields
[j
].iov_base
,
5418 1, c
->log_extra_fields
[j
].iov_len
,
5423 if (c
->log_namespace
)
5424 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
5426 if (c
->secure_bits
) {
5427 _cleanup_free_
char *str
= NULL
;
5429 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
5431 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
5434 if (c
->capability_bounding_set
!= CAP_ALL
) {
5435 _cleanup_free_
char *str
= NULL
;
5437 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
5439 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
5442 if (c
->capability_ambient_set
!= 0) {
5443 _cleanup_free_
char *str
= NULL
;
5445 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
5447 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
5451 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
5453 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
5455 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
5457 if (!strv_isempty(c
->supplementary_groups
)) {
5458 fprintf(f
, "%sSupplementaryGroups:", prefix
);
5459 strv_fprintf(f
, c
->supplementary_groups
);
5464 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
5466 if (!strv_isempty(c
->read_write_paths
)) {
5467 fprintf(f
, "%sReadWritePaths:", prefix
);
5468 strv_fprintf(f
, c
->read_write_paths
);
5472 if (!strv_isempty(c
->read_only_paths
)) {
5473 fprintf(f
, "%sReadOnlyPaths:", prefix
);
5474 strv_fprintf(f
, c
->read_only_paths
);
5478 if (!strv_isempty(c
->inaccessible_paths
)) {
5479 fprintf(f
, "%sInaccessiblePaths:", prefix
);
5480 strv_fprintf(f
, c
->inaccessible_paths
);
5484 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
5485 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
5486 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
5487 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
5488 c
->bind_mounts
[i
].source
,
5489 c
->bind_mounts
[i
].destination
,
5490 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
5492 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
5493 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
5495 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
5497 isempty(t
->options
) ? "" : ":",
5498 strempty(t
->options
));
5503 "%sUtmpIdentifier: %s\n",
5504 prefix
, c
->utmp_id
);
5506 if (c
->selinux_context
)
5508 "%sSELinuxContext: %s%s\n",
5509 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
5511 if (c
->apparmor_profile
)
5513 "%sAppArmorProfile: %s%s\n",
5514 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
5516 if (c
->smack_process_label
)
5518 "%sSmackProcessLabel: %s%s\n",
5519 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
5521 if (c
->personality
!= PERSONALITY_INVALID
)
5523 "%sPersonality: %s\n",
5524 prefix
, strna(personality_to_string(c
->personality
)));
5527 "%sLockPersonality: %s\n",
5528 prefix
, yes_no(c
->lock_personality
));
5530 if (c
->syscall_filter
) {
5537 "%sSystemCallFilter: ",
5540 if (!c
->syscall_allow_list
)
5544 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
5545 _cleanup_free_
char *name
= NULL
;
5546 const char *errno_name
= NULL
;
5547 int num
= PTR_TO_INT(val
);
5554 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
5555 fputs(strna(name
), f
);
5558 errno_name
= seccomp_errno_or_action_to_string(num
);
5560 fprintf(f
, ":%s", errno_name
);
5562 fprintf(f
, ":%d", num
);
5570 if (c
->syscall_archs
) {
5576 "%sSystemCallArchitectures:",
5580 SET_FOREACH(id
, c
->syscall_archs
)
5581 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
5586 if (exec_context_restrict_namespaces_set(c
)) {
5587 _cleanup_free_
char *s
= NULL
;
5589 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
5591 fprintf(f
, "%sRestrictNamespaces: %s\n",
5595 if (c
->network_namespace_path
)
5597 "%sNetworkNamespacePath: %s\n",
5598 prefix
, c
->network_namespace_path
);
5600 if (c
->syscall_errno
> 0) {
5602 const char *errno_name
;
5605 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
5608 errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
5610 fputs(errno_name
, f
);
5612 fprintf(f
, "%d", c
->syscall_errno
);
5617 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
5620 fprintf(f
, "%sMountImages: %s%s:%s%s", prefix
,
5621 c
->mount_images
[i
].ignore_enoent
? "-": "",
5622 c
->mount_images
[i
].source
,
5623 c
->mount_images
[i
].destination
,
5624 LIST_IS_EMPTY(c
->mount_images
[i
].mount_options
) ? "": ":");
5625 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
5627 partition_designator_to_string(o
->partition_designator
),
