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"
57 #include "hexdecoct.h"
64 #include "memory-util.h"
65 #include "missing_fs.h"
67 #include "namespace.h"
68 #include "parse-util.h"
69 #include "path-util.h"
70 #include "process-util.h"
71 #include "rlimit-util.h"
74 #include "seccomp-util.h"
76 #include "securebits-util.h"
77 #include "selinux-util.h"
78 #include "signal-util.h"
79 #include "smack-util.h"
80 #include "socket-util.h"
82 #include "stat-util.h"
83 #include "string-table.h"
84 #include "string-util.h"
86 #include "syslog-util.h"
87 #include "terminal-util.h"
88 #include "umask-util.h"
90 #include "user-util.h"
91 #include "utmp-wtmp.h"
93 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
94 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
96 #define SNDBUF_SIZE (8*1024*1024)
98 static int shift_fds(int fds
[], size_t n_fds
) {
99 int start
, restart_from
;
104 /* Modifies the fds array! (sorts it) */
114 for (i
= start
; i
< (int) n_fds
; i
++) {
117 /* Already at right index? */
121 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
128 /* Hmm, the fd we wanted isn't free? Then
129 * let's remember that and try again from here */
130 if (nfd
!= i
+3 && restart_from
< 0)
134 if (restart_from
< 0)
137 start
= restart_from
;
143 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
147 n_fds
= n_socket_fds
+ n_storage_fds
;
153 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
154 * O_NONBLOCK only applies to socket activation though. */
156 for (i
= 0; i
< n_fds
; i
++) {
158 if (i
< n_socket_fds
) {
159 r
= fd_nonblock(fds
[i
], nonblock
);
164 /* We unconditionally drop FD_CLOEXEC from the fds,
165 * since after all we want to pass these fds to our
168 r
= fd_cloexec(fds
[i
], false);
176 static const char *exec_context_tty_path(const ExecContext
*context
) {
179 if (context
->stdio_as_fds
)
182 if (context
->tty_path
)
183 return context
->tty_path
;
185 return "/dev/console";
188 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
193 path
= exec_context_tty_path(context
);
195 if (context
->tty_vhangup
) {
196 if (p
&& p
->stdin_fd
>= 0)
197 (void) terminal_vhangup_fd(p
->stdin_fd
);
199 (void) terminal_vhangup(path
);
202 if (context
->tty_reset
) {
203 if (p
&& p
->stdin_fd
>= 0)
204 (void) reset_terminal_fd(p
->stdin_fd
, true);
206 (void) reset_terminal(path
);
209 if (context
->tty_vt_disallocate
&& path
)
210 (void) vt_disallocate(path
);
213 static bool is_terminal_input(ExecInput i
) {
216 EXEC_INPUT_TTY_FORCE
,
217 EXEC_INPUT_TTY_FAIL
);
220 static bool is_terminal_output(ExecOutput o
) {
223 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
224 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
227 static bool is_kmsg_output(ExecOutput o
) {
230 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
233 static bool exec_context_needs_term(const ExecContext
*c
) {
236 /* Return true if the execution context suggests we should set $TERM to something useful. */
238 if (is_terminal_input(c
->std_input
))
241 if (is_terminal_output(c
->std_output
))
244 if (is_terminal_output(c
->std_error
))
247 return !!c
->tty_path
;
250 static int open_null_as(int flags
, int nfd
) {
255 fd
= open("/dev/null", flags
|O_NOCTTY
);
259 return move_fd(fd
, nfd
, false);
262 static int connect_journal_socket(
264 const char *log_namespace
,
268 union sockaddr_union sa
;
270 uid_t olduid
= UID_INVALID
;
271 gid_t oldgid
= GID_INVALID
;
276 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
277 "/run/systemd/journal/stdout";
278 r
= sockaddr_un_set_path(&sa
.un
, j
);
283 if (gid_is_valid(gid
)) {
286 if (setegid(gid
) < 0)
290 if (uid_is_valid(uid
)) {
293 if (seteuid(uid
) < 0) {
299 r
= connect(fd
, &sa
.sa
, sa_len
) < 0 ? -errno
: 0;
301 /* If we fail to restore the uid or gid, things will likely
302 fail later on. This should only happen if an LSM interferes. */
304 if (uid_is_valid(uid
))
305 (void) seteuid(olduid
);
308 if (gid_is_valid(gid
))
309 (void) setegid(oldgid
);
314 static int connect_logger_as(
316 const ExecContext
*context
,
317 const ExecParameters
*params
,
324 _cleanup_close_
int fd
= -1;
329 assert(output
< _EXEC_OUTPUT_MAX
);
333 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
337 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
341 if (shutdown(fd
, SHUT_RD
) < 0)
344 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
354 context
->syslog_identifier
?: ident
,
355 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
356 context
->syslog_priority
,
357 !!context
->syslog_level_prefix
,
359 is_kmsg_output(output
),
360 is_terminal_output(output
)) < 0)
363 return move_fd(TAKE_FD(fd
), nfd
, false);
366 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
372 fd
= open_terminal(path
, flags
| O_NOCTTY
);
376 return move_fd(fd
, nfd
, false);
379 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
380 union sockaddr_union sa
;
382 _cleanup_close_
int fd
= -1;
387 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
390 fd
= open(path
, flags
|O_NOCTTY
, mode
);
394 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
397 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
399 r
= sockaddr_un_set_path(&sa
.un
, path
);
401 return r
== -EINVAL
? -ENXIO
: r
;
404 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
408 if (connect(fd
, &sa
.sa
, sa_len
) < 0)
409 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
410 * indication that his wasn't an AF_UNIX socket after all */
412 if ((flags
& O_ACCMODE
) == O_RDONLY
)
413 r
= shutdown(fd
, SHUT_WR
);
414 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
415 r
= shutdown(fd
, SHUT_RD
);
424 static int fixup_input(
425 const ExecContext
*context
,
427 bool apply_tty_stdin
) {
433 std_input
= context
->std_input
;
435 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
436 return EXEC_INPUT_NULL
;
438 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
439 return EXEC_INPUT_NULL
;
441 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
442 return EXEC_INPUT_NULL
;
447 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
449 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
450 return EXEC_OUTPUT_INHERIT
;
455 static int setup_input(
456 const ExecContext
*context
,
457 const ExecParameters
*params
,
459 const int named_iofds
[static 3]) {
467 if (params
->stdin_fd
>= 0) {
468 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
471 /* Try to make this the controlling tty, if it is a tty, and reset it */
472 if (isatty(STDIN_FILENO
)) {
473 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
474 (void) reset_terminal_fd(STDIN_FILENO
, true);
480 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
484 case EXEC_INPUT_NULL
:
485 return open_null_as(O_RDONLY
, STDIN_FILENO
);
488 case EXEC_INPUT_TTY_FORCE
:
489 case EXEC_INPUT_TTY_FAIL
: {
492 fd
= acquire_terminal(exec_context_tty_path(context
),
493 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
494 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
495 ACQUIRE_TERMINAL_WAIT
,
500 return move_fd(fd
, STDIN_FILENO
, false);
