1 /* SPDX-License-Identifier: LGPL-2.1+ */
6 #include <sys/eventfd.h>
9 #include <sys/personality.h>
10 #include <sys/prctl.h>
12 #include <sys/types.h>
18 #include <security/pam_appl.h>
22 #include <selinux/selinux.h>
30 #include <sys/apparmor.h>
33 #include "sd-messages.h"
36 #include "alloc-util.h"
38 #include "apparmor-util.h"
43 #include "capability-util.h"
44 #include "chown-recursive.h"
45 #include "cgroup-setup.h"
46 #include "cpu-set-util.h"
50 #include "errno-list.h"
52 #include "exit-status.h"
54 #include "format-util.h"
56 #include "glob-util.h"
63 #include "memory-util.h"
64 #include "missing_fs.h"
66 #include "namespace.h"
67 #include "parse-util.h"
68 #include "path-util.h"
69 #include "process-util.h"
70 #include "rlimit-util.h"
73 #include "seccomp-util.h"
75 #include "securebits-util.h"
76 #include "selinux-util.h"
77 #include "signal-util.h"
78 #include "smack-util.h"
79 #include "socket-util.h"
81 #include "stat-util.h"
82 #include "string-table.h"
83 #include "string-util.h"
85 #include "syslog-util.h"
86 #include "terminal-util.h"
87 #include "umask-util.h"
89 #include "user-util.h"
90 #include "utmp-wtmp.h"
92 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
93 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
95 #define SNDBUF_SIZE (8*1024*1024)
97 static int shift_fds(int fds
[], size_t n_fds
) {
98 int start
, restart_from
;
103 /* Modifies the fds array! (sorts it) */
113 for (i
= start
; i
< (int) n_fds
; i
++) {
116 /* Already at right index? */
120 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
127 /* Hmm, the fd we wanted isn't free? Then
128 * let's remember that and try again from here */
129 if (nfd
!= i
+3 && restart_from
< 0)
133 if (restart_from
< 0)
136 start
= restart_from
;
142 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
146 n_fds
= n_socket_fds
+ n_storage_fds
;
152 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
153 * O_NONBLOCK only applies to socket activation though. */
155 for (i
= 0; i
< n_fds
; i
++) {
157 if (i
< n_socket_fds
) {
158 r
= fd_nonblock(fds
[i
], nonblock
);
163 /* We unconditionally drop FD_CLOEXEC from the fds,
164 * since after all we want to pass these fds to our
167 r
= fd_cloexec(fds
[i
], false);
175 static const char *exec_context_tty_path(const ExecContext
*context
) {
178 if (context
->stdio_as_fds
)
181 if (context
->tty_path
)
182 return context
->tty_path
;
184 return "/dev/console";
187 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
192 path
= exec_context_tty_path(context
);
194 if (context
->tty_vhangup
) {
195 if (p
&& p
->stdin_fd
>= 0)
196 (void) terminal_vhangup_fd(p
->stdin_fd
);
198 (void) terminal_vhangup(path
);
201 if (context
->tty_reset
) {
202 if (p
&& p
->stdin_fd
>= 0)
203 (void) reset_terminal_fd(p
->stdin_fd
, true);
205 (void) reset_terminal(path
);
208 if (context
->tty_vt_disallocate
&& path
)
209 (void) vt_disallocate(path
);
212 static bool is_terminal_input(ExecInput i
) {
215 EXEC_INPUT_TTY_FORCE
,
216 EXEC_INPUT_TTY_FAIL
);
219 static bool is_terminal_output(ExecOutput o
) {
222 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
223 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
224 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
227 static bool is_syslog_output(ExecOutput o
) {
230 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
233 static bool is_kmsg_output(ExecOutput o
) {
236 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
239 static bool exec_context_needs_term(const ExecContext
*c
) {
242 /* Return true if the execution context suggests we should set $TERM to something useful. */
244 if (is_terminal_input(c
->std_input
))
247 if (is_terminal_output(c
->std_output
))
250 if (is_terminal_output(c
->std_error
))
253 return !!c
->tty_path
;
256 static int open_null_as(int flags
, int nfd
) {
261 fd
= open("/dev/null", flags
|O_NOCTTY
);
265 return move_fd(fd
, nfd
, false);
268 static int connect_journal_socket(
270 const char *log_namespace
,
274 union sockaddr_union sa
= {
275 .un
.sun_family
= AF_UNIX
,
277 uid_t olduid
= UID_INVALID
;
278 gid_t oldgid
= GID_INVALID
;
283 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
284 "/run/systemd/journal/stdout";
285 r
= sockaddr_un_set_path(&sa
.un
, j
);
289 if (gid_is_valid(gid
)) {
292 if (setegid(gid
) < 0)
296 if (uid_is_valid(uid
)) {
299 if (seteuid(uid
) < 0) {
305 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
307 /* If we fail to restore the uid or gid, things will likely
308 fail later on. This should only happen if an LSM interferes. */
310 if (uid_is_valid(uid
))
311 (void) seteuid(olduid
);
314 if (gid_is_valid(gid
))
315 (void) setegid(oldgid
);
320 static int connect_logger_as(
322 const ExecContext
*context
,
323 const ExecParameters
*params
,
330 _cleanup_close_
int fd
= -1;
335 assert(output
< _EXEC_OUTPUT_MAX
);
339 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
343 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
347 if (shutdown(fd
, SHUT_RD
) < 0)
350 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
360 context
->syslog_identifier
?: ident
,
361 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
362 context
->syslog_priority
,
363 !!context
->syslog_level_prefix
,
364 is_syslog_output(output
),
365 is_kmsg_output(output
),
366 is_terminal_output(output
)) < 0)
369 return move_fd(TAKE_FD(fd
), nfd
, false);
372 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
378 fd
= open_terminal(path
, flags
| O_NOCTTY
);
382 return move_fd(fd
, nfd
, false);
385 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
386 union sockaddr_union sa
= {};
387 _cleanup_close_
int fd
= -1;
392 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
395 fd
= open(path
, flags
|O_NOCTTY
, mode
);
399 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
401 if (strlen(path
) >= sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
404 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
406 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
410 salen
= sockaddr_un_set_path(&sa
.un
, path
);
414 if (connect(fd
, &sa
.sa
, salen
) < 0)
415 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
416 * indication that his wasn't an AF_UNIX socket after all */
418 if ((flags
& O_ACCMODE
) == O_RDONLY
)
419 r
= shutdown(fd
, SHUT_WR
);
420 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
421 r
= shutdown(fd
, SHUT_RD
);
430 static int fixup_input(
431 const ExecContext
*context
,
433 bool apply_tty_stdin
) {
439 std_input
= context
->std_input
;
441 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
442 return EXEC_INPUT_NULL
;
444 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
445 return EXEC_INPUT_NULL
;
447 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
448 return EXEC_INPUT_NULL
;
453 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
455 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
456 return EXEC_OUTPUT_INHERIT
;
461 static int setup_input(
462 const ExecContext
*context
,
463 const ExecParameters
*params
,
465 const int named_iofds
[static 3]) {
473 if (params
->stdin_fd
>= 0) {
474 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
477 /* Try to make this the controlling tty, if it is a tty, and reset it */
478 if (isatty(STDIN_FILENO
)) {
479 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
480 (void) reset_terminal_fd(STDIN_FILENO
, true);
486 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
490 case EXEC_INPUT_NULL
:
491 return open_null_as(O_RDONLY
, STDIN_FILENO
);
494 case EXEC_INPUT_TTY_FORCE
:
495 case EXEC_INPUT_TTY_FAIL
: {
498 fd
= acquire_terminal(exec_context_tty_path(context
),
499 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
500 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
501 ACQUIRE_TERMINAL_WAIT
,
506 return move_fd(fd
, STDIN_FILENO
, false);
509 case EXEC_INPUT_SOCKET
:
510 assert(socket_fd
>= 0);
512 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
514 case EXEC_INPUT_NAMED_FD
:
515 assert(named_iofds
[STDIN_FILENO
] >= 0);
517 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
518 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
520 case EXEC_INPUT_DATA
: {
523 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
527 return move_fd(fd
, STDIN_FILENO
, false);
530 case EXEC_INPUT_FILE
: {
534 assert(context
->stdio_file
[STDIN_FILENO
]);
536 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
537 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
539 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
543 return move_fd(fd
, STDIN_FILENO
, false);
547 assert_not_reached("Unknown input type");
551 static bool can_inherit_stderr_from_stdout(
552 const ExecContext
*context
,
558 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
561 if (e
== EXEC_OUTPUT_INHERIT
)
566 if (e
== EXEC_OUTPUT_NAMED_FD
)
567 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
569 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
570 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
575 static int setup_output(
577 const ExecContext
*context
,
578 const ExecParameters
*params
,
581 const int named_iofds
[static 3],
585 dev_t
*journal_stream_dev
,
586 ino_t
*journal_stream_ino
) {
596 assert(journal_stream_dev
);
597 assert(journal_stream_ino
);
599 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
601 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
604 return STDOUT_FILENO
;
607 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
608 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
611 return STDERR_FILENO
;
614 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
615 o
= fixup_output(context
->std_output
, socket_fd
);
617 if (fileno
== STDERR_FILENO
) {
619 e
= fixup_output(context
->std_error
, socket_fd
);
621 /* This expects the input and output are already set up */
623 /* Don't change the stderr file descriptor if we inherit all
624 * the way and are not on a tty */
625 if (e
== EXEC_OUTPUT_INHERIT
&&
626 o
== EXEC_OUTPUT_INHERIT
&&
627 i
== EXEC_INPUT_NULL
&&
628 !is_terminal_input(context
->std_input
) &&
632 /* Duplicate from stdout if possible */
633 if (can_inherit_stderr_from_stdout(context
, o
, e
))
634 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
638 } else if (o
== EXEC_OUTPUT_INHERIT
) {
639 /* If input got downgraded, inherit the original value */
640 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
641 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
643 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
644 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
645 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
647 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
651 /* We need to open /dev/null here anew, to get the right access mode. */
652 return open_null_as(O_WRONLY
, fileno
);
657 case EXEC_OUTPUT_NULL
:
658 return open_null_as(O_WRONLY
, fileno
);
660 case EXEC_OUTPUT_TTY
:
661 if (is_terminal_input(i
))
662 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
664 /* We don't reset the terminal if this is just about output */
665 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
667 case EXEC_OUTPUT_SYSLOG
:
668 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
669 case EXEC_OUTPUT_KMSG
:
670 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
671 case EXEC_OUTPUT_JOURNAL
:
672 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
673 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
675 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
676 r
= open_null_as(O_WRONLY
, fileno
);
680 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
681 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
682 * services to detect whether they are connected to the journal or not.