5633 bool exec_context_maintains_privileges(const ExecContext
*c
) {
5636 /* Returns true if the process forked off would run under
5637 * an unchanged UID or as root. */
5642 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
5648 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
5656 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5658 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5663 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
5666 /* Explicit setting wins */
5667 if (c
->mount_apivfs_set
)
5668 return c
->mount_apivfs
;
5670 /* Default to "yes" if root directory or image are specified */
5671 if (exec_context_with_rootfs(c
))
5677 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5680 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
5681 free(c
->log_extra_fields
[l
].iov_base
);
5682 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5683 c
->n_log_extra_fields
= 0;
5686 void exec_context_revert_tty(ExecContext
*c
) {
5691 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5692 exec_context_tty_reset(c
, NULL
);
5694 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5695 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5696 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5698 if (exec_context_may_touch_tty(c
)) {
5701 path
= exec_context_tty_path(c
);
5703 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
5704 if (r
< 0 && r
!= -ENOENT
)
5705 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5710 int exec_context_get_clean_directories(
5716 _cleanup_strv_free_
char **l
= NULL
;
5723 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5726 if (!FLAGS_SET(mask
, 1U << t
))
5732 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5735 j
= path_join(prefix
[t
], *i
);
5739 r
= strv_consume(&l
, j
);
5743 /* Also remove private directories unconditionally. */
5744 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5745 j
= path_join(prefix
[t
], "private", *i
);
5749 r
= strv_consume(&l
, j
);
5760 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5761 ExecCleanMask mask
= 0;
5766 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5767 if (!strv_isempty(c
->directories
[t
].paths
))
5774 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5781 dual_timestamp_get(&s
->start_timestamp
);
5784 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5792 dual_timestamp_get(&s
->exit_timestamp
);
5797 if (context
&& context
->utmp_id
)
5798 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5801 void exec_status_reset(ExecStatus
*s
) {
5804 *s
= (ExecStatus
) {};
5807 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5808 char buf
[FORMAT_TIMESTAMP_MAX
];
5816 prefix
= strempty(prefix
);
5819 "%sPID: "PID_FMT
"\n",
5822 if (dual_timestamp_is_set(&s
->start_timestamp
))
5824 "%sStart Timestamp: %s\n",
5825 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5827 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5829 "%sExit Timestamp: %s\n"
5831 "%sExit Status: %i\n",
5832 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5833 prefix
, sigchld_code_to_string(s
->code
),
5837 static char *exec_command_line(char **argv
) {
5845 STRV_FOREACH(a
, argv
)
5853 STRV_FOREACH(a
, argv
) {
5860 if (strpbrk(*a
, WHITESPACE
)) {
5871 /* FIXME: this doesn't really handle arguments that have
5872 * spaces and ticks in them */
5877 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5878 _cleanup_free_
char *cmd
= NULL
;
5879 const char *prefix2
;
5884 prefix
= strempty(prefix
);
5885 prefix2
= strjoina(prefix
, "\t");
5887 cmd
= exec_command_line(c
->argv
);
5889 "%sCommand Line: %s\n",
5890 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
5892 exec_status_dump(&c
->exec_status
, f
, prefix2
);
5895 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5898 prefix
= strempty(prefix
);
5900 LIST_FOREACH(command
, c
, c
)
5901 exec_command_dump(c
, f
, prefix
);
5904 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
5911 /* It's kind of important, that we keep the order here */
5912 LIST_FIND_TAIL(command
, *l
, end
);
5913 LIST_INSERT_AFTER(command
, *l
, end
, e
);
5918 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
5926 l
= strv_new_ap(path
, ap
);
5938 free_and_replace(c
->path
, p
);
5940 return strv_free_and_replace(c
->argv
, l
);
5943 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
5944 _cleanup_strv_free_
char **l
= NULL
;
5952 l
= strv_new_ap(path
, ap
);
5958 r
= strv_extend_strv(&c
->argv
, l
, false);
5965 static void *remove_tmpdir_thread(void *p
) {