503 case EXEC_INPUT_SOCKET
:
504 assert(socket_fd
>= 0);
506 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
508 case EXEC_INPUT_NAMED_FD
:
509 assert(named_iofds
[STDIN_FILENO
] >= 0);
511 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
512 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
514 case EXEC_INPUT_DATA
: {
517 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
521 return move_fd(fd
, STDIN_FILENO
, false);
524 case EXEC_INPUT_FILE
: {
528 assert(context
->stdio_file
[STDIN_FILENO
]);
530 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
531 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
533 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
537 return move_fd(fd
, STDIN_FILENO
, false);
541 assert_not_reached("Unknown input type");
545 static bool can_inherit_stderr_from_stdout(
546 const ExecContext
*context
,
552 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
555 if (e
== EXEC_OUTPUT_INHERIT
)
560 if (e
== EXEC_OUTPUT_NAMED_FD
)
561 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
563 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
564 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
569 static int setup_output(
571 const ExecContext
*context
,
572 const ExecParameters
*params
,
575 const int named_iofds
[static 3],
579 dev_t
*journal_stream_dev
,
580 ino_t
*journal_stream_ino
) {
590 assert(journal_stream_dev
);
591 assert(journal_stream_ino
);
593 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
595 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
598 return STDOUT_FILENO
;
601 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
602 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
605 return STDERR_FILENO
;
608 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
609 o
= fixup_output(context
->std_output
, socket_fd
);
611 if (fileno
== STDERR_FILENO
) {
613 e
= fixup_output(context
->std_error
, socket_fd
);
615 /* This expects the input and output are already set up */
617 /* Don't change the stderr file descriptor if we inherit all
618 * the way and are not on a tty */
619 if (e
== EXEC_OUTPUT_INHERIT
&&
620 o
== EXEC_OUTPUT_INHERIT
&&
621 i
== EXEC_INPUT_NULL
&&
622 !is_terminal_input(context
->std_input
) &&
626 /* Duplicate from stdout if possible */
627 if (can_inherit_stderr_from_stdout(context
, o
, e
))
628 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
632 } else if (o
== EXEC_OUTPUT_INHERIT
) {
633 /* If input got downgraded, inherit the original value */
634 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
635 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
637 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
638 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
639 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
641 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
645 /* We need to open /dev/null here anew, to get the right access mode. */
646 return open_null_as(O_WRONLY
, fileno
);
651 case EXEC_OUTPUT_NULL
:
652 return open_null_as(O_WRONLY
, fileno
);
654 case EXEC_OUTPUT_TTY
:
655 if (is_terminal_input(i
))
656 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
658 /* We don't reset the terminal if this is just about output */
659 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
661 case EXEC_OUTPUT_KMSG
:
662 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
663 case EXEC_OUTPUT_JOURNAL
:
664 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
665 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
667 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
668 r
= open_null_as(O_WRONLY
, fileno
);
672 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
673 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
674 * services to detect whether they are connected to the journal or not.
676 * If both stdout and stderr are connected to a stream then let's make sure to store the data
677 * about STDERR as that's usually the best way to do logging. */
679 if (fstat(fileno
, &st
) >= 0 &&
680 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
681 *journal_stream_dev
= st
.st_dev
;
682 *journal_stream_ino
= st
.st_ino
;
687 case EXEC_OUTPUT_SOCKET
:
688 assert(socket_fd
>= 0);
690 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
692 case EXEC_OUTPUT_NAMED_FD
:
693 assert(named_iofds
[fileno
] >= 0);
695 (void) fd_nonblock(named_iofds
[fileno
], false);
696 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
698 case EXEC_OUTPUT_FILE
:
699 case EXEC_OUTPUT_FILE_APPEND
: {
703 assert(context
->stdio_file
[fileno
]);
705 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
706 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
709 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
712 if (o
== EXEC_OUTPUT_FILE_APPEND
)
715 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
719 return move_fd(fd
, fileno
, 0);
723 assert_not_reached("Unknown error type");
727 static int chown_terminal(int fd
, uid_t uid
) {
732 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
733 if (isatty(fd
) < 1) {
734 if (IN_SET(errno
, EINVAL
, ENOTTY
))
735 return 0; /* not a tty */
740 /* This might fail. What matters are the results. */
741 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, -1);
748 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
749 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
752 assert(_saved_stdin
);
753 assert(_saved_stdout
);
755 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
759 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
760 if (saved_stdout
< 0)
763 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
767 r
= chown_terminal(fd
, getuid());
771 r
= reset_terminal_fd(fd
, true);
775 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
780 *_saved_stdin
= saved_stdin
;
781 *_saved_stdout
= saved_stdout
;
783 saved_stdin
= saved_stdout
= -1;
788 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
791 if (err
== -ETIMEDOUT
)
792 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
795 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
799 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
800 _cleanup_close_
int fd
= -1;
804 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
808 write_confirm_error_fd(err
, fd
, u
);
811 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
815 assert(saved_stdout
);
819 if (*saved_stdin
>= 0)
820 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
823 if (*saved_stdout
>= 0)
824 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
827 *saved_stdin
= safe_close(*saved_stdin
);
828 *saved_stdout
= safe_close(*saved_stdout
);
834 CONFIRM_PRETEND_FAILURE
= -1,
835 CONFIRM_PRETEND_SUCCESS
= 0,
839 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
840 int saved_stdout
= -1, saved_stdin
= -1, r
;
841 _cleanup_free_
char *e
= NULL
;
844 /* For any internal errors, assume a positive response. */
845 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
847 write_confirm_error(r
, vc
, u
);
848 return CONFIRM_EXECUTE
;
851 /* confirm_spawn might have been disabled while we were sleeping. */