684 * If both stdout and stderr are connected to a stream then let's make sure to store the data
685 * about STDERR as that's usually the best way to do logging. */
687 if (fstat(fileno
, &st
) >= 0 &&
688 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
689 *journal_stream_dev
= st
.st_dev
;
690 *journal_stream_ino
= st
.st_ino
;
695 case EXEC_OUTPUT_SOCKET
:
696 assert(socket_fd
>= 0);
698 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
700 case EXEC_OUTPUT_NAMED_FD
:
701 assert(named_iofds
[fileno
] >= 0);
703 (void) fd_nonblock(named_iofds
[fileno
], false);
704 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
706 case EXEC_OUTPUT_FILE
:
707 case EXEC_OUTPUT_FILE_APPEND
: {
711 assert(context
->stdio_file
[fileno
]);
713 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
714 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
717 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
720 if (o
== EXEC_OUTPUT_FILE_APPEND
)
723 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
727 return move_fd(fd
, fileno
, 0);
731 assert_not_reached("Unknown error type");
735 static int chown_terminal(int fd
, uid_t uid
) {
740 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
741 if (isatty(fd
) < 1) {
742 if (IN_SET(errno
, EINVAL
, ENOTTY
))
743 return 0; /* not a tty */
748 /* This might fail. What matters are the results. */
749 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, -1);
756 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
757 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
760 assert(_saved_stdin
);
761 assert(_saved_stdout
);
763 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
767 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
768 if (saved_stdout
< 0)
771 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
775 r
= chown_terminal(fd
, getuid());
779 r
= reset_terminal_fd(fd
, true);
783 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
788 *_saved_stdin
= saved_stdin
;
789 *_saved_stdout
= saved_stdout
;
791 saved_stdin
= saved_stdout
= -1;
796 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
799 if (err
== -ETIMEDOUT
)
800 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
803 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
807 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
808 _cleanup_close_
int fd
= -1;
812 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
816 write_confirm_error_fd(err
, fd
, u
);
819 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
823 assert(saved_stdout
);
827 if (*saved_stdin
>= 0)
828 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
831 if (*saved_stdout
>= 0)
832 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
835 *saved_stdin
= safe_close(*saved_stdin
);
836 *saved_stdout
= safe_close(*saved_stdout
);
842 CONFIRM_PRETEND_FAILURE
= -1,
843 CONFIRM_PRETEND_SUCCESS
= 0,
847 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
848 int saved_stdout
= -1, saved_stdin
= -1, r
;
849 _cleanup_free_
char *e
= NULL
;
852 /* For any internal errors, assume a positive response. */
853 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
855 write_confirm_error(r
, vc
, u
);
856 return CONFIRM_EXECUTE
;
859 /* confirm_spawn might have been disabled while we were sleeping. */
860 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
865 e
= ellipsize(cmdline
, 60, 100);
873 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
875 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
882 printf("Resuming normal execution.\n");
883 manager_disable_confirm_spawn();
887 unit_dump(u
, stdout
, " ");
888 continue; /* ask again */
890 printf("Failing execution.\n");
891 r
= CONFIRM_PRETEND_FAILURE
;
894 printf(" c - continue, proceed without asking anymore\n"
895 " D - dump, show the state of the unit\n"
896 " f - fail, don't execute the command and pretend it failed\n"
898 " i - info, show a short summary of the unit\n"
899 " j - jobs, show jobs that are in progress\n"
900 " s - skip, don't execute the command and pretend it succeeded\n"
901 " y - yes, execute the command\n");
902 continue; /* ask again */
904 printf(" Description: %s\n"
907 u
->id
, u
->description
, cmdline
);
908 continue; /* ask again */
910 manager_dump_jobs(u
->manager
, stdout
, " ");
911 continue; /* ask again */
913 /* 'n' was removed in favor of 'f'. */
914 printf("Didn't understand 'n', did you mean 'f'?\n");
915 continue; /* ask again */
917 printf("Skipping execution.\n");
918 r
= CONFIRM_PRETEND_SUCCESS
;
924 assert_not_reached("Unhandled choice");
930 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
934 static int get_fixed_user(const ExecContext
*c
, const char **user
,
935 uid_t
*uid
, gid_t
*gid
,
936 const char **home
, const char **shell
) {
945 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
946 * (i.e. are "/" or "/bin/nologin"). */
949 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
957 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
967 r
= get_group_creds(&name
, gid
, 0);
975 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
976 const char *group
, gid_t gid
,
977 gid_t
**supplementary_gids
, int *ngids
) {
981 bool keep_groups
= false;
982 gid_t
*groups
= NULL
;
983 _cleanup_free_ gid_t
*l_gids
= NULL
;
988 * If user is given, then lookup GID and supplementary groups list.
989 * We avoid NSS lookups for gid=0. Also we have to initialize groups
990 * here and as early as possible so we keep the list of supplementary
991 * groups of the caller.
993 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
994 /* First step, initialize groups from /etc/groups */
995 if (initgroups(user
, gid
) < 0)
1001 if (strv_isempty(c
->supplementary_groups
))
1005 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1006 * be positive, otherwise fail.
1009 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1010 if (ngroups_max
<= 0)
1011 return errno_or_else(EOPNOTSUPP
);
1013 l_gids
= new(gid_t
, ngroups_max
);
1019 * Lookup the list of groups that the user belongs to, we
1020 * avoid NSS lookups here too for gid=0.
1023 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1028 STRV_FOREACH(i
, c
->supplementary_groups
) {
1031 if (k
>= ngroups_max
)
1035 r
= get_group_creds(&g
, l_gids
+k
, 0);
1043 * Sets ngids to zero to drop all supplementary groups, happens
1044 * when we are under root and SupplementaryGroups= is empty.
1051 /* Otherwise get the final list of supplementary groups */
1052 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1056 *supplementary_gids
= groups
;
1064 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1067 /* Handle SupplementaryGroups= if it is not empty */
1069 r
= maybe_setgroups(ngids
, supplementary_gids
);
1074 if (gid_is_valid(gid
)) {
1075 /* Then set our gids */
1076 if (setresgid(gid
, gid
, gid
) < 0)
1083 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1086 if (!uid_is_valid(uid
))
1089 /* Sets (but doesn't look up) the uid and make sure we keep the
1090 * capabilities while doing so. */
1092 if (context
->capability_ambient_set
!= 0) {
1094 /* First step: If we need to keep capabilities but
1095 * drop privileges we need to make sure we keep our
1096 * caps, while we drop privileges. */
1098 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1100 if (prctl(PR_GET_SECUREBITS
) != sb
)
1101 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1106 /* Second step: actually set the uids */
1107 if (setresuid(uid
, uid
, uid
) < 0)
1110 /* At this point we should have all necessary capabilities but
1111 are otherwise a normal user. However, the caps might got
1112 corrupted due to the setresuid() so we need clean them up
1113 later. This is done outside of this call. */
1120 static int null_conv(
1122 const struct pam_message
**msg
,
1123 struct pam_response
**resp
,
1124 void *appdata_ptr
) {
1126 /* We don't support conversations */
1128 return PAM_CONV_ERR
;
1133 static int setup_pam(
1140 const int fds
[], size_t n_fds
) {
1144 static const struct pam_conv conv
= {
1149 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1150 pam_handle_t
*handle
= NULL
;
1152 int pam_code
= PAM_SUCCESS
, r
;
1153 char **nv
, **e
= NULL
;
1154 bool close_session
= false;
1155 pid_t pam_pid
= 0, parent_pid
;
1162 /* We set up PAM in the parent process, then fork. The child
1163 * will then stay around until killed via PR_GET_PDEATHSIG or
1164 * systemd via the cgroup logic. It will then remove the PAM
1165 * session again. The parent process will exec() the actual
1166 * daemon. We do things this way to ensure that the main PID
1167 * of the daemon is the one we initially fork()ed. */
1169 r
= barrier_create(&barrier
);
1173 if (log_get_max_level() < LOG_DEBUG
)
1174 flags
|= PAM_SILENT
;
1176 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1177 if (pam_code
!= PAM_SUCCESS
) {
1183 _cleanup_free_
char *q
= NULL
;
1185 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1186 * out if that's the case, and read the TTY off it. */
1188 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1189 tty
= strjoina("/dev/", q
);
1193 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1194 if (pam_code
!= PAM_SUCCESS
)
1198 STRV_FOREACH(nv
, *env
) {
1199 pam_code
= pam_putenv(handle
, *nv
);
1200 if (pam_code
!= PAM_SUCCESS
)
1204 pam_code
= pam_acct_mgmt(handle
, flags
);
1205 if (pam_code
!= PAM_SUCCESS
)
1208 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1209 if (pam_code
!= PAM_SUCCESS
)
1210 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1212 pam_code
= pam_open_session(handle
, flags
);
1213 if (pam_code
!= PAM_SUCCESS
)
1216 close_session
= true;
1218 e
= pam_getenvlist(handle
);
1220 pam_code
= PAM_BUF_ERR
;
1224 /* Block SIGTERM, so that we know that it won't get lost in
1227 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1229 parent_pid
= getpid_cached();
1231 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1235 int sig
, ret
= EXIT_PAM
;
1237 /* The child's job is to reset the PAM session on
1239 barrier_set_role(&barrier
, BARRIER_CHILD
);
1241 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1242 * are open here that have been opened by PAM. */
1243 (void) close_many(fds
, n_fds
);
1245 /* Drop privileges - we don't need any to pam_close_session
1246 * and this will make PR_SET_PDEATHSIG work in most cases.
1247 * If this fails, ignore the error - but expect sd-pam threads
1248 * to fail to exit normally */
1250 r
= maybe_setgroups(0, NULL
);
1252 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1253 if (setresgid(gid
, gid
, gid
) < 0)
1254 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1255 if (setresuid(uid
, uid
, uid
) < 0)
1256 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1258 (void) ignore_signals(SIGPIPE
, -1);
1260 /* Wait until our parent died. This will only work if
1261 * the above setresuid() succeeds, otherwise the kernel
1262 * will not allow unprivileged parents kill their privileged
1263 * children this way. We rely on the control groups kill logic
1264 * to do the rest for us. */
1265 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1268 /* Tell the parent that our setup is done. This is especially
1269 * important regarding dropping privileges. Otherwise, unit
1270 * setup might race against our setresuid(2) call.