5966 _cleanup_free_
char *path
= p
;
5968 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
5972 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
5979 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
5981 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
5983 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
5984 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
5986 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
5988 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
5993 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
5994 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
5996 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
5998 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
6000 rt
->var_tmp_dir
= NULL
;
6003 rt
->id
= mfree(rt
->id
);
6004 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
6005 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
6006 safe_close_pair(rt
->netns_storage_socket
);
6010 static void exec_runtime_freep(ExecRuntime
**rt
) {
6011 (void) exec_runtime_free(*rt
, false);
6014 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
6015 _cleanup_free_
char *id_copy
= NULL
;
6020 id_copy
= strdup(id
);
6024 n
= new(ExecRuntime
, 1);
6028 *n
= (ExecRuntime
) {
6029 .id
= TAKE_PTR(id_copy
),
6030 .netns_storage_socket
= { -1, -1 },
6037 static int exec_runtime_add(
6042 int netns_storage_socket
[2],
6043 ExecRuntime
**ret
) {
6045 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6051 /* tmp_dir, var_tmp_dir, netns_storage_socket fds are donated on success */
6053 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
6057 r
= exec_runtime_allocate(&rt
, id
);
6061 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
6065 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6066 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6067 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6069 if (netns_storage_socket
) {
6070 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6071 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6078 /* do not remove created ExecRuntime object when the operation succeeds. */
6083 static int exec_runtime_make(
6085 const ExecContext
*c
,
6087 ExecRuntime
**ret
) {
6089 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6090 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
6097 /* It is not necessary to create ExecRuntime object. */
6098 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
) {
6103 if (c
->private_tmp
&&
6104 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6105 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6106 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6107 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6112 if (c
->private_network
|| c
->network_namespace_path
) {
6113 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6117 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ret
);
6124 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6132 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6134 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
6142 /* If not found, then create a new object. */
6143 r
= exec_runtime_make(m
, c
, id
, &rt
);
6147 /* When r == 0, it is not necessary to create ExecRuntime object. */
6153 /* increment reference counter. */
6159 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6163 assert(rt
->n_ref
> 0);
6169 return exec_runtime_free(rt
, destroy
);
6172 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6179 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6180 fprintf(f
, "exec-runtime=%s", rt
->id
);
6183 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6185 if (rt
->var_tmp_dir
)
6186 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6188 if (rt
->netns_storage_socket
[0] >= 0) {
6191 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6195 fprintf(f
, " netns-socket-0=%i", copy
);
6198 if (rt
->netns_storage_socket
[1] >= 0) {
6201 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6205 fprintf(f
, " netns-socket-1=%i", copy
);
6214 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6215 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6219 /* This is for the migration from old (v237 or earlier) deserialization text.
6220 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6221 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6222 * so or not from the serialized text, then we always creates a new object owned by this. */
6228 /* Manager manages ExecRuntime objects by the unit id.