852 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
857 e
= ellipsize(cmdline
, 60, 100);
865 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
867 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
874 printf("Resuming normal execution.\n");
875 manager_disable_confirm_spawn();
879 unit_dump(u
, stdout
, " ");
880 continue; /* ask again */
882 printf("Failing execution.\n");
883 r
= CONFIRM_PRETEND_FAILURE
;
886 printf(" c - continue, proceed without asking anymore\n"
887 " D - dump, show the state of the unit\n"
888 " f - fail, don't execute the command and pretend it failed\n"
890 " i - info, show a short summary of the unit\n"
891 " j - jobs, show jobs that are in progress\n"
892 " s - skip, don't execute the command and pretend it succeeded\n"
893 " y - yes, execute the command\n");
894 continue; /* ask again */
896 printf(" Description: %s\n"
899 u
->id
, u
->description
, cmdline
);
900 continue; /* ask again */
902 manager_dump_jobs(u
->manager
, stdout
, " ");
903 continue; /* ask again */
905 /* 'n' was removed in favor of 'f'. */
906 printf("Didn't understand 'n', did you mean 'f'?\n");
907 continue; /* ask again */
909 printf("Skipping execution.\n");
910 r
= CONFIRM_PRETEND_SUCCESS
;
916 assert_not_reached("Unhandled choice");
922 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
926 static int get_fixed_user(const ExecContext
*c
, const char **user
,
927 uid_t
*uid
, gid_t
*gid
,
928 const char **home
, const char **shell
) {
937 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
938 * (i.e. are "/" or "/bin/nologin"). */
941 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
949 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
959 r
= get_group_creds(&name
, gid
, 0);
967 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
968 const char *group
, gid_t gid
,
969 gid_t
**supplementary_gids
, int *ngids
) {
973 bool keep_groups
= false;
974 gid_t
*groups
= NULL
;
975 _cleanup_free_ gid_t
*l_gids
= NULL
;
980 * If user is given, then lookup GID and supplementary groups list.
981 * We avoid NSS lookups for gid=0. Also we have to initialize groups
982 * here and as early as possible so we keep the list of supplementary
983 * groups of the caller.
985 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
986 /* First step, initialize groups from /etc/groups */
987 if (initgroups(user
, gid
) < 0)
993 if (strv_isempty(c
->supplementary_groups
))
997 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
998 * be positive, otherwise fail.
1001 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1002 if (ngroups_max
<= 0)
1003 return errno_or_else(EOPNOTSUPP
);
1005 l_gids
= new(gid_t
, ngroups_max
);
1011 * Lookup the list of groups that the user belongs to, we
1012 * avoid NSS lookups here too for gid=0.
1015 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1020 STRV_FOREACH(i
, c
->supplementary_groups
) {
1023 if (k
>= ngroups_max
)
1027 r
= get_group_creds(&g
, l_gids
+k
, 0);
1035 * Sets ngids to zero to drop all supplementary groups, happens
1036 * when we are under root and SupplementaryGroups= is empty.
1043 /* Otherwise get the final list of supplementary groups */
1044 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1048 *supplementary_gids
= groups
;
1056 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1059 /* Handle SupplementaryGroups= if it is not empty */
1061 r
= maybe_setgroups(ngids
, supplementary_gids
);
1066 if (gid_is_valid(gid
)) {
1067 /* Then set our gids */
1068 if (setresgid(gid
, gid
, gid
) < 0)
1075 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1078 if (!uid_is_valid(uid
))
1081 /* Sets (but doesn't look up) the uid and make sure we keep the
1082 * capabilities while doing so. */
1084 if (context
->capability_ambient_set
!= 0) {
1086 /* First step: If we need to keep capabilities but
1087 * drop privileges we need to make sure we keep our
1088 * caps, while we drop privileges. */
1090 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1092 if (prctl(PR_GET_SECUREBITS
) != sb
)
1093 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1098 /* Second step: actually set the uids */
1099 if (setresuid(uid
, uid
, uid
) < 0)
1102 /* At this point we should have all necessary capabilities but
1103 are otherwise a normal user. However, the caps might got
1104 corrupted due to the setresuid() so we need clean them up
1105 later. This is done outside of this call. */
1112 static int null_conv(
1114 const struct pam_message
**msg
,
1115 struct pam_response
**resp
,
1116 void *appdata_ptr
) {
1118 /* We don't support conversations */
1120 return PAM_CONV_ERR
;
1125 static int setup_pam(
1132 const int fds
[], size_t n_fds
) {
1136 static const struct pam_conv conv
= {
1141 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1142 pam_handle_t
*handle
= NULL
;
1144 int pam_code
= PAM_SUCCESS
, r
;
1145 char **nv
, **e
= NULL
;
1146 bool close_session
= false;
1147 pid_t pam_pid
= 0, parent_pid
;
1154 /* We set up PAM in the parent process, then fork. The child
1155 * will then stay around until killed via PR_GET_PDEATHSIG or
1156 * systemd via the cgroup logic. It will then remove the PAM
1157 * session again. The parent process will exec() the actual
1158 * daemon. We do things this way to ensure that the main PID
1159 * of the daemon is the one we initially fork()ed. */
1161 r
= barrier_create(&barrier
);
1165 if (log_get_max_level() < LOG_DEBUG
)
1166 flags
|= PAM_SILENT
;
1168 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1169 if (pam_code
!= PAM_SUCCESS
) {
1175 _cleanup_free_
char *q
= NULL
;
1177 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1178 * out if that's the case, and read the TTY off it. */
1180 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1181 tty
= strjoina("/dev/", q
);
1185 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1186 if (pam_code
!= PAM_SUCCESS
)
1190 STRV_FOREACH(nv
, *env
) {
1191 pam_code
= pam_putenv(handle
, *nv
);
1192 if (pam_code
!= PAM_SUCCESS
)
1196 pam_code
= pam_acct_mgmt(handle
, flags
);
1197 if (pam_code
!= PAM_SUCCESS
)
1200 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1201 if (pam_code
!= PAM_SUCCESS
)
1202 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1204 pam_code
= pam_open_session(handle
, flags
);
1205 if (pam_code
!= PAM_SUCCESS
)
1208 close_session
= true;
1210 e
= pam_getenvlist(handle
);
1212 pam_code
= PAM_BUF_ERR
;
1216 /* Block SIGTERM, so that we know that it won't get lost in
1219 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1221 parent_pid
= getpid_cached();
1223 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1227 int sig
, ret
= EXIT_PAM
;
1229 /* The child's job is to reset the PAM session on
1231 barrier_set_role(&barrier
, BARRIER_CHILD
);
1233 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1234 * are open here that have been opened by PAM. */
1235 (void) close_many(fds
, n_fds
);
1237 /* Drop privileges - we don't need any to pam_close_session
1238 * and this will make PR_SET_PDEATHSIG work in most cases.