1272 * If the parent aborted, we'll detect this below, hence ignore
1273 * return failure here. */
1274 (void) barrier_place(&barrier
);
1276 /* Check if our parent process might already have died? */
1277 if (getppid() == parent_pid
) {
1280 assert_se(sigemptyset(&ss
) >= 0);
1281 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1284 if (sigwait(&ss
, &sig
) < 0) {
1291 assert(sig
== SIGTERM
);
1296 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1297 if (pam_code
!= PAM_SUCCESS
)
1300 /* If our parent died we'll end the session */
1301 if (getppid() != parent_pid
) {
1302 pam_code
= pam_close_session(handle
, flags
);
1303 if (pam_code
!= PAM_SUCCESS
)
1310 pam_end(handle
, pam_code
| flags
);
1314 barrier_set_role(&barrier
, BARRIER_PARENT
);
1316 /* If the child was forked off successfully it will do all the
1317 * cleanups, so forget about the handle here. */
1320 /* Unblock SIGTERM again in the parent */
1321 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1323 /* We close the log explicitly here, since the PAM modules
1324 * might have opened it, but we don't want this fd around. */
1327 /* Synchronously wait for the child to initialize. We don't care for
1328 * errors as we cannot recover. However, warn loudly if it happens. */
1329 if (!barrier_place_and_sync(&barrier
))
1330 log_error("PAM initialization failed");
1332 return strv_free_and_replace(*env
, e
);
1335 if (pam_code
!= PAM_SUCCESS
) {
1336 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1337 r
= -EPERM
; /* PAM errors do not map to errno */
1339 log_error_errno(r
, "PAM failed: %m");
1343 pam_code
= pam_close_session(handle
, flags
);
1345 pam_end(handle
, pam_code
| flags
);
1357 static void rename_process_from_path(const char *path
) {
1358 char process_name
[11];
1362 /* This resulting string must fit in 10 chars (i.e. the length
1363 * of "/sbin/init") to look pretty in /bin/ps */
1367 rename_process("(...)");
1373 /* The end of the process name is usually more
1374 * interesting, since the first bit might just be
1380 process_name
[0] = '(';
1381 memcpy(process_name
+1, p
, l
);
1382 process_name
[1+l
] = ')';
1383 process_name
[1+l
+1] = 0;
1385 rename_process(process_name
);
1388 static bool context_has_address_families(const ExecContext
*c
) {
1391 return c
->address_families_whitelist
||
1392 !set_isempty(c
->address_families
);
1395 static bool context_has_syscall_filters(const ExecContext
*c
) {
1398 return c
->syscall_whitelist
||
1399 !hashmap_isempty(c
->syscall_filter
);
1402 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1405 if (c
->no_new_privileges
)
1408 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1411 /* We need NNP if we have any form of seccomp and are unprivileged */
1412 return context_has_address_families(c
) ||
1413 c
->memory_deny_write_execute
||
1414 c
->restrict_realtime
||
1415 c
->restrict_suid_sgid
||
1416 exec_context_restrict_namespaces_set(c
) ||
1418 c
->protect_kernel_tunables
||
1419 c
->protect_kernel_modules
||
1420 c
->protect_kernel_logs
||
1421 c
->private_devices
||
1422 context_has_syscall_filters(c
) ||
1423 !set_isempty(c
->syscall_archs
) ||
1424 c
->lock_personality
||
1425 c
->protect_hostname
;
1430 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1432 if (is_seccomp_available())
1435 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1439 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1440 uint32_t negative_action
, default_action
, action
;
1446 if (!context_has_syscall_filters(c
))
1449 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1452 negative_action
= c
->syscall_errno
== 0 ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1454 if (c
->syscall_whitelist
) {
1455 default_action
= negative_action
;
1456 action
= SCMP_ACT_ALLOW
;
1458 default_action
= SCMP_ACT_ALLOW
;
1459 action
= negative_action
;
1462 if (needs_ambient_hack
) {
1463 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1468 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1471 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1475 if (set_isempty(c
->syscall_archs
))
1478 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1481 return seccomp_restrict_archs(c
->syscall_archs
);
1484 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1488 if (!context_has_address_families(c
))
1491 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1494 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1497 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1501 if (!c
->memory_deny_write_execute
)
1504 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1507 return seccomp_memory_deny_write_execute();
1510 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1514 if (!c
->restrict_realtime
)
1517 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1520 return seccomp_restrict_realtime();
1523 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1527 if (!c
->restrict_suid_sgid
)
1530 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1533 return seccomp_restrict_suid_sgid();
1536 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1540 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1541 * let's protect even those systems where this is left on in the kernel. */
1543 if (!c
->protect_kernel_tunables
)
1546 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1549 return seccomp_protect_sysctl();
1552 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1556 /* Turn off module syscalls on ProtectKernelModules=yes */
1558 if (!c
->protect_kernel_modules
)
1561 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1564 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1567 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1571 if (!c
->protect_kernel_logs
)
1574 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1577 return seccomp_protect_syslog();
1580 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1584 if (!c
->protect_clock
)
1587 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1590 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1593 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1597 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1599 if (!c
->private_devices
)
1602 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1605 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1608 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1612 if (!exec_context_restrict_namespaces_set(c
))
1615 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1618 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1621 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1622 unsigned long personality
;
1628 if (!c
->lock_personality
)
1631 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1634 personality
= c
->personality
;
1636 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1637 if (personality
== PERSONALITY_INVALID
) {
1639 r
= opinionated_personality(&personality
);
1644 return seccomp_lock_personality(personality
);
1649 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1652 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1653 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1655 if (idle_pipe
[0] >= 0) {
1658 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1660 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1663 /* Signal systemd that we are bored and want to continue. */
1664 n
= write(idle_pipe
[3], "x", 1);
1666 /* Wait for systemd to react to the signal above. */
1667 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1670 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1674 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1677 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1679 static int build_environment(
1681 const ExecContext
*c
,
1682 const ExecParameters
*p
,
1685 const char *username
,
1687 dev_t journal_stream_dev
,
1688 ino_t journal_stream_ino
,
1691 _cleanup_strv_free_
char **our_env
= NULL
;
1692 ExecDirectoryType t
;
1701 our_env
= new0(char*, 15 + _EXEC_DIRECTORY_TYPE_MAX
);
1706 _cleanup_free_
char *joined
= NULL
;
1708 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1710 our_env
[n_env
++] = x
;
1712 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1714 our_env
[n_env
++] = x
;
1716 joined
= strv_join(p
->fd_names
, ":");
1720 x
= strjoin("LISTEN_FDNAMES=", joined
);
1723 our_env
[n_env
++] = x
;
1726 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1727 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1729 our_env
[n_env
++] = x
;
1731 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1733 our_env
[n_env
++] = x
;
1736 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1737 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1738 * check the database directly. */
1739 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1740 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1743 our_env
[n_env
++] = x
;
1747 x
= strjoin("HOME=", home
);
1751 path_simplify(x
+ 5, true);
1752 our_env
[n_env
++] = x
;
1756 x
= strjoin("LOGNAME=", username
);
1759 our_env
[n_env
++] = x
;
1761 x
= strjoin("USER=", username
);
1764 our_env
[n_env
++] = x
;
1768 x
= strjoin("SHELL=", shell
);
1772 path_simplify(x
+ 6, true);
1773 our_env
[n_env
++] = x
;
1776 if (!sd_id128_is_null(u
->invocation_id
)) {
1777 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1780 our_env
[n_env
++] = x
;
1783 if (exec_context_needs_term(c
)) {
1784 const char *tty_path
, *term
= NULL
;
1786 tty_path
= exec_context_tty_path(c
);
1788 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1789 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1790 * passes to PID 1 ends up all the way in the console login shown. */
1792 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1793 term
= getenv("TERM");
1795 term
= default_term_for_tty(tty_path
);
1797 x
= strjoin("TERM=", term
);
1800 our_env
[n_env
++] = x
;
1803 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1804 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1807 our_env
[n_env
++] = x
;
1810 if (c
->log_namespace
) {
1811 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1815 our_env
[n_env
++] = x
;
1818 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1819 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1825 if (strv_isempty(c
->directories
[t
].paths
))
1828 n
= exec_directory_env_name_to_string(t
);
1832 pre
= strjoin(p
->prefix
[t
], "/");
1836 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1840 x
= strjoin(n
, "=", joined
);
1844 our_env
[n_env
++] = x
;
1847 our_env
[n_env
++] = NULL
;
1848 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1850 *ret
= TAKE_PTR(our_env
);
1855 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1856 _cleanup_strv_free_
char **pass_env
= NULL
;
1857 size_t n_env
= 0, n_bufsize
= 0;
1860 STRV_FOREACH(i
, c
->pass_environment
) {
1861 _cleanup_free_
char *x
= NULL
;
1867 x
= strjoin(*i
, "=", v
);
1871 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1874 pass_env
[n_env
++] = TAKE_PTR(x
);
1875 pass_env
[n_env
] = NULL
;
1878 *ret
= TAKE_PTR(pass_env
);
1883 static bool exec_needs_mount_namespace(
1884 const ExecContext
*context
,
1885 const ExecParameters
*params
,
1886 const ExecRuntime
*runtime
) {
1891 if (context
->root_image
)
1894 if (!strv_isempty(context
->read_write_paths
) ||
1895 !strv_isempty(context
->read_only_paths
) ||
1896 !strv_isempty(context
->inaccessible_paths
))
1899 if (context
->n_bind_mounts
> 0)
1902 if (context
->n_temporary_filesystems
> 0)
1905 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
1908 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1911 if (context
->private_devices
||
1912 context
->private_mounts
||
1913 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1914 context
->protect_home
!= PROTECT_HOME_NO
||
1915 context
->protect_kernel_tunables
||
1916 context
->protect_kernel_modules
||
1917 context
->protect_kernel_logs
||
1918 context
->protect_control_groups
)
1921 if (context
->root_directory
) {
1922 ExecDirectoryType t
;
1924 if (context
->mount_apivfs
)
1927 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1928 if (!params
->prefix
[t
])
1931 if (!strv_isempty(context
->directories
[t
].paths
))
1936 if (context
->dynamic_user
&&
1937 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1938 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1939 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1942 if (context
->log_namespace
)
1948 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
1949 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1950 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1951 _cleanup_close_
int unshare_ready_fd
= -1;
1952 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1957 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
1958 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
1959 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1960 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1961 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1962 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1963 * continues execution normally.
1964 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
1965 * does not need CAP_SETUID to write the single line mapping to itself. */
1967 /* Can only set up multiple mappings with CAP_SETUID. */
1968 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
1969 r
= asprintf(&uid_map
,
1970 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
1971 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1972 ouid
, ouid
, uid
, uid
);
1974 r
= asprintf(&uid_map
,
1975 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
1981 /* Can only set up multiple mappings with CAP_SETGID. */
1982 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
1983 r
= asprintf(&gid_map
,
1984 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
1985 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1986 ogid
, ogid
, gid
, gid
);
1988 r
= asprintf(&gid_map
,
1989 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
1995 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1997 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1998 if (unshare_ready_fd
< 0)
2001 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2003 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2006 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2010 _cleanup_close_
int fd
= -1;
2014 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2015 * here, after the parent opened its own user namespace. */
2018 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2020 /* Wait until the parent unshared the user namespace */
2021 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2026 /* Disable the setgroups() system call in the child user namespace, for good. */
2027 a
= procfs_file_alloca(ppid
, "setgroups");
2028 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2030 if (errno
!= ENOENT
) {
2035 /* If the file is missing the kernel is too old, let's continue anyway. */
2037 if (write(fd
, "deny\n", 5) < 0) {
2042 fd
= safe_close(fd
);
2045 /* First write the GID map */
2046 a
= procfs_file_alloca(ppid
, "gid_map");
2047 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2052 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2056 fd
= safe_close(fd
);
2058 /* The write the UID map */
2059 a
= procfs_file_alloca(ppid
, "uid_map");
2060 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2065 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2070 _exit(EXIT_SUCCESS
);
2073 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2074 _exit(EXIT_FAILURE
);
2077 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2079 if (unshare(CLONE_NEWUSER
) < 0)
2082 /* Let the child know that the namespace is ready now */
2083 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2086 /* Try to read an error code from the child */
2087 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2090 if (n
== sizeof(r
)) { /* an error code was sent to us */
2095 if (n
!= 0) /* on success we should have read 0 bytes */
2098 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2102 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2108 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2109 if (!context
->dynamic_user
)
2112 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2115 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2121 static int setup_exec_directory(
2122 const ExecContext
*context
,
2123 const ExecParameters
*params
,
2126 ExecDirectoryType type
,
2129 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2130 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2131 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2132 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2133 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2134 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2141 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2142 assert(exit_status
);
2144 if (!params
->prefix
[type
])
2147 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2148 if (!uid_is_valid(uid
))
2150 if (!gid_is_valid(gid
))
2154 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2155 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2157 p
= path_join(params
->prefix
[type
], *rt
);
2163 r
= mkdir_parents_label(p
, 0755);
2167 if (exec_directory_is_private(context
, type
)) {
2168 _cleanup_free_
char *private_root
= NULL
;
2170 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2171 * case we want to avoid leaving a directory around fully accessible that is owned by
2172 * a dynamic user whose UID is later on reused. To lock this down we use the same
2173 * trick used by container managers to prohibit host users to get access to files of
2174 * the same UID in containers: we place everything inside a directory that has an
2175 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2176 * for unprivileged host code. We then use fs namespacing to make this directory
2177 * permeable for the service itself.
2179 * Specifically: for a service which wants a special directory "foo/" we first create
2180 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2181 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2182 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2183 * unprivileged host users can't look into it. Inside of the namespace of the unit
2184 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2185 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2186 * for the service and making sure it only gets access to the dirs it needs but no
2187 * others. Tricky? Yes, absolutely, but it works!
2189 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2190 * to be owned by the service itself.
2192 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2193 * for sharing files or sockets with other services. */
2195 private_root
= path_join(params
->prefix
[type
], "private");
2196 if (!private_root
) {
2201 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2202 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2206 pp
= path_join(private_root
, *rt
);
2212 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2213 r
= mkdir_parents_label(pp
, 0755);
2217 if (is_dir(p
, false) > 0 &&
2218 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2220 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2221 * it over. Most likely the service has been upgraded from one that didn't use
2222 * DynamicUser=1, to one that does. */
2224 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2225 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2226 exec_directory_type_to_string(type
), p
, pp
);
2228 if (rename(p
, pp
) < 0) {
2233 /* Otherwise, create the actual directory for the service */
2235 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2236 if (r
< 0 && r
!= -EEXIST
)
2240 /* And link it up from the original place */
2241 r
= symlink_idempotent(pp
, p
, true);
2246 _cleanup_free_
char *target
= NULL
;
2248 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2249 readlink_and_make_absolute(p
, &target
) >= 0) {
2250 _cleanup_free_
char *q
= NULL
;
2252 /* This already exists and is a symlink? Interesting. Maybe it's one created
2253 * by DynamicUser=1 (see above)?