6229 * So, we omit the serialized text when the unit does not have id (yet?)... */
6230 if (isempty(u
->id
)) {
6231 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6235 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
6237 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
6241 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6243 r
= exec_runtime_allocate(&rt_create
, u
->id
);
6250 if (streq(key
, "tmp-dir")) {
6253 copy
= strdup(value
);
6257 free_and_replace(rt
->tmp_dir
, copy
);
6259 } else if (streq(key
, "var-tmp-dir")) {
6262 copy
= strdup(value
);
6266 free_and_replace(rt
->var_tmp_dir
, copy
);
6268 } else if (streq(key
, "netns-socket-0")) {
6271 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6272 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6276 safe_close(rt
->netns_storage_socket
[0]);
6277 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6279 } else if (streq(key
, "netns-socket-1")) {
6282 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6283 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6287 safe_close(rt
->netns_storage_socket
[1]);
6288 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6292 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6294 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6296 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6300 rt_create
->manager
= u
->manager
;
6303 TAKE_PTR(rt_create
);
6309 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6310 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6312 int r
, fdpair
[] = {-1, -1};
6313 const char *p
, *v
= value
;
6320 n
= strcspn(v
, " ");
6321 id
= strndupa(v
, n
);
6326 v
= startswith(p
, "tmp-dir=");
6328 n
= strcspn(v
, " ");
6329 tmp_dir
= strndup(v
, n
);
6337 v
= startswith(p
, "var-tmp-dir=");
6339 n
= strcspn(v
, " ");
6340 var_tmp_dir
= strndup(v
, n
);
6348 v
= startswith(p
, "netns-socket-0=");
6352 n
= strcspn(v
, " ");
6353 buf
= strndupa(v
, n
);
6355 r
= safe_atoi(buf
, &fdpair
[0]);
6357 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6358 if (!fdset_contains(fds
, fdpair
[0]))
6359 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6360 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", fdpair
[0]);
6361 fdpair
[0] = fdset_remove(fds
, fdpair
[0]);
6367 v
= startswith(p
, "netns-socket-1=");
6371 n
= strcspn(v
, " ");
6372 buf
= strndupa(v
, n
);
6373 r
= safe_atoi(buf
, &fdpair
[1]);
6375 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6376 if (!fdset_contains(fds
, fdpair
[1]))
6377 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6378 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", fdpair
[1]);
6379 fdpair
[1] = fdset_remove(fds
, fdpair
[1]);
6383 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, fdpair
, NULL
);
6385 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
6389 void exec_runtime_vacuum(Manager
*m
) {
6394 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
6396 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6400 (void) exec_runtime_free(rt
, false);
6404 void exec_params_clear(ExecParameters
*p
) {
6408 p
->environment
= strv_free(p
->environment
);
6409 p
->fd_names
= strv_free(p
->fd_names
);
6410 p
->fds
= mfree(p
->fds
);
6411 p
->exec_fd
= safe_close(p
->exec_fd
);
6414 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
6423 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
6425 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
6426 [EXEC_INPUT_NULL
] = "null",
6427 [EXEC_INPUT_TTY
] = "tty",
6428 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
6429 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
6430 [EXEC_INPUT_SOCKET
] = "socket",
6431 [EXEC_INPUT_NAMED_FD
] = "fd",
6432 [EXEC_INPUT_DATA
] = "data",
6433 [EXEC_INPUT_FILE
] = "file",
6436 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
6438 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
6439 [EXEC_OUTPUT_INHERIT
] = "inherit",
6440 [EXEC_OUTPUT_NULL
] = "null",
6441 [EXEC_OUTPUT_TTY
] = "tty",
6442 [EXEC_OUTPUT_KMSG
] = "kmsg",
6443 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
6444 [EXEC_OUTPUT_JOURNAL
] = "journal",
6445 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
6446 [EXEC_OUTPUT_SOCKET
] = "socket",
6447 [EXEC_OUTPUT_NAMED_FD
] = "fd",
6448 [EXEC_OUTPUT_FILE
] = "file",
6449 [EXEC_OUTPUT_FILE_APPEND
] = "append",
6452 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
6454 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
6455 [EXEC_UTMP_INIT
] = "init",
6456 [EXEC_UTMP_LOGIN
] = "login",
6457 [EXEC_UTMP_USER
] = "user",
6460 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
6462 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
6463 [EXEC_PRESERVE_NO
] = "no",
6464 [EXEC_PRESERVE_YES
] = "yes",
6465 [EXEC_PRESERVE_RESTART
] = "restart",
6468 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
6470 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
6471 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6472 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
6473 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
6474 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
6475 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
6476 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
6479 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
6481 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
6482 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
6483 * directories, specifically .timer units with their timestamp touch file. */
6484 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6485 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
6486 [EXEC_DIRECTORY_STATE
] = "state",
6487 [EXEC_DIRECTORY_CACHE
] = "cache",
6488 [EXEC_DIRECTORY_LOGS
] = "logs",
6489 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
6492 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
6494 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
6495 * the service payload in. */
6496 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6497 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
6498 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
6499 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
6500 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
6501 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
6504 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
6506 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
6507 [EXEC_KEYRING_INHERIT
] = "inherit",
6508 [EXEC_KEYRING_PRIVATE
] = "private",
6509 [EXEC_KEYRING_SHARED
] = "shared",
6512 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);