1239 * If this fails, ignore the error - but expect sd-pam threads
1240 * to fail to exit normally */
1242 r
= maybe_setgroups(0, NULL
);
1244 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1245 if (setresgid(gid
, gid
, gid
) < 0)
1246 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1247 if (setresuid(uid
, uid
, uid
) < 0)
1248 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1250 (void) ignore_signals(SIGPIPE
, -1);
1252 /* Wait until our parent died. This will only work if
1253 * the above setresuid() succeeds, otherwise the kernel
1254 * will not allow unprivileged parents kill their privileged
1255 * children this way. We rely on the control groups kill logic
1256 * to do the rest for us. */
1257 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1260 /* Tell the parent that our setup is done. This is especially
1261 * important regarding dropping privileges. Otherwise, unit
1262 * setup might race against our setresuid(2) call.
1264 * If the parent aborted, we'll detect this below, hence ignore
1265 * return failure here. */
1266 (void) barrier_place(&barrier
);
1268 /* Check if our parent process might already have died? */
1269 if (getppid() == parent_pid
) {
1272 assert_se(sigemptyset(&ss
) >= 0);
1273 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1276 if (sigwait(&ss
, &sig
) < 0) {
1283 assert(sig
== SIGTERM
);
1288 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1289 if (pam_code
!= PAM_SUCCESS
)
1292 /* If our parent died we'll end the session */
1293 if (getppid() != parent_pid
) {
1294 pam_code
= pam_close_session(handle
, flags
);
1295 if (pam_code
!= PAM_SUCCESS
)
1302 pam_end(handle
, pam_code
| flags
);
1306 barrier_set_role(&barrier
, BARRIER_PARENT
);
1308 /* If the child was forked off successfully it will do all the
1309 * cleanups, so forget about the handle here. */
1312 /* Unblock SIGTERM again in the parent */
1313 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1315 /* We close the log explicitly here, since the PAM modules
1316 * might have opened it, but we don't want this fd around. */
1319 /* Synchronously wait for the child to initialize. We don't care for
1320 * errors as we cannot recover. However, warn loudly if it happens. */
1321 if (!barrier_place_and_sync(&barrier
))
1322 log_error("PAM initialization failed");
1324 return strv_free_and_replace(*env
, e
);
1327 if (pam_code
!= PAM_SUCCESS
) {
1328 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1329 r
= -EPERM
; /* PAM errors do not map to errno */
1331 log_error_errno(r
, "PAM failed: %m");
1335 pam_code
= pam_close_session(handle
, flags
);
1337 pam_end(handle
, pam_code
| flags
);
1349 static void rename_process_from_path(const char *path
) {
1350 char process_name
[11];
1354 /* This resulting string must fit in 10 chars (i.e. the length
1355 * of "/sbin/init") to look pretty in /bin/ps */
1359 rename_process("(...)");
1365 /* The end of the process name is usually more
1366 * interesting, since the first bit might just be
1372 process_name
[0] = '(';
1373 memcpy(process_name
+1, p
, l
);
1374 process_name
[1+l
] = ')';
1375 process_name
[1+l
+1] = 0;
1377 rename_process(process_name
);
1380 static bool context_has_address_families(const ExecContext
*c
) {
1383 return c
->address_families_allow_list
||
1384 !set_isempty(c
->address_families
);
1387 static bool context_has_syscall_filters(const ExecContext
*c
) {
1390 return c
->syscall_allow_list
||
1391 !hashmap_isempty(c
->syscall_filter
);
1394 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1397 if (c
->no_new_privileges
)
1400 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1403 /* We need NNP if we have any form of seccomp and are unprivileged */
1404 return context_has_address_families(c
) ||
1405 c
->memory_deny_write_execute
||
1406 c
->restrict_realtime
||
1407 c
->restrict_suid_sgid
||
1408 exec_context_restrict_namespaces_set(c
) ||
1410 c
->protect_kernel_tunables
||
1411 c
->protect_kernel_modules
||
1412 c
->protect_kernel_logs
||
1413 c
->private_devices
||
1414 context_has_syscall_filters(c
) ||
1415 !set_isempty(c
->syscall_archs
) ||
1416 c
->lock_personality
||
1417 c
->protect_hostname
;
1422 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1424 if (is_seccomp_available())
1427 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1431 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1432 uint32_t negative_action
, default_action
, action
;
1438 if (!context_has_syscall_filters(c
))
1441 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1444 negative_action
= c
->syscall_errno
== 0 ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1446 if (c
->syscall_allow_list
) {
1447 default_action
= negative_action
;
1448 action
= SCMP_ACT_ALLOW
;
1450 default_action
= SCMP_ACT_ALLOW
;
1451 action
= negative_action
;
1454 if (needs_ambient_hack
) {
1455 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1460 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1463 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1467 if (set_isempty(c
->syscall_archs
))
1470 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1473 return seccomp_restrict_archs(c
->syscall_archs
);
1476 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1480 if (!context_has_address_families(c
))
1483 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1486 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1489 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1493 if (!c
->memory_deny_write_execute
)
1496 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1499 return seccomp_memory_deny_write_execute();
1502 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1506 if (!c
->restrict_realtime
)
1509 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1512 return seccomp_restrict_realtime();
1515 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1519 if (!c
->restrict_suid_sgid
)
1522 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1525 return seccomp_restrict_suid_sgid();
1528 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1532 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1533 * let's protect even those systems where this is left on in the kernel. */
1535 if (!c
->protect_kernel_tunables
)
1538 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1541 return seccomp_protect_sysctl();
1544 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1548 /* Turn off module syscalls on ProtectKernelModules=yes */
1550 if (!c
->protect_kernel_modules
)
1553 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1556 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1559 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1563 if (!c
->protect_kernel_logs
)
1566 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1569 return seccomp_protect_syslog();
1572 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1576 if (!c
->protect_clock
)
1579 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1582 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1585 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1589 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1591 if (!c
->private_devices
)
1594 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1597 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1600 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1604 if (!exec_context_restrict_namespaces_set(c