2255 * We do this for all directory types except for ConfigurationDirectory=,
2256 * since they all support the private/ symlink logic at least in some
2257 * configurations, see above. */
2259 q
= path_join(params
->prefix
[type
], "private", *rt
);
2265 if (path_equal(q
, target
)) {
2267 /* Hmm, apparently DynamicUser= was once turned on for this service,
2268 * but is no longer. Let's move the directory back up. */
2270 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2271 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2272 exec_directory_type_to_string(type
), q
, p
);
2274 if (unlink(p
) < 0) {
2279 if (rename(q
, p
) < 0) {
2286 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2291 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2294 /* Don't change the owner/access mode of the configuration directory,
2295 * as in the common case it is not written to by a service, and shall
2296 * not be writable. */
2298 if (stat(p
, &st
) < 0) {
2303 /* Still complain if the access mode doesn't match */
2304 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2305 log_warning("%s \'%s\' already exists but the mode is different. "
2306 "(File system: %o %sMode: %o)",
2307 exec_directory_type_to_string(type
), *rt
,
2308 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2315 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2316 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2317 * current UID/GID ownership.) */
2318 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2322 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2323 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2324 * assignments to exist.*/
2325 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2333 *exit_status
= exit_status_table
[type
];
2338 static int setup_smack(
2339 const ExecContext
*context
,
2340 const ExecCommand
*command
) {
2347 if (context
->smack_process_label
) {
2348 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2352 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2354 _cleanup_free_
char *exec_label
= NULL
;
2356 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2357 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2360 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2370 static int compile_bind_mounts(
2371 const ExecContext
*context
,
2372 const ExecParameters
*params
,
2373 BindMount
**ret_bind_mounts
,
2374 size_t *ret_n_bind_mounts
,
2375 char ***ret_empty_directories
) {
2377 _cleanup_strv_free_
char **empty_directories
= NULL
;
2378 BindMount
*bind_mounts
;
2380 ExecDirectoryType t
;
2385 assert(ret_bind_mounts
);
2386 assert(ret_n_bind_mounts
);
2387 assert(ret_empty_directories
);
2389 n
= context
->n_bind_mounts
;
2390 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2391 if (!params
->prefix
[t
])
2394 n
+= strv_length(context
->directories
[t
].paths
);
2398 *ret_bind_mounts
= NULL
;
2399 *ret_n_bind_mounts
= 0;
2400 *ret_empty_directories
= NULL
;
2404 bind_mounts
= new(BindMount
, n
);
2408 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2409 BindMount
*item
= context
->bind_mounts
+ i
;
2412 s
= strdup(item
->source
);
2418 d
= strdup(item
->destination
);
2425 bind_mounts
[h
++] = (BindMount
) {
2428 .read_only
= item
->read_only
,
2429 .recursive
= item
->recursive
,
2430 .ignore_enoent
= item
->ignore_enoent
,
2434 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2437 if (!params
->prefix
[t
])
2440 if (strv_isempty(context
->directories
[t
].paths
))
2443 if (exec_directory_is_private(context
, t
) &&
2444 !(context
->root_directory
|| context
->root_image
)) {
2447 /* So this is for a dynamic user, and we need to make sure the process can access its own
2448 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2449 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2451 private_root
= path_join(params
->prefix
[t
], "private");
2452 if (!private_root
) {
2457 r
= strv_consume(&empty_directories
, private_root
);
2462 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2465 if (exec_directory_is_private(context
, t
))
2466 s
= path_join(params
->prefix
[t
], "private", *suffix
);
2468 s
= path_join(params
->prefix
[t
], *suffix
);
2474 if (exec_directory_is_private(context
, t
) &&
2475 (context
->root_directory
|| context
->root_image
))
2476 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2477 * directory is not created on the root directory. So, let's bind-mount the directory
2478 * on the 'non-private' place. */
2479 d
= path_join(params
->prefix
[t
], *suffix
);
2488 bind_mounts
[h
++] = (BindMount
) {
2492 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
2494 .ignore_enoent
= false,
2501 *ret_bind_mounts
= bind_mounts
;
2502 *ret_n_bind_mounts
= n
;
2503 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2508 bind_mount_free_many(bind_mounts
, h
);
2512 static bool insist_on_sandboxing(
2513 const ExecContext
*context
,
2514 const char *root_dir
,
2515 const char *root_image
,
2516 const BindMount
*bind_mounts
,
2517 size_t n_bind_mounts
) {
2522 assert(n_bind_mounts
== 0 || bind_mounts
);
2524 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
2525 * would alter the view on the file system beyond making things read-only or invisble, i.e. would
2526 * rearrange stuff in a way we cannot ignore gracefully. */
2528 if (context
->n_temporary_filesystems
> 0)
2531 if (root_dir
|| root_image
)
2534 if (context
->dynamic_user
)
2537 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
2539 for (i
= 0; i
< n_bind_mounts
; i
++)
2540 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
2543 if (context
->log_namespace
)
2549 static int apply_mount_namespace(
2551 const ExecCommand
*command
,
2552 const ExecContext
*context
,
2553 const ExecParameters
*params
,
2554 const ExecRuntime
*runtime
,
2555 char **error_path
) {
2557 _cleanup_strv_free_
char **empty_directories
= NULL
;
2558 char *tmp
= NULL
, *var
= NULL
;
2559 const char *root_dir
= NULL
, *root_image
= NULL
;
2560 NamespaceInfo ns_info
;
2561 bool needs_sandboxing
;
2562 BindMount
*bind_mounts
= NULL
;
2563 size_t n_bind_mounts
= 0;
2568 /* The runtime struct only contains the parent of the private /tmp,
2569 * which is non-accessible to world users. Inside of it there's a /tmp
2570 * that is sticky, and that's the one we want to use here. */
2572 if (context
->private_tmp
&& runtime
) {
2573 if (runtime
->tmp_dir
)
2574 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2575 if (runtime
->var_tmp_dir
)
2576 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2579 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2580 root_image
= context
->root_image
;
2583 root_dir
= context
->root_directory
;
2586 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2590 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2591 if (needs_sandboxing
)
2592 ns_info
= (NamespaceInfo
) {
2593 .ignore_protect_paths
= false,
2594 .private_dev
= context
->private_devices
,
2595 .protect_control_groups
= context
->protect_control_groups
,
2596 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2597 .protect_kernel_modules
= context
->protect_kernel_modules
,
2598 .protect_kernel_logs
= context
->protect_kernel_logs
,
2599 .protect_hostname
= context
->protect_hostname
,
2600 .mount_apivfs
= context
->mount_apivfs
,
2601 .private_mounts
= context
->private_mounts
,
2603 else if (!context
->dynamic_user
&& root_dir
)
2605 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2606 * sandbox info, otherwise enforce it, don't ignore protected paths and
2607 * fail if we are enable to apply the sandbox inside the mount namespace.
2609 ns_info
= (NamespaceInfo
) {
2610 .ignore_protect_paths
= true,
2613 ns_info
= (NamespaceInfo
) {};
2615 if (context
->mount_flags
== MS_SHARED
)
2616 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2618 r
= setup_namespace(root_dir
, root_image
,
2619 &ns_info
, context
->read_write_paths
,
2620 needs_sandboxing
? context
->read_only_paths
: NULL
,
2621 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2625 context
->temporary_filesystems
,
2626 context
->n_temporary_filesystems
,
2629 context
->log_namespace
,
2630 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2631 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2632 context
->mount_flags
,
2633 DISSECT_IMAGE_DISCARD_ON_LOOP
|DISSECT_IMAGE_RELAX_VAR_CHECK
|DISSECT_IMAGE_FSCK
,
2636 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2637 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
2638 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2639 * completely different execution environment. */
2641 if (insist_on_sandboxing(
2643 root_dir
, root_image
,
2646 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2647 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2648 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2652 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2657 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2661 static int apply_working_directory(
2662 const ExecContext
*context
,
2663 const ExecParameters
*params
,
2670 assert(exit_status
);
2672 if (context
->working_directory_home
) {
2675 *exit_status
= EXIT_CHDIR
;
2681 } else if (context
->working_directory
)
2682 wd
= context
->working_directory
;
2686 if (params
->flags
& EXEC_APPLY_CHROOT
)
2689 d
= prefix_roota(context
->root_directory
, wd
);
2691 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2692 *exit_status
= EXIT_CHDIR
;
2699 static int apply_root_directory(
2700 const ExecContext
*context
,
2701 const ExecParameters
*params
,
2702 const bool needs_mount_ns
,
2706 assert(exit_status
);
2708 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2709 if (!needs_mount_ns
&& context
->root_directory
)
2710 if (chroot(context
->root_directory
) < 0) {
2711 *exit_status
= EXIT_CHROOT
;
2719 static int setup_keyring(
2721 const ExecContext
*context
,
2722 const ExecParameters
*p
,
2723 uid_t uid
, gid_t gid
) {
2725 key_serial_t keyring
;
2734 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2735 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2736 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2737 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2738 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2739 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2741 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2744 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2745 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2746 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2747 * & group is just as nasty as acquiring a reference to the user keyring. */
2749 saved_uid
= getuid();
2750 saved_gid
= getgid();
2752 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2753 if (setregid(gid
, -1) < 0)
2754 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2757 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2758 if (setreuid(uid
, -1) < 0) {
2759 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2764 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2765 if (keyring
== -1) {
2766 if (errno
== ENOSYS
)
2767 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2768 else if (IN_SET(errno
, EACCES
, EPERM
))
2769 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2770 else if (errno
== EDQUOT
)
2771 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2773 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2778 /* When requested link the user keyring into the session keyring. */
2779 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2781 if (keyctl(KEYCTL_LINK
,
2782 KEY_SPEC_USER_KEYRING
,
2783 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2784 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2789 /* Restore uid/gid back */
2790 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2791 if (setreuid(saved_uid
, -1) < 0) {
2792 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2797 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2798 if (setregid(saved_gid
, -1) < 0)
2799 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2802 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2803 if (!sd_id128_is_null(u
->invocation_id
)) {
2806 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2808 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2810 if (keyctl(KEYCTL_SETPERM
, key
,
2811 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2812 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2813 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2818 /* Revert back uid & gid for the the last time, and exit */
2819 /* no extra logging, as only the first already reported error matters */
2820 if (getuid() != saved_uid
)
2821 (void) setreuid(saved_uid
, -1);
2823 if (getgid() != saved_gid
)
2824 (void) setregid(saved_gid
, -1);
2829 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2835 array
[(*n
)++] = pair
[0];
2837 array
[(*n
)++] = pair
[1];
2840 static int close_remaining_fds(
2841 const ExecParameters
*params
,
2842 const ExecRuntime
*runtime
,
2843 const DynamicCreds
*dcreds
,
2847 const int *fds
, size_t n_fds
) {
2849 size_t n_dont_close
= 0;
2850 int dont_close
[n_fds
+ 12];
2854 if (params
->stdin_fd
>= 0)
2855 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2856 if (params
->stdout_fd
>= 0)
2857 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2858 if (params
->stderr_fd
>= 0)
2859 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2862 dont_close
[n_dont_close
++] = socket_fd
;
2864 dont_close
[n_dont_close
++] = exec_fd
;
2866 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2867 n_dont_close
+= n_fds
;
2871 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2875 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2877 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2880 if (user_lookup_fd
>= 0)
2881 dont_close
[n_dont_close
++] = user_lookup_fd
;
2883 return close_all_fds(dont_close
, n_dont_close
);
2886 static int send_user_lookup(
2894 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2895 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2898 if (user_lookup_fd
< 0)
2901 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2904 if (writev(user_lookup_fd
,
2906 IOVEC_INIT(&uid
, sizeof(uid
)),
2907 IOVEC_INIT(&gid
, sizeof(gid
)),
2908 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2914 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2921 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2926 if (!c
->working_directory_home
)
2929 r
= get_home_dir(buf
);
2937 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2938 _cleanup_strv_free_
char ** list
= NULL
;
2939 ExecDirectoryType t
;
2946 assert(c
->dynamic_user
);
2948 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2949 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2952 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2955 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2961 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2964 if (exec_directory_is_private(c
, t
))
2965 e
= path_join(p
->prefix
[t
], "private", *i
);
2967 e
= path_join(p
->prefix
[t
], *i
);
2971 r
= strv_consume(&list
, e
);
2977 *ret
= TAKE_PTR(list
);
2982 static char *exec_command_line(char **argv
);
2984 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2985 bool using_subcgroup
;
2991 if (!params
->cgroup_path
)
2994 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2995 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2996 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2997 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2998 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2999 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3000 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3001 * flag, which is only passed for the former statements, not for the latter. */
3003 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3004 if (using_subcgroup
)
3005 p
= path_join(params
->cgroup_path
, ".control");
3007 p
= strdup(params
->cgroup_path
);
3012 return using_subcgroup
;
3015 static int exec_child(
3017 const ExecCommand
*command
,
3018 const ExecContext
*context
,
3019 const ExecParameters
*params
,
3020 ExecRuntime
*runtime
,
3021 DynamicCreds
*dcreds
,
3023 const int named_iofds
[static 3],
3025 size_t n_socket_fds
,
3026 size_t n_storage_fds
,
3031 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3032 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
3033 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3034 const char *username
= NULL
, *groupname
= NULL
;
3035 _cleanup_free_
char *home_buffer
= NULL
;
3036 const char *home
= NULL
, *shell
= NULL
;
3037 char **final_argv
= NULL
;
3038 dev_t journal_stream_dev
= 0;
3039 ino_t journal_stream_ino
= 0;
3040 bool userns_set_up
= false;
3041 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3042 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3043 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3044 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3046 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3047 bool use_selinux
= false;
3050 bool use_smack
= false;
3053 bool use_apparmor
= false;
3055 uid_t saved_uid
= getuid();
3056 gid_t saved_gid
= getgid();
3057 uid_t uid
= UID_INVALID
;
3058 gid_t gid
= GID_INVALID
;
3060 ExecDirectoryType dt
;
3062 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3063 int ngids_after_pam
= 0;
3069 assert(exit_status
);
3071 rename_process_from_path(command
->path
);
3073 /* We reset exactly these signals, since they are the
3074 * only ones we set to SIG_IGN in the main daemon. All
3075 * others we leave untouched because we set them to
3076 * SIG_DFL or a valid handler initially, both of which
3077 * will be demoted to SIG_DFL. */
3078 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3079 SIGNALS_IGNORE
, -1);
3081 if (context
->ignore_sigpipe
)
3082 (void) ignore_signals(SIGPIPE
, -1);
3084 r
= reset_signal_mask();
3086 *exit_status
= EXIT_SIGNAL_MASK
;
3087 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3090 if (params
->idle_pipe
)
3091 do_idle_pipe_dance(params
->idle_pipe
);
3093 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3094 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3095 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3096 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3099 log_set_open_when_needed(true);
3101 /* In case anything used libc syslog(), close this here, too */
3104 n_fds
= n_socket_fds
+ n_storage_fds
;
3105 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
3107 *exit_status
= EXIT_FDS
;
3108 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3111 if (!context
->same_pgrp
)
3113 *exit_status
= EXIT_SETSID
;
3114 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3117 exec_context_tty_reset(context
, params
);
3119 if (unit_shall_confirm_spawn(unit
)) {
3120 const char *vc
= params
->confirm_spawn
;
3121 _cleanup_free_
char *cmdline
= NULL
;
3123 cmdline
= exec_command_line(command
->argv
);
3125 *exit_status
= EXIT_MEMORY
;
3129 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3130 if (r
!= CONFIRM_EXECUTE
) {
3131 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3132 *exit_status
= EXIT_SUCCESS
;
3135 *exit_status
= EXIT_CONFIRM
;
3136 log_unit_error(unit
, "Execution cancelled by the user");
3141 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3142 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3143 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3144 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3145 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3146 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3147 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3148 *exit_status
= EXIT_MEMORY
;
3149 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3152 if (context
->dynamic_user
&& dcreds
) {
3153 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3155 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3156 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
3157 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3158 *exit_status
= EXIT_USER
;
3159 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3162 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3164 *exit_status
= EXIT_MEMORY
;
3168 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3170 *exit_status
= EXIT_USER
;
3172 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3175 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3178 if (!uid_is_valid(uid
)) {
3179 *exit_status
= EXIT_USER
;
3180 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3184 if (!gid_is_valid(gid
)) {
3185 *exit_status
= EXIT_USER
;
3186 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3191 username
= dcreds
->user
->name
;
3194 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3196 *exit_status
= EXIT_USER
;
3197 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3200 r
= get_fixed_group(context
, &groupname
, &gid
);
3202 *exit_status
= EXIT_GROUP
;
3203 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3207 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3208 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3209 &supplementary_gids
, &ngids
);
3211 *exit_status
= EXIT_GROUP
;
3212 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3215 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3217 *exit_status
= EXIT_USER
;
3218 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3221 user_lookup_fd
= safe_close(user_lookup_fd
);
3223 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3225 *exit_status
= EXIT_CHDIR
;
3226 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3229 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3230 * must sure to drop O_NONBLOCK */
3232 (void) fd_nonblock(socket_fd
, false);
3234 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3235 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3236 if (params
->cgroup_path
) {
3237 _cleanup_free_
char *p
= NULL
;
3239 r
= exec_parameters_get_cgroup_path(params
, &p
);
3241 *exit_status
= EXIT_CGROUP
;
3242 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3245 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3247 *exit_status
= EXIT_CGROUP
;
3248 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3252 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3253 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3255 *exit_status
= EXIT_NETWORK
;
3256 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3260 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3262 *exit_status
= EXIT_STDIN
;
3263 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3266 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3268 *exit_status
= EXIT_STDOUT
;
3269 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3272 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3274 *exit_status
= EXIT_STDERR
;
3275 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3278 if (context
->oom_score_adjust_set
) {
3279 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3280 * prohibit write access to this file, and we shouldn't trip up over that. */
3281 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3282 if (IN_SET(r
, -EPERM
, -EACCES
))
3283 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3285 *exit_status
= EXIT_OOM_ADJUST
;
3286 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3290 if (context
->nice_set
) {
3291 r
= setpriority_closest(context
->nice
);
3293 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3296 if (context
->cpu_sched_set
) {
3297 struct sched_param param
= {
3298 .sched_priority
= context
->cpu_sched_priority
,
3301 r
= sched_setscheduler(0,
3302 context
->cpu_sched_policy
|
3303 (context
->cpu_sched_reset_on_fork
?
3304 SCHED_RESET_ON_FORK
: 0),
3307 *exit_status
= EXIT_SETSCHEDULER
;
3308 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3312 if (context
->cpu_set
.set
)
3313 if (sched_setaffinity(0, context
->cpu_set
.allocated
, context
->cpu_set
.set
) < 0) {
3314 *exit_status
= EXIT_CPUAFFINITY
;
3315 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3318 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
3319 r
= apply_numa_policy(&context
->numa_policy
);
3320 if (r
== -EOPNOTSUPP
)
3321 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
3323 *exit_status
= EXIT_NUMA_POLICY
;
3324 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
3328 if (context
->ioprio_set
)
3329 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3330 *exit_status
= EXIT_IOPRIO
;
3331 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3334 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3335 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3336 *exit_status
= EXIT_TIMERSLACK
;
3337 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3340 if (context
->personality
!= PERSONALITY_INVALID
) {
3341 r
= safe_personality(context
->personality
);
3343 *exit_status
= EXIT_PERSONALITY
;
3344 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3348 if (context
->utmp_id
)
3349 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3351 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3352 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3356 if (uid_is_valid(uid
)) {
3357 r
= chown_terminal(STDIN_FILENO
, uid
);
3359 *exit_status
= EXIT_STDIN
;
3360 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3364 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3365 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3366 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3367 * touch a single hierarchy too. */
3368 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3369 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3371 *exit_status
= EXIT_CGROUP
;
3372 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3376 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3377 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3379 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3382 r
= build_environment(
3394 *exit_status
= EXIT_MEMORY
;
3398 r
= build_pass_environment(context
, &pass_env
);
3400 *exit_status
= EXIT_MEMORY
;
3404 accum_env
= strv_env_merge(5,
3405 params
->environment
,
3408 context
->environment
,
3412 *exit_status
= EXIT_MEMORY
;
3415 accum_env
= strv_env_clean(accum_env
);
3417 (void) umask(context
->umask
);
3419 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3421 *exit_status
= EXIT_KEYRING
;
3422 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3425 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3426 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3428 /* 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 */
3429 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3431 /* 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 */
3432 if (needs_ambient_hack
)
3433 needs_setuid
= false;
3435 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3437 if (needs_sandboxing
) {
3438 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3439 * present. The actual MAC context application will happen later, as late as possible, to avoid
3440 * impacting our own code paths. */
3443 use_selinux
= mac_selinux_use();
3446 use_smack
= mac_smack_use();
3449 use_apparmor
= mac_apparmor_use();
3453 if (needs_sandboxing
) {
3456 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3457 * is set here. (See below.) */
3459 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3461 *exit_status
= EXIT_LIMITS
;
3462 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3468 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3469 * wins here. (See above.) */
3471 if (context
->pam_name
&& username
) {
3472 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3474 *exit_status
= EXIT_PAM
;
3475 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3478 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
3479 if (ngids_after_pam
< 0) {
3480 *exit_status
= EXIT_MEMORY
;
3481 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
3486 if (needs_sandboxing
) {
3488 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3489 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3491 *exit_status
= EXIT_SELINUX_CONTEXT
;
3492 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3497 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
3498 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
3499 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
3500 if (context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
3501 userns_set_up
= true;
3502 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
3504 *exit_status
= EXIT_USER
;
3505 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
3510 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3512 if (ns_type_supported(NAMESPACE_NET
)) {
3513 r
= setup_netns(runtime
->netns_storage_socket
);
3515 *exit_status
= EXIT_NETWORK
;
3516 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3518 } else if (context
->network_namespace_path
) {
3519 *exit_status
= EXIT_NETWORK
;
3520 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
), "NetworkNamespacePath= is not supported, refusing.");
3522 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3525 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3526 if (needs_mount_namespace
) {
3527 _cleanup_free_
char *error_path
= NULL
;
3529 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
, &error_path
);
3531 *exit_status
= EXIT_NAMESPACE
;
3532 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
3533 error_path
? ": " : "", strempty(error_path
));
3537 if (context
->protect_hostname
) {
3538 if (ns_type_supported(NAMESPACE_UTS
)) {
3539 if (unshare(CLONE_NEWUTS
) < 0) {
3540 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
3541 *exit_status
= EXIT_NAMESPACE
;
3542 return log_unit_error_errno(unit
, errno
, "Failed to set up UTS namespacing: %m");
3545 log_unit_warning(unit
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
3548 log_unit_warning(unit
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
3550 r
= seccomp_protect_hostname();
3552 *exit_status
= EXIT_SECCOMP
;
3553 return log_unit_error_errno(unit
, r
, "Failed to apply hostname restrictions: %m");
3558 /* Drop groups as early as possible.
3559 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
3560 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
3562 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
3563 int ngids_to_enforce
= 0;
3565 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
3570 if (ngids_to_enforce
< 0) {
3571 *exit_status
= EXIT_MEMORY
;
3572 return log_unit_error_errno(unit
,
3574 "Failed to merge group lists. Group membership might be incorrect: %m");
3577 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
3579 *exit_status
= EXIT_GROUP
;
3580 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3584 /* If the user namespace was not set up above, try to do it now.