))
1607 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1610 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1613 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1614 unsigned long personality
;
1620 if (!c
->lock_personality
)
1623 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1626 personality
= c
->personality
;
1628 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1629 if (personality
== PERSONALITY_INVALID
) {
1631 r
= opinionated_personality(&personality
);
1636 return seccomp_lock_personality(personality
);
1641 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1645 if (!c
->protect_hostname
)
1648 if (ns_type_supported(NAMESPACE_UTS
)) {
1649 if (unshare(CLONE_NEWUTS
) < 0) {
1650 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1651 *ret_exit_status
= EXIT_NAMESPACE
;
1652 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1655 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1658 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1663 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1666 r
= seccomp_protect_hostname();
1668 *ret_exit_status
= EXIT_SECCOMP
;
1669 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1676 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1679 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1680 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1682 if (idle_pipe
[0] >= 0) {
1685 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1687 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1690 /* Signal systemd that we are bored and want to continue. */
1691 n
= write(idle_pipe
[3], "x", 1);
1693 /* Wait for systemd to react to the signal above. */
1694 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1697 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1701 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1704 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1706 static int build_environment(
1708 const ExecContext
*c
,
1709 const ExecParameters
*p
,
1712 const char *username
,
1714 dev_t journal_stream_dev
,
1715 ino_t journal_stream_ino
,
1718 _cleanup_strv_free_
char **our_env
= NULL
;
1719 ExecDirectoryType t
;
1728 our_env
= new0(char*, 15 + _EXEC_DIRECTORY_TYPE_MAX
);
1733 _cleanup_free_
char *joined
= NULL
;
1735 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1737 our_env
[n_env
++] = x
;
1739 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1741 our_env
[n_env
++] = x
;
1743 joined
= strv_join(p
->fd_names
, ":");
1747 x
= strjoin("LISTEN_FDNAMES=", joined
);
1750 our_env
[n_env
++] = x
;
1753 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1754 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1756 our_env
[n_env
++] = x
;
1758 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1760 our_env
[n_env
++] = x
;
1763 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1764 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1765 * check the database directly. */
1766 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1767 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1770 our_env
[n_env
++] = x
;
1774 x
= strjoin("HOME=", home
);
1778 path_simplify(x
+ 5, true);
1779 our_env
[n_env
++] = x
;
1783 x
= strjoin("LOGNAME=", username
);
1786 our_env
[n_env
++] = x
;
1788 x
= strjoin("USER=", username
);
1791 our_env
[n_env
++] = x
;
1795 x
= strjoin("SHELL=", shell
);
1799 path_simplify(x
+ 6, true);
1800 our_env
[n_env
++] = x
;
1803 if (!sd_id128_is_null(u
->invocation_id
)) {
1804 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1807 our_env
[n_env
++] = x
;
1810 if (exec_context_needs_term(c
)) {
1811 const char *tty_path
, *term
= NULL
;
1813 tty_path
= exec_context_tty_path(c
);
1815 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1816 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1817 * container manager passes to PID 1 ends up all the way in the console login shown. */
1819 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1820 term
= getenv("TERM");
1823 term
= default_term_for_tty(tty_path
);
1825 x
= strjoin("TERM=", term
);
1828 our_env
[n_env
++] = x
;
1831 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1832 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1835 our_env
[n_env
++] = x
;
1838 if (c
->log_namespace
) {
1839 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1843 our_env
[n_env
++] = x
;
1846 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1847 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1853 if (strv_isempty(c
->directories
[t
].paths
))
1856 n
= exec_directory_env_name_to_string(t
);
1860 pre
= strjoin(p
->prefix
[t
], "/");
1864 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1868 x
= strjoin(n
, "=", joined
);
1872 our_env
[n_env
++] = x
;
1875 our_env
[n_env
++] = NULL
;
1876 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1878 *ret
= TAKE_PTR(our_env
);
1883 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1884 _cleanup_strv_free_
char **pass_env
= NULL
;
1885 size_t n_env
= 0, n_bufsize
= 0;
1888 STRV_FOREACH(i
, c
->pass_environment
) {
1889 _cleanup_free_
char *x
= NULL
;
1895 x
= strjoin(*i
, "=", v
);
1899 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1902 pass_env
[n_env
++] = TAKE_PTR(x
);
1903 pass_env
[n_env
] = NULL
;
1906 *ret
= TAKE_PTR(pass_env
);
1911 static bool exec_needs_mount_namespace(
1912 const ExecContext
*context
,
1913 const ExecParameters
*params
,
1914 const ExecRuntime
*runtime
) {
1919 if (context
->root_image
)
1922 if (!strv_isempty(context
->read_write_paths
) ||
1923 !strv_isempty(context
->read_only_paths
) ||
1924 !strv_isempty(context
->inaccessible_paths
))
1927 if (context
->n_bind_mounts
> 0)
1930 if (context
->n_temporary_filesystems
> 0)
1933 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
1936 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1939 if (context
->private_devices
||
1940 context
->private_mounts
||
1941 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1942 context
->protect_home
!= PROTECT_HOME_NO
||
1943 context
->protect_kernel_tunables
||
1944 context
->protect_kernel_modules
||
1945 context
->protect_kernel_logs
||
1946 context
->protect_control_groups
)
1949 if (context
->root_directory
) {
1950 ExecDirectoryType t
;
1952 if (context
->mount_apivfs
)
1955 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1956 if (!params
->prefix
[t
])
1959 if (!strv_isempty(context
->directories
[t
].paths
))
1964 if (context
->dynamic_user
&&
1965 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1966 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1967 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1970 if (context
->log_namespace
)
1976 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
1977 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1978 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1979 _cleanup_close_
int unshare_ready_fd
= -1;
1980 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1985 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
1986 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
1987 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1988 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1989 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1990 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1991 * continues execution normally.
1992 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
1993 * does not need CAP_SETUID to write the single line mapping to itself. */
1995 /* Can only set up multiple mappings with CAP_SETUID. */
1996 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
1997 r
= asprintf(&uid_map
,
1998 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
1999 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2000 ouid
, ouid
, uid
, uid
);
2002 r
= asprintf(&uid_map
,
2003 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2009 /* Can only set up multiple mappings with CAP_SETGID. */
2010 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2011 r
= asprintf(&gid_map
,
2012 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2013 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2014 ogid
, ogid
, gid
, gid
);
2016 r
= asprintf(&gid_map
,
2017 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2023 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2025 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2026 if (unshare_ready_fd
< 0)
2029 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2031 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2034 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2038 _cleanup_close_
int fd
= -1;
2042 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2043 * here, after the parent opened its own user namespace. */
2046 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2048 /* Wait until the parent unshared the user namespace */
2049 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2054 /* Disable the setgroups() system call in the child user namespace, for good. */
2055 a
= procfs_file_alloca(ppid
, "setgroups");
2056 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2058 if (errno
!= ENOENT
) {
2063 /* If the file is missing the kernel is too old, let's continue anyway. */
2065 if (write(fd
, "deny\n", 5) < 0) {
2070 fd
= safe_close(fd
);
2073 /* First write the GID map */
2074 a
= procfs_file_alloca(ppid
, "gid_map");
2075 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2080 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2084 fd
= safe_close(fd
);
2086 /* The write the UID map */
2087 a
= procfs_file_alloca(ppid
, "uid_map");
2088 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2093 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2098 _exit(EXIT_SUCCESS
);
2101 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2102 _exit(EXIT_FAILURE
);
2105 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2107 if (unshare(CLONE_NEWUSER
) < 0)
2110 /* Let the child know that the namespace is ready now */
2111 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2114 /* Try to read an error code from the child */
2115 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2118 if (n
== sizeof(r
)) { /* an error code was sent to us */
2123 if (n
!= 0) /* on success we should have read 0 bytes */
2126 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2130 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2136 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2137 if (!context
->dynamic_user
)
2140 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2143 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2149 static int setup_exec_directory(
2150 const ExecContext
*context
,
2151 const ExecParameters
*params
,
2154 ExecDirectoryType type
,
2157 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2158 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2159 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2160 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2161 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2162 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2169 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2170 assert(exit_status
);
2172 if (!params
->prefix
[type
])
2175 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2176 if (!uid_is_valid(uid
))
2178 if (!gid_is_valid(gid
))
2182 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2183 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2185 p
= path_join(params
->prefix
[type
], *rt
);
2191 r
= mkdir_parents_label(p
, 0755);
2195 if (exec_directory_is_private(context
, type
)) {
2196 _cleanup_free_
char *private_root
= NULL
;
2198 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2199 * case we want to avoid leaving a directory around fully accessible that is owned by
2200 * a dynamic user whose UID is later on reused. To lock this down we use the same
2201 * trick used by container managers to prohibit host users to get access to files of
2202 * the same UID in containers: we place everything inside a directory that has an
2203 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2204 * for unprivileged host code. We then use fs namespacing to make this directory
2205 * permeable for the service itself.
2207 * Specifically: for a service which wants a special directory "foo/" we first create
2208 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2209 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2210 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2211 * unprivileged host users can't look into it. Inside of the namespace of the unit
2212 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2213 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2214 * for the service and making sure it only gets access to the dirs it needs but no
2215 * others. Tricky? Yes, absolutely, but it works!
2217 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2218 * to be owned by the service itself.
2220 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2221 * for sharing files or sockets with other services. */
2223 private_root
= path_join(params
->prefix
[type
], "private");
2224 if (!private_root
) {
2229 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2230 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2234 pp
= path_join(private_root
, *rt
);
2240 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2241 r
= mkdir_parents_label(pp
, 0755);
2245 if (is_dir(p
, false) > 0 &&
2246 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2248 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2249 * it over. Most likely the service has been upgraded from one that didn't use
2250 * DynamicUser=1, to one that does. */
2252 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2253 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2254 exec_directory_type_to_string(type
), p
, pp
);
2256 if (rename(p
, pp
) < 0) {
2261 /* Otherwise, create the actual directory for the service */
2263 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2264 if (r
< 0 && r
!= -EEXIST
)
2268 /* And link it up from the original place */
2269 r
= symlink_idempotent(pp
, p
, true);
2274 _cleanup_free_
char *target
= NULL
;
2276 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2277 readlink_and_make_absolute(p
, &target
) >= 0) {
2278 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2280 /* This already exists and is a symlink? Interesting. Maybe it's one created
2281 * by DynamicUser=1 (see above)?