3585 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
3586 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
3587 * case of mount namespaces being less privileged when the mount point list is copied from a
3588 * different user namespace). */
3590 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
3591 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
3593 *exit_status
= EXIT_USER
;
3594 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3598 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3599 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3600 * however if we have it as we want to keep it open until the final execve(). */
3602 if (params
->exec_fd
>= 0) {
3603 exec_fd
= params
->exec_fd
;
3605 if (exec_fd
< 3 + (int) n_fds
) {
3608 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3609 * process we are about to execute. */
3611 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3613 *exit_status
= EXIT_FDS
;
3614 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3617 safe_close(exec_fd
);
3620 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3621 r
= fd_cloexec(exec_fd
, true);
3623 *exit_status
= EXIT_FDS
;
3624 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3628 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3629 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3630 fds_with_exec_fd
[n_fds
] = exec_fd
;
3631 n_fds_with_exec_fd
= n_fds
+ 1;
3633 fds_with_exec_fd
= fds
;
3634 n_fds_with_exec_fd
= n_fds
;
3637 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3639 r
= shift_fds(fds
, n_fds
);
3641 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3643 *exit_status
= EXIT_FDS
;
3644 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3647 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3648 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3649 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3652 secure_bits
= context
->secure_bits
;
3654 if (needs_sandboxing
) {
3657 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3658 * requested. (Note this is placed after the general resource limit initialization, see
3659 * above, in order to take precedence.) */
3660 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3661 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3662 *exit_status
= EXIT_LIMITS
;
3663 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3668 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3669 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3671 r
= setup_smack(context
, command
);
3673 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3674 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3679 bset
= context
->capability_bounding_set
;
3680 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3681 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3682 * instead of us doing that */
3683 if (needs_ambient_hack
)
3684 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3685 (UINT64_C(1) << CAP_SETUID
) |
3686 (UINT64_C(1) << CAP_SETGID
);
3688 if (!cap_test_all(bset
)) {
3689 r
= capability_bounding_set_drop(bset
, false);
3691 *exit_status
= EXIT_CAPABILITIES
;
3692 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3696 /* This is done before enforce_user, but ambient set
3697 * does not survive over setresuid() if keep_caps is not set. */
3698 if (!needs_ambient_hack
) {
3699 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3701 *exit_status
= EXIT_CAPABILITIES
;
3702 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3707 /* chroot to root directory first, before we lose the ability to chroot */
3708 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
3710 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
3713 if (uid_is_valid(uid
)) {
3714 r
= enforce_user(context
, uid
);
3716 *exit_status
= EXIT_USER
;
3717 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3720 if (!needs_ambient_hack
&&
3721 context
->capability_ambient_set
!= 0) {
3723 /* Fix the ambient capabilities after user change. */
3724 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3726 *exit_status
= EXIT_CAPABILITIES
;
3727 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3730 /* If we were asked to change user and ambient capabilities
3731 * were requested, we had to add keep-caps to the securebits
3732 * so that we would maintain the inherited capability set
3733 * through the setresuid(). Make sure that the bit is added
3734 * also to the context secure_bits so that we don't try to
3735 * drop the bit away next. */
3737 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3742 /* Apply working directory here, because the working directory might be on NFS and only the user running
3743 * this service might have the correct privilege to change to the working directory */
3744 r
= apply_working_directory(context
, params
, home
, exit_status
);
3746 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3748 if (needs_sandboxing
) {
3749 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3750 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3751 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3752 * are restricted. */
3756 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3759 r
= setexeccon(exec_context
);
3761 *exit_status
= EXIT_SELINUX_CONTEXT
;
3762 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3769 if (use_apparmor
&& context
->apparmor_profile
) {
3770 r
= aa_change_onexec(context
->apparmor_profile
);
3771 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3772 *exit_status
= EXIT_APPARMOR_PROFILE
;
3773 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3778 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3779 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3780 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3781 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3782 *exit_status
= EXIT_SECUREBITS
;
3783 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3786 if (context_has_no_new_privileges(context
))
3787 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3788 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3789 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3793 r
= apply_address_families(unit
, context
);
3795 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3796 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3799 r
= apply_memory_deny_write_execute(unit
, context
);
3801 *exit_status
= EXIT_SECCOMP
;
3802 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3805 r
= apply_restrict_realtime(unit
, context
);
3807 *exit_status
= EXIT_SECCOMP
;
3808 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3811 r
= apply_restrict_suid_sgid(unit
, context
);
3813 *exit_status
= EXIT_SECCOMP
;
3814 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
3817 r
= apply_restrict_namespaces(unit
, context
);
3819 *exit_status
= EXIT_SECCOMP
;
3820 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3823 r
= apply_protect_sysctl(unit
, context
);
3825 *exit_status
= EXIT_SECCOMP
;
3826 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3829 r
= apply_protect_kernel_modules(unit
, context
);
3831 *exit_status
= EXIT_SECCOMP
;
3832 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3835 r
= apply_protect_kernel_logs(unit
, context
);
3837 *exit_status
= EXIT_SECCOMP
;
3838 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
3841 r
= apply_protect_clock(unit
, context
);
3843 *exit_status
= EXIT_SECCOMP
;
3844 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
3847 r
= apply_private_devices(unit
, context
);
3849 *exit_status
= EXIT_SECCOMP
;
3850 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3853 r
= apply_syscall_archs(unit
, context
);
3855 *exit_status
= EXIT_SECCOMP
;
3856 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3859 r
= apply_lock_personality(unit
, context
);
3861 *exit_status
= EXIT_SECCOMP
;
3862 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3865 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3866 * by the filter as little as possible. */
3867 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3869 *exit_status
= EXIT_SECCOMP
;
3870 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3875 if (!strv_isempty(context
->unset_environment
)) {
3878 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3880 *exit_status
= EXIT_MEMORY
;
3884 strv_free_and_replace(accum_env
, ee
);
3887 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3888 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3889 if (!replaced_argv
) {
3890 *exit_status
= EXIT_MEMORY
;
3893 final_argv
= replaced_argv
;
3895 final_argv
= command
->argv
;
3897 if (DEBUG_LOGGING
) {
3898 _cleanup_free_
char *line
;
3900 line
= exec_command_line(final_argv
);
3902 log_struct(LOG_DEBUG
,
3903 "EXECUTABLE=%s", command
->path
,
3904 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3906 LOG_UNIT_INVOCATION_ID(unit
));
3912 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3913 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3915 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3916 *exit_status
= EXIT_EXEC
;
3917 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3921 execve(command
->path
, final_argv
, accum_env
);
3927 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3928 * that POLLHUP on it no longer means execve() succeeded. */
3930 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3931 *exit_status
= EXIT_EXEC
;
3932 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3936 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3937 log_struct_errno(LOG_INFO
, r
,
3938 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3940 LOG_UNIT_INVOCATION_ID(unit
),
3941 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3943 "EXECUTABLE=%s", command
->path
);
3947 *exit_status
= EXIT_EXEC
;
3948 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3951 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3952 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
3954 int exec_spawn(Unit
*unit
,
3955 ExecCommand
*command
,
3956 const ExecContext
*context
,
3957 const ExecParameters
*params
,
3958 ExecRuntime
*runtime
,
3959 DynamicCreds
*dcreds
,
3962 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3963 _cleanup_free_
char *subcgroup_path
= NULL
;
3964 _cleanup_strv_free_
char **files_env
= NULL
;
3965 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3966 _cleanup_free_
char *line
= NULL
;
3974 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3976 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3977 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3978 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3980 if (params
->n_socket_fds
> 1) {
3981 log_unit_error(unit
, "Got more than one socket.");
3985 if (params
->n_socket_fds
== 0) {
3986 log_unit_error(unit
, "Got no socket.");
3990 socket_fd
= params
->fds
[0];
3994 n_socket_fds
= params
->n_socket_fds
;
3995 n_storage_fds
= params
->n_storage_fds
;
3998 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4000 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4002 r
= exec_context_load_environment(unit
, context
, &files_env
);
4004 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4006 line
= exec_command_line(command
->argv
);
4010 log_struct(LOG_DEBUG
,
4011 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
4012 "EXECUTABLE=%s", command
->path
,
4014 LOG_UNIT_INVOCATION_ID(unit
));
4016 if (params
->cgroup_path
) {
4017 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4019 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4020 if (r
> 0) { /* We are using a child cgroup */
4021 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4023 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4029 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4032 int exit_status
= EXIT_SUCCESS
;
4034 r
= exec_child(unit
,
4046 unit
->manager
->user_lookup_fds
[1],
4050 const char *status
=
4051 exit_status_to_string(exit_status
,
4052 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4054 log_struct_errno(LOG_ERR
, r
,
4055 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4057 LOG_UNIT_INVOCATION_ID(unit
),
4058 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4059 status
, command
->path
),
4060 "EXECUTABLE=%s", command
->path
);
4066 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4068 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4069 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4070 * process will be killed too). */
4072 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4074 exec_status_start(&command
->exec_status
, pid
);
4080 void exec_context_init(ExecContext
*c
) {
4081 ExecDirectoryType i
;
4086 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4087 c
->cpu_sched_policy
= SCHED_OTHER
;
4088 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4089 c
->syslog_level_prefix
= true;
4090 c
->ignore_sigpipe
= true;
4091 c
->timer_slack_nsec
= NSEC_INFINITY
;
4092 c
->personality
= PERSONALITY_INVALID
;
4093 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
4094 c
->directories
[i
].mode
= 0755;
4095 c
->timeout_clean_usec
= USEC_INFINITY
;
4096 c
->capability_bounding_set
= CAP_ALL
;
4097 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4098 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4099 c
->log_level_max
= -1;
4100 numa_policy_reset(&c
->numa_policy
);
4103 void exec_context_done(ExecContext
*c
) {
4104 ExecDirectoryType i
;
4109 c
->environment
= strv_free(c
->environment
);
4110 c
->environment_files
= strv_free(c
->environment_files
);
4111 c
->pass_environment
= strv_free(c
->pass_environment
);
4112 c
->unset_environment
= strv_free(c
->unset_environment
);
4114 rlimit_free_all(c
->rlimit
);
4116 for (l
= 0; l
< 3; l
++) {
4117 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4118 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4121 c
->working_directory
= mfree(c
->working_directory
);
4122 c
->root_directory
= mfree(c
->root_directory
);
4123 c
->root_image
= mfree(c
->root_image
);
4124 c
->tty_path
= mfree(c
->tty_path
);
4125 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4126 c
->user
= mfree(c
->user
);
4127 c
->group
= mfree(c
->group
);
4129 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4131 c
->pam_name
= mfree(c
->pam_name
);
4133 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4134 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4135 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4137 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4138 c
->bind_mounts
= NULL
;
4139 c
->n_bind_mounts
= 0;
4140 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4141 c
->temporary_filesystems
= NULL
;
4142 c
->n_temporary_filesystems
= 0;
4144 cpu_set_reset(&c
->cpu_set
);
4145 numa_policy_reset(&c
->numa_policy
);
4147 c
->utmp_id
= mfree(c
->utmp_id
);
4148 c
->selinux_context
= mfree(c
->selinux_context
);
4149 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4150 c
->smack_process_label
= mfree(c
->smack_process_label
);
4152 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4153 c
->syscall_archs
= set_free(c
->syscall_archs
);
4154 c
->address_families
= set_free(c
->address_families
);
4156 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
4157 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
4159 c
->log_level_max
= -1;
4161 exec_context_free_log_extra_fields(c
);
4163 c
->log_ratelimit_interval_usec
= 0;
4164 c
->log_ratelimit_burst
= 0;
4166 c
->stdin_data
= mfree(c
->stdin_data
);
4167 c
->stdin_data_size
= 0;