2283 * We do this for all directory types except for ConfigurationDirectory=,
2284 * since they all support the private/ symlink logic at least in some
2285 * configurations, see above. */
2287 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2291 q
= path_join(params
->prefix
[type
], "private", *rt
);
2297 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2298 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2302 if (path_equal(q_resolved
, target_resolved
)) {
2304 /* Hmm, apparently DynamicUser= was once turned on for this service,
2305 * but is no longer. Let's move the directory back up. */
2307 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2308 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2309 exec_directory_type_to_string(type
), q
, p
);
2311 if (unlink(p
) < 0) {
2316 if (rename(q
, p
) < 0) {
2323 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2328 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2331 /* Don't change the owner/access mode of the configuration directory,
2332 * as in the common case it is not written to by a service, and shall
2333 * not be writable. */
2335 if (stat(p
, &st
) < 0) {
2340 /* Still complain if the access mode doesn't match */
2341 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2342 log_warning("%s \'%s\' already exists but the mode is different. "
2343 "(File system: %o %sMode: %o)",
2344 exec_directory_type_to_string(type
), *rt
,
2345 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2352 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2353 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2354 * current UID/GID ownership.) */
2355 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2359 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2360 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2361 * assignments to exist.*/
2362 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2370 *exit_status
= exit_status_table
[type
];
2375 static int setup_smack(
2376 const ExecContext
*context
,
2377 const ExecCommand
*command
) {
2384 if (context
->smack_process_label
) {
2385 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2389 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2391 _cleanup_free_
char *exec_label
= NULL
;
2393 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2394 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2397 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2407 static int compile_bind_mounts(
2408 const ExecContext
*context
,
2409 const ExecParameters
*params
,
2410 BindMount
**ret_bind_mounts
,
2411 size_t *ret_n_bind_mounts
,
2412 char ***ret_empty_directories
) {
2414 _cleanup_strv_free_
char **empty_directories
= NULL
;
2415 BindMount
*bind_mounts
;
2417 ExecDirectoryType t
;
2422 assert(ret_bind_mounts
);
2423 assert(ret_n_bind_mounts
);
2424 assert(ret_empty_directories
);
2426 n
= context
->n_bind_mounts
;
2427 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2428 if (!params
->prefix
[t
])
2431 n
+= strv_length(context
->directories
[t
].paths
);
2435 *ret_bind_mounts
= NULL
;
2436 *ret_n_bind_mounts
= 0;
2437 *ret_empty_directories
= NULL
;
2441 bind_mounts
= new(BindMount
, n
);
2445 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2446 BindMount
*item
= context
->bind_mounts
+ i
;
2449 s
= strdup(item
->source
);
2455 d
= strdup(item
->destination
);
2462 bind_mounts
[h
++] = (BindMount
) {
2465 .read_only
= item
->read_only
,
2466 .recursive
= item
->recursive
,
2467 .ignore_enoent
= item
->ignore_enoent
,
2471 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2474 if (!params
->prefix
[t
])
2477 if (strv_isempty(context
->directories
[t
].paths
))
2480 if (exec_directory_is_private(context
, t
) &&
2481 !(context
->root_directory
|| context
->root_image
)) {
2484 /* So this is for a dynamic user, and we need to make sure the process can access its own
2485 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2486 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2488 private_root
= path_join(params
->prefix
[t
], "private");
2489 if (!private_root
) {
2494 r
= strv_consume(&empty_directories
, private_root
);
2499 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2502 if (exec_directory_is_private(context
, t
))
2503 s
= path_join(params
->prefix
[t
], "private", *suffix
);
2505 s
= path_join(params
->prefix
[t
], *suffix
);
2511 if (exec_directory_is_private(context
, t
) &&
2512 (context
->root_directory
|| context
->root_image
))
2513 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2514 * directory is not created on the root directory. So, let's bind-mount the directory
2515 * on the 'non-private' place. */
2516 d
= path_join(params
->prefix
[t
], *suffix
);
2525 bind_mounts
[h
++] = (BindMount
) {
2529 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
2531 .ignore_enoent
= false,
2538 *ret_bind_mounts
= bind_mounts
;
2539 *ret_n_bind_mounts
= n
;
2540 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2545 bind_mount_free_many(bind_mounts
, h
);
2549 static bool insist_on_sandboxing(
2550 const ExecContext
*context
,
2551 const char *root_dir
,
2552 const char *root_image
,
2553 const BindMount
*bind_mounts
,
2554 size_t n_bind_mounts
) {
2559 assert(n_bind_mounts
== 0 || bind_mounts
);
2561 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
2562 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
2563 * rearrange stuff in a way we cannot ignore gracefully. */
2565 if (context
->n_temporary_filesystems
> 0)
2568 if (root_dir
|| root_image
)
2571 if (context
->dynamic_user
)
2574 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
2576 for (i
= 0; i
< n_bind_mounts
; i
++)
2577 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
2580 if (context
->log_namespace
)
2586 static int apply_mount_namespace(
2588 const ExecCommand
*command
,
2589 const ExecContext
*context
,
2590 const ExecParameters
*params
,
2591 const ExecRuntime
*runtime
,
2592 char **error_path
) {
2594 _cleanup_strv_free_
char **empty_directories
= NULL
;
2595 char *tmp
= NULL
, *var
= NULL
;
2596 const char *root_dir
= NULL
, *root_image
= NULL
;
2597 NamespaceInfo ns_info
;
2598 bool needs_sandboxing
;
2599 BindMount
*bind_mounts
= NULL
;
2600 size_t n_bind_mounts
= 0;
2605 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2606 root_image
= context
->root_image
;
2609 root_dir
= context
->root_directory
;
2612 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2616 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2617 if (needs_sandboxing
) {
2618 /* The runtime struct only contains the parent of the private /tmp,
2619 * which is non-accessible to world users. Inside of it there's a /tmp
2620 * that is sticky, and that's the one we want to use here. */
2622 if (context
->private_tmp
&& runtime
) {
2623 if (runtime
->tmp_dir
)
2624 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2625 if (runtime
->var_tmp_dir
)
2626 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2629 ns_info
= (NamespaceInfo
) {
2630 .ignore_protect_paths
= false,
2631 .private_dev
= context
->private_devices
,
2632 .protect_control_groups
= context
->protect_control_groups
,
2633 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2634 .protect_kernel_modules
= context
->protect_kernel_modules
,
2635 .protect_kernel_logs
= context
->protect_kernel_logs
,
2636 .protect_hostname
= context
->protect_hostname
,
2637 .mount_apivfs
= context
->mount_apivfs
,
2638 .private_mounts
= context
->private_mounts
,
2640 } else if (!context
->dynamic_user
&& root_dir
)
2642 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2643 * sandbox info, otherwise enforce it, don't ignore protected paths and
2644 * fail if we are enable to apply the sandbox inside the mount namespace.