4169 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4171 c
->log_namespace
= mfree(c
->log_namespace
);
4174 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4179 if (!runtime_prefix
)
4182 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4183 _cleanup_free_
char *p
;
4185 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4186 p
= path_join(runtime_prefix
, "private", *i
);
4188 p
= path_join(runtime_prefix
, *i
);
4192 /* We execute this synchronously, since we need to be sure this is gone when we start the
4194 (void) rm_rf(p
, REMOVE_ROOT
);
4200 static void exec_command_done(ExecCommand
*c
) {
4203 c
->path
= mfree(c
->path
);
4204 c
->argv
= strv_free(c
->argv
);
4207 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4210 for (i
= 0; i
< n
; i
++)
4211 exec_command_done(c
+i
);
4214 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4218 LIST_REMOVE(command
, c
, i
);
4219 exec_command_done(i
);
4226 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4229 for (i
= 0; i
< n
; i
++)
4230 c
[i
] = exec_command_free_list(c
[i
]);
4233 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4236 for (i
= 0; i
< n
; i
++)
4237 exec_status_reset(&c
[i
].exec_status
);
4240 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
4243 for (i
= 0; i
< n
; i
++) {
4246 LIST_FOREACH(command
, z
, c
[i
])
4247 exec_status_reset(&z
->exec_status
);
4251 typedef struct InvalidEnvInfo
{
4256 static void invalid_env(const char *p
, void *userdata
) {
4257 InvalidEnvInfo
*info
= userdata
;
4259 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4262 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4268 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4271 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4274 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4277 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4280 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4283 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4290 static int exec_context_named_iofds(
4291 const ExecContext
*c
,
4292 const ExecParameters
*p
,
4293 int named_iofds
[static 3]) {
4296 const char* stdio_fdname
[3];
4301 assert(named_iofds
);
4303 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4304 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4305 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4307 for (i
= 0; i
< 3; i
++)
4308 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4310 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4312 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4313 if (named_iofds
[STDIN_FILENO
] < 0 &&
4314 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4315 stdio_fdname
[STDIN_FILENO
] &&
4316 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4318 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4321 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4322 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4323 stdio_fdname
[STDOUT_FILENO
] &&
4324 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4326 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4329 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4330 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4331 stdio_fdname
[STDERR_FILENO
] &&
4332 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4334 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4338 return targets
== 0 ? 0 : -ENOENT
;
4341 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4342 char **i
, **r
= NULL
;
4347 STRV_FOREACH(i
, c
->environment_files
) {
4351 bool ignore
= false;
4353 _cleanup_globfree_ glob_t pglob
= {};
4362 if (!path_is_absolute(fn
)) {
4370 /* Filename supports globbing, take all matching files */
4371 k
= safe_glob(fn
, 0, &pglob
);
4380 /* When we don't match anything, -ENOENT should be returned */
4381 assert(pglob
.gl_pathc
> 0);
4383 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4384 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4392 /* Log invalid environment variables with filename */
4394 InvalidEnvInfo info
= {
4396 .path
= pglob
.gl_pathv
[n
]
4399 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4407 m
= strv_env_merge(2, r
, p
);
4423 static bool tty_may_match_dev_console(const char *tty
) {
4424 _cleanup_free_
char *resolved
= NULL
;
4429 tty
= skip_dev_prefix(tty
);
4431 /* trivial identity? */
4432 if (streq(tty
, "console"))
4435 if (resolve_dev_console(&resolved
) < 0)
4436 return true; /* if we could not resolve, assume it may */
4438 /* "tty0" means the active VC, so it may be the same sometimes */
4439 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4442 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
4445 return ec
->tty_reset
||
4447 ec
->tty_vt_disallocate
||
4448 is_terminal_input(ec
->std_input
) ||
4449 is_terminal_output(ec
->std_output
) ||
4450 is_terminal_output(ec
->std_error
);
4453 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4455 return exec_context_may_touch_tty(ec
) &&
4456 tty_may_match_dev_console(exec_context_tty_path(ec
));
4459 static void strv_fprintf(FILE *f
, char **l
) {
4465 fprintf(f
, " %s", *g
);
4468 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4469 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
4470 ExecDirectoryType dt
;
4477 prefix
= strempty(prefix
);
4481 "%sWorkingDirectory: %s\n"
4482 "%sRootDirectory: %s\n"
4483 "%sNonBlocking: %s\n"
4484 "%sPrivateTmp: %s\n"
4485 "%sPrivateDevices: %s\n"
4486 "%sProtectKernelTunables: %s\n"
4487 "%sProtectKernelModules: %s\n"
4488 "%sProtectKernelLogs: %s\n"
4489 "%sProtectClock: %s\n"
4490 "%sProtectControlGroups: %s\n"
4491 "%sPrivateNetwork: %s\n"
4492 "%sPrivateUsers: %s\n"
4493 "%sProtectHome: %s\n"
4494 "%sProtectSystem: %s\n"
4495 "%sMountAPIVFS: %s\n"
4496 "%sIgnoreSIGPIPE: %s\n"
4497 "%sMemoryDenyWriteExecute: %s\n"
4498 "%sRestrictRealtime: %s\n"
4499 "%sRestrictSUIDSGID: %s\n"
4500 "%sKeyringMode: %s\n"
4501 "%sProtectHostname: %s\n",
4503 prefix
, c
->working_directory
? c
->working_directory
: "/",
4504 prefix
, c
->root_directory
? c
->root_directory
: "/",
4505 prefix
, yes_no(c
->non_blocking
),
4506 prefix
, yes_no(c
->private_tmp
),
4507 prefix
, yes_no(c
->private_devices
),
4508 prefix
, yes_no(c
->protect_kernel_tunables
),
4509 prefix
, yes_no(c
->protect_kernel_modules
),
4510 prefix
, yes_no(c
->protect_kernel_logs
),
4511 prefix
, yes_no(c
->protect_clock
),
4512 prefix
, yes_no(c
->protect_control_groups
),
4513 prefix
, yes_no(c
->private_network
),
4514 prefix
, yes_no(c
->private_users
),
4515 prefix
, protect_home_to_string(c
->protect_home
),
4516 prefix
, protect_system_to_string(c
->protect_system
),
4517 prefix
, yes_no(c
->mount_apivfs
),
4518 prefix
, yes_no(c
->ignore_sigpipe
),
4519 prefix
, yes_no(c
->memory_deny_write_execute
),
4520 prefix
, yes_no(c
->restrict_realtime
),
4521 prefix
, yes_no(c
->restrict_suid_sgid
),
4522 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4523 prefix
, yes_no(c
->protect_hostname
));
4526 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4528 STRV_FOREACH(e
, c
->environment
)
4529 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4531 STRV_FOREACH(e
, c
->environment_files
)
4532 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4534 STRV_FOREACH(e
, c
->pass_environment
)
4535 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4537 STRV_FOREACH(e
, c
->unset_environment
)
4538 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4540 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4542 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4543 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4545 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4546 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4550 "%sTimeoutCleanSec: %s\n",
4551 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
4558 if (c
->oom_score_adjust_set
)
4560 "%sOOMScoreAdjust: %i\n",
4561 prefix
, c
->oom_score_adjust
);
4563 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4565 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4566 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4567 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4568 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4571 if (c
->ioprio_set
) {
4572 _cleanup_free_
char *class_str
= NULL
;
4574 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4576 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4578 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4581 if (c
->cpu_sched_set
) {
4582 _cleanup_free_
char *policy_str
= NULL
;
4584 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4586 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4589 "%sCPUSchedulingPriority: %i\n"
4590 "%sCPUSchedulingResetOnFork: %s\n",
4591 prefix
, c
->cpu_sched_priority
,
4592 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4595 if (c
->cpu_set
.set
) {
4596 _cleanup_free_
char *affinity
= NULL
;
4598 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
4599 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
4602 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
4603 _cleanup_free_
char *nodes
= NULL
;
4605 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
4606 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
4607 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
4610 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4611 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4614 "%sStandardInput: %s\n"
4615 "%sStandardOutput: %s\n"
4616 "%sStandardError: %s\n",
4617 prefix
, exec_input_to_string(c
->std_input
),
4618 prefix
, exec_output_to_string(c
->std_output
),
4619 prefix
, exec_output_to_string(c
->std_error
));
4621 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4622 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4623 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4624 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4625 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4626 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4628 if (c
->std_input
== EXEC_INPUT_FILE
)
4629 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4630 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4631 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4632 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4633 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4634 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4635 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4636 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4637 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4643 "%sTTYVHangup: %s\n"
4644 "%sTTYVTDisallocate: %s\n",
4645 prefix
, c
->tty_path
,
4646 prefix
, yes_no(c
->tty_reset
),
4647 prefix
, yes_no(c
->tty_vhangup
),
4648 prefix
, yes_no(c
->tty_vt_disallocate
));
4650 if (IN_SET(c
->std_output
,
4653 EXEC_OUTPUT_JOURNAL
,
4654 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4655 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4656 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4657 IN_SET(c
->std_error
,
4660 EXEC_OUTPUT_JOURNAL
,
4661 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4662 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4663 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4665 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4667 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4669 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4671 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4673 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4676 if (c
->log_level_max
>= 0) {
4677 _cleanup_free_
char *t
= NULL
;
4679 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4681 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4684 if (c
->log_ratelimit_interval_usec
> 0) {
4685 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4688 "%sLogRateLimitIntervalSec: %s\n",
4689 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
4692 if (c
->log_ratelimit_burst
> 0)
4693 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
4695 if (c
->n_log_extra_fields
> 0) {
4698 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4699 fprintf(f
, "%sLogExtraFields: ", prefix
);
4700 fwrite(c
->log_extra_fields
[j
].iov_base
,
4701 1, c
->log_extra_fields
[j
].iov_len
,
4707 if (c
->log_namespace
)
4708 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
4710 if (c
->secure_bits
) {
4711 _cleanup_free_
char *str
= NULL
;
4713 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4715 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4718 if (c
->capability_bounding_set
!= CAP_ALL
) {
4719 _cleanup_free_
char *str
= NULL
;
4721 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4723 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4726 if (c
->capability_ambient_set
!= 0) {
4727 _cleanup_free_
char *str
= NULL
;
4729 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4731 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4735 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4737 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4739 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4741 if (!strv_isempty(c
->supplementary_groups
)) {
4742 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4743 strv_fprintf(f
, c
->supplementary_groups
);
4748 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4750 if (!strv_isempty(c
->read_write_paths
)) {
4751 fprintf(f
, "%sReadWritePaths:", prefix
);
4752 strv_fprintf(f
, c
->read_write_paths
);
4756 if (!strv_isempty(c
->read_only_paths
)) {
4757 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4758 strv_fprintf(f
, c
->read_only_paths
);
4762 if (!strv_isempty(c
->inaccessible_paths
)) {
4763 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4764 strv_fprintf(f
, c
->inaccessible_paths
);
4768 if (c
->n_bind_mounts
> 0)
4769 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4770 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4771 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4772 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4773 c
->bind_mounts
[i
].source
,
4774 c
->bind_mounts
[i
].destination
,
4775 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4777 if (c
->n_temporary_filesystems
> 0)
4778 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4779 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4781 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4783 isempty(t
->options
) ? "" : ":",
4784 strempty(t
->options
));
4789 "%sUtmpIdentifier: %s\n",
4790 prefix
, c
->utmp_id
);
4792 if (c
->selinux_context
)
4794 "%sSELinuxContext: %s%s\n",
4795 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4797 if (c
->apparmor_profile
)
4799 "%sAppArmorProfile: %s%s\n",
4800 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4802 if (c
->smack_process_label
)
4804 "%sSmackProcessLabel: %s%s\n",
4805 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4807 if (c
->personality
!= PERSONALITY_INVALID
)
4809 "%sPersonality: %s\n",
4810 prefix
, strna(personality_to_string(c
->personality
)));
4813 "%sLockPersonality: %s\n",
4814 prefix
, yes_no(c
->lock_personality
));
4816 if (c
->syscall_filter
) {
4824 "%sSystemCallFilter: ",
4827 if (!c
->syscall_whitelist
)
4831 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4832 _cleanup_free_
char *name
= NULL
;
4833 const char *errno_name
= NULL
;
4834 int num
= PTR_TO_INT(val
);
4841 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4842 fputs(strna(name
), f
);
4845 errno_name
= errno_to_name(num
);
4847 fprintf(f
, ":%s", errno_name
);
4849 fprintf(f
, ":%d", num
);
4857 if (c
->syscall_archs
) {
4864 "%sSystemCallArchitectures:",
4868 SET_FOREACH(id
, c
->syscall_archs
, j
)
4869 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4874 if (exec_context_restrict_namespaces_set(c
)) {
4875 _cleanup_free_
char *s
= NULL
;
4877 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4879 fprintf(f
, "%sRestrictNamespaces: %s\n",
4883 if (c
->network_namespace_path
)
4885 "%sNetworkNamespacePath: %s\n",
4886 prefix
, c
->network_namespace_path
);
4888 if (c
->syscall_errno
> 0) {
4889 const char *errno_name
;
4891 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4893 errno_name
= errno_to_name(c
->syscall_errno
);
4895 fprintf(f
, "%s\n", errno_name
);
4897 fprintf(f
, "%d\n", c
->syscall_errno
);
4901 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4904 /* Returns true if the process forked off would run under
4905 * an unchanged UID or as root. */
4910 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4916 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4924 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4926 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4931 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4936 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4937 free(c
->log_extra_fields
[l
].iov_base
);
4938 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4939 c
->n_log_extra_fields
= 0;
4942 void exec_context_revert_tty(ExecContext
*c
) {
4947 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
4948 exec_context_tty_reset(c
, NULL
);
4950 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
4951 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
4952 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
4954 if (exec_context_may_touch_tty(c
)) {
4957 path
= exec_context_tty_path(c
);
4959 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
4960 if (r
< 0 && r
!= -ENOENT
)
4961 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
4966 int exec_context_get_clean_directories(
4972 _cleanup_strv_free_
char **l
= NULL
;
4973 ExecDirectoryType t
;
4980 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
4983 if (!FLAGS_SET(mask
, 1U << t
))
4989 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
4992 j
= path_join(prefix
[t
], *i
);
4996 r
= strv_consume(&l
, j
);
5000 /* Also remove private directories unconditionally. */
5001 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5002 j
= path_join(prefix
[t
], "private", *i
);
5006 r
= strv_consume(&l
, j
);
5017 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5018 ExecCleanMask mask
= 0;
5023 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5024 if (!strv_isempty(c
->directories
[t
].paths
))
5031 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5038 dual_timestamp_get(&s
->start_timestamp
);
5041 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5044 if (s
->pid
!= pid
) {
5050 dual_timestamp_get(&s
->exit_timestamp
);
5055 if (context
&& context
->utmp_id
)
5056 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5059 void exec_status_reset(ExecStatus
*s
) {
5062 *s
= (ExecStatus
) {};
5065 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5066 char buf
[FORMAT_TIMESTAMP_MAX
];
5074 prefix
= strempty(prefix
);
5077 "%sPID: "PID_FMT
"\n",
5080 if (dual_timestamp_is_set(&s
->start_timestamp
))
5082 "%sStart Timestamp: %s\n",
5083 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5085 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5087 "%sExit Timestamp: %s\n"
5089 "%sExit Status: %i\n",
5090 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5091 prefix
, sigchld_code_to_string(s
->code
),
5095 static char *exec_command_line(char **argv
) {
5103 STRV_FOREACH(a
, argv
)
5111 STRV_FOREACH(a
, argv
) {
5118 if (strpbrk(*a
, WHITESPACE
)) {
5129 /* FIXME: this doesn't really handle arguments that have
5130 * spaces and ticks in them */
5135 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5136 _cleanup_free_
char *cmd
= NULL
;
5137 const char *prefix2
;
5142 prefix
= strempty(prefix
);
5143 prefix2
= strjoina(prefix
, "\t");
5145 cmd
= exec_command_line(c
->argv
);
5147 "%sCommand Line: %s\n",
5148 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
5150 exec_status_dump(&c
->exec_status
, f
, prefix2
);
5153 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5156 prefix
= strempty(prefix
);
5158 LIST_FOREACH(command
, c
, c
)
5159 exec_command_dump(c
, f
, prefix
);
5162 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
5169 /* It's kind of important, that we keep the order here */
5170 LIST_FIND_TAIL(command
, *l
, end
);
5171 LIST_INSERT_AFTER(command
, *l
, end
, e
);
5176 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
5184 l
= strv_new_ap(path
, ap
);
5196 free_and_replace(c
->path
, p
);
5198 return strv_free_and_replace(c
->argv
, l
);
5201 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
5202 _cleanup_strv_free_
char **l
= NULL
;
5210 l
= strv_new_ap(path
, ap
);
5216 r
= strv_extend_strv(&c
->argv
, l
, false);
5223 static void *remove_tmpdir_thread(void *p
) {
5224 _cleanup_free_
char *path
= p
;
5226 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
5230 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
5237 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
5239 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
5240 if (destroy
&& rt
->tmp_dir
) {
5241 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
5243 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
5245 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
5252 if (destroy
&& rt
->var_tmp_dir
) {
5253 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
5255 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
5257 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
5258 free(rt
->var_tmp_dir
);
5261 rt
->var_tmp_dir
= NULL
;
5264 rt
->id
= mfree(rt
->id
);
5265 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
5266 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
5267 safe_close_pair(rt
->netns_storage_socket
);
5271 static void exec_runtime_freep(ExecRuntime
**rt
) {
5272 (void) exec_runtime_free(*rt
, false);
5275 static int exec_runtime_allocate(ExecRuntime
**ret
) {
5280 n
= new(ExecRuntime
, 1);
5284 *n
= (ExecRuntime
) {
5285 .netns_storage_socket
= { -1, -1 },
5292 static int exec_runtime_add(
5295 const char *tmp_dir
,
5296 const char *var_tmp_dir
,
5297 const int netns_storage_socket
[2],
5298 ExecRuntime
**ret
) {
5300 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
5306 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
5310 r
= exec_runtime_allocate(&rt
);
5314 rt
->id
= strdup(id
);
5319 rt
->tmp_dir
= strdup(tmp_dir
);
5323 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
5324 assert(var_tmp_dir
);
5325 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
5326 if (!rt
->var_tmp_dir
)
5330 if (netns_storage_socket
) {
5331 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
5332 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
5335 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
5344 /* do not remove created ExecRuntime object when the operation succeeds. */
5349 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
5350 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5351 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
5358 /* It is not necessary to create ExecRuntime object. */
5359 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
5362 if (c
->private_tmp
) {
5363 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
5368 if (c
->private_network
|| c
->network_namespace_path
) {
5369 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
5373 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
5378 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
5382 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
5390 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
5392 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5398 /* If not found, then create a new object. */
5399 r
= exec_runtime_make(m
, c
, id
, &rt
);
5401 /* When r == 0, it is not necessary to create ExecRuntime object. */
5405 /* increment reference counter. */
5411 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5415 assert(rt
->n_ref
> 0);
5421 return exec_runtime_free(rt
, destroy
);
5424 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5432 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5433 fprintf(f
, "exec-runtime=%s", rt
->id
);
5436 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5438 if (rt
->var_tmp_dir
)
5439 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5441 if (rt
->netns_storage_socket
[0] >= 0) {
5444 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5448 fprintf(f
, " netns-socket-0=%i", copy
);
5451 if (rt
->netns_storage_socket
[1] >= 0) {
5454 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5458 fprintf(f
, " netns-socket-1=%i", copy
);
5467 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5468 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5472 /* This is for the migration from old (v237 or earlier) deserialization text.
5473 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5474 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5475 * so or not from the serialized text, then we always creates a new object owned by this. */
5481 /* Manager manages ExecRuntime objects by the unit id.
5482 * So, we omit the serialized text when the unit does not have id (yet?)... */
5483 if (isempty(u
->id
)) {
5484 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5488 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5490 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5494 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5496 r
= exec_runtime_allocate(&rt_create
);
5500 rt_create
->id
= strdup(u
->id
);
5507 if (streq(key
, "tmp-dir")) {
5510 copy
= strdup(value
);
5514 free_and_replace(rt
->tmp_dir
, copy
);
5516 } else if (streq(key
, "var-tmp-dir")) {
5519 copy
= strdup(value
);
5523 free_and_replace(rt
->var_tmp_dir
, copy
);
5525 } else if (streq(key
, "netns-socket-0")) {
5528 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5529 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5533 safe_close(rt
->netns_storage_socket
[0]);
5534 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5536 } else if (streq(key
, "netns-socket-1")) {
5539 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5540 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5544 safe_close(rt
->netns_storage_socket
[1]);
5545 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5549 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5551 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5553 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5557 rt_create
->manager
= u
->manager
;
5566 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5567 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5568 int r
, fd0
= -1, fd1
= -1;
5569 const char *p
, *v
= value
;
5576 n
= strcspn(v
, " ");
5577 id
= strndupa(v
, n
);
5582 v
= startswith(p
, "tmp-dir=");
5584 n
= strcspn(v
, " ");
5585 tmp_dir
= strndupa(v
, n
);
5591 v
= startswith(p
, "var-tmp-dir=");
5593 n
= strcspn(v
, " ");
5594 var_tmp_dir
= strndupa(v
, n
);
5600 v
= startswith(p
, "netns-socket-0=");
5604 n
= strcspn(v
, " ");
5605 buf
= strndupa(v
, n
);
5606 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5607 log_debug("Unable to process exec-runtime netns fd specification.");
5610 fd0
= fdset_remove(fds
, fd0
);
5616 v
= startswith(p
, "netns-socket-1=");
5620 n
= strcspn(v
, " ");
5621 buf
= strndupa(v
, n
);
5622 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5623 log_debug("Unable to process exec-runtime netns fd specification.");
5626 fd1
= fdset_remove(fds
, fd1
);
5631 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5633 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5636 void exec_runtime_vacuum(Manager
*m
) {
5642 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5644 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5648 (void) exec_runtime_free(rt
, false);
5652 void exec_params_clear(ExecParameters
*p
) {
5656 strv_free(p
->environment
);
5659 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5660 [EXEC_INPUT_NULL
] = "null",
5661 [EXEC_INPUT_TTY
] = "tty",
5662 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5663 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5664 [EXEC_INPUT_SOCKET
] = "socket",
5665 [EXEC_INPUT_NAMED_FD
] = "fd",
5666 [EXEC_INPUT_DATA
] = "data",
5667 [EXEC_INPUT_FILE
] = "file",
5670 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5672 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5673 [EXEC_OUTPUT_INHERIT
] = "inherit",
5674 [EXEC_OUTPUT_NULL
] = "null",
5675 [EXEC_OUTPUT_TTY
] = "tty",
5676 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5677 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5678 [EXEC_OUTPUT_KMSG
] = "kmsg",
5679 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5680 [EXEC_OUTPUT_JOURNAL
] = "journal",
5681 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5682 [EXEC_OUTPUT_SOCKET
] = "socket",
5683 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5684 [EXEC_OUTPUT_FILE
] = "file",
5685 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5688 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5690 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5691 [EXEC_UTMP_INIT
] = "init",
5692 [EXEC_UTMP_LOGIN
] = "login",
5693 [EXEC_UTMP_USER
] = "user",
5696 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5698 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5699 [EXEC_PRESERVE_NO
] = "no",
5700 [EXEC_PRESERVE_YES
] = "yes",
5701 [EXEC_PRESERVE_RESTART
] = "restart",
5704 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5706 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
5707 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5708 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5709 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5710 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5711 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5712 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5715 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5717 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
5718 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
5719 * directories, specifically .timer units with their timestamp touch file. */
5720 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5721 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
5722 [EXEC_DIRECTORY_STATE
] = "state",
5723 [EXEC_DIRECTORY_CACHE
] = "cache",
5724 [EXEC_DIRECTORY_LOGS
] = "logs",
5725 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
5728 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
5730 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
5731 * the service payload in. */
5732 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5733 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5734 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5735 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5736 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5737 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5740 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5742 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5743 [EXEC_KEYRING_INHERIT
] = "inherit",
5744 [EXEC_KEYRING_PRIVATE
] = "private",
5745 [EXEC_KEYRING_SHARED
] = "shared",
5748 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);