2646 ns_info
= (NamespaceInfo
) {
2647 .ignore_protect_paths
= true,
2650 ns_info
= (NamespaceInfo
) {};
2652 if (context
->mount_flags
== MS_SHARED
)
2653 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2655 r
= setup_namespace(root_dir
, root_image
,
2656 &ns_info
, context
->read_write_paths
,
2657 needs_sandboxing
? context
->read_only_paths
: NULL
,
2658 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2662 context
->temporary_filesystems
,
2663 context
->n_temporary_filesystems
,
2666 context
->log_namespace
,
2667 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2668 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2669 context
->mount_flags
,
2670 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
2671 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
2672 context
->root_verity
,
2673 DISSECT_IMAGE_DISCARD_ON_LOOP
|DISSECT_IMAGE_RELAX_VAR_CHECK
|DISSECT_IMAGE_FSCK
,
2676 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2677 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
2678 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2679 * completely different execution environment. */
2681 if (insist_on_sandboxing(
2683 root_dir
, root_image
,
2686 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2687 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2688 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2692 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2697 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2701 static int apply_working_directory(
2702 const ExecContext
*context
,
2703 const ExecParameters
*params
,
2710 assert(exit_status
);
2712 if (context
->working_directory_home
) {
2715 *exit_status
= EXIT_CHDIR
;
2721 } else if (context
->working_directory
)
2722 wd
= context
->working_directory
;
2726 if (params
->flags
& EXEC_APPLY_CHROOT
)
2729 d
= prefix_roota(context
->root_directory
, wd
);
2731 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2732 *exit_status
= EXIT_CHDIR
;
2739 static int apply_root_directory(
2740 const ExecContext
*context
,
2741 const ExecParameters
*params
,
2742 const bool needs_mount_ns
,
2746 assert(exit_status
);
2748 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2749 if (!needs_mount_ns
&& context
->root_directory
)
2750 if (chroot(context
->root_directory
) < 0) {
2751 *exit_status
= EXIT_CHROOT
;
2759 static int setup_keyring(
2761 const ExecContext
*context
,
2762 const ExecParameters
*p
,
2763 uid_t uid
, gid_t gid
) {
2765 key_serial_t keyring
;
2774 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2775 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2776 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2777 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2778 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2779 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2781 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2784 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2785 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2786 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2787 * & group is just as nasty as acquiring a reference to the user keyring. */
2789 saved_uid
= getuid();
2790 saved_gid
= getgid();
2792 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2793 if (setregid(gid
, -1) < 0)
2794 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2797 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2798 if (setreuid(uid
, -1) < 0) {
2799 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2804 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2805 if (keyring
== -1) {
2806 if (errno
== ENOSYS
)
2807 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2808 else if (IN_SET(errno
, EACCES
, EPERM
))
2809 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2810 else if (errno
== EDQUOT
)
2811 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2813 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2818 /* When requested link the user keyring into the session keyring. */
2819 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2821 if (keyctl(KEYCTL_LINK
,
2822 KEY_SPEC_USER_KEYRING
,
2823 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2824 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2829 /* Restore uid/gid back */
2830 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2831 if (setreuid(saved_uid
, -1) < 0) {
2832 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2837 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2838 if (setregid(saved_gid
, -1) < 0)
2839 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2842 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2843 if (!sd_id128_is_null(u
->invocation_id
)) {
2846 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2848 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2850 if (keyctl(KEYCTL_SETPERM
, key
,
2851 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2852 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2853 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2858 /* Revert back uid & gid for the last time, and exit */
2859 /* no extra logging, as only the first already reported error matters */
2860 if (getuid() != saved_uid
)
2861 (void) setreuid(saved_uid
, -1);
2863 if (getgid() != saved_gid
)
2864 (void) setregid(saved_gid
, -1);
2869 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2875 array
[(*n
)++] = pair
[0];
2877 array
[(*n
)++] = pair
[1];
2880 static int close_remaining_fds(
2881 const ExecParameters
*params
,
2882 const ExecRuntime
*runtime
,
2883 const DynamicCreds
*dcreds
,
2887 const int *fds
, size_t n_fds
) {
2889 size_t n_dont_close
= 0;
2890 int dont_close
[n_fds
+ 12];
2894 if (params
->stdin_fd
>= 0)
2895 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2896 if (params
->stdout_fd
>= 0)
2897 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2898 if (params
->stderr_fd
>= 0)
2899 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2902 dont_close
[n_dont_close
++] = socket_fd
;
2904 dont_close
[n_dont_close
++] = exec_fd
;
2906 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2907 n_dont_close
+= n_fds
;
2911 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2915 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2917 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2920 if (user_lookup_fd
>= 0)
2921 dont_close
[n_dont_close
++] = user_lookup_fd
;
2923 return close_all_fds(dont_close
, n_dont_close
);
2926 static int send_user_lookup(
2934 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2935 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2938 if (user_lookup_fd
< 0)
2941 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2944 if (writev(user_lookup_fd
,
2946 IOVEC_INIT(&uid
, sizeof(uid
)),
2947 IOVEC_INIT(&gid
, sizeof(gid
)),
2948 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2954 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2961 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2966 if (!c
->working_directory_home
)
2969 r
= get_home_dir(buf
);
2977 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2978 _cleanup_strv_free_
char ** list
= NULL
;
2979 ExecDirectoryType t
;
2986 assert(c
->dynamic_user
);
2988 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2989 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2992 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2995 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3001 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3004 if (exec_directory_is_private(c
, t
))
3005 e
= path_join(p
->prefix
[t
], "private", *i
);
3007 e
= path_join(p
->prefix
[t
], *i
);
3011 r
= strv_consume(&list
, e
);
3017 *ret
= TAKE_PTR(list
);
3022 static char *exec_command_line(char **argv
);
3024 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3025 bool using_subcgroup
;
3031 if (!params
->cgroup_path
)
3034 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3035 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3036 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3037 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3038 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3039 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3040 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3041 * flag, which is only passed for the former statements, not for the latter. */
3043 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3044 if (using_subcgroup
)
3045 p
= path_join(params
->cgroup_path
, ".control");
3047 p
= strdup(params
->cgroup_path
);
3052 return using_subcgroup
;
3055 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3056 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3062 if (!c
->numa_policy
.nodes
.set
) {
3063 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3067 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3073 return cpu_set_add_all(ret
, &s
);
3076 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3079 return c
->cpu_affinity_from_numa
;
3082 static int exec_child(
3084 const ExecCommand
*command
,
3085 const ExecContext
*context
,
3086 const ExecParameters
*params
,
3087 ExecRuntime
*runtime
,
3088 DynamicCreds
*dcreds
,
3090 const int named_iofds
[static 3],
3092 size_t n_socket_fds
,
3093 size_t n_storage_fds
,
3098 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3099 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
3100 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3101 const char *username
= NULL
, *groupname
= NULL
;
3102 _cleanup_free_
char *home_buffer
= NULL
;
3103 const char *home
= NULL
, *shell
= NULL
;
3104 char **final_argv
= NULL
;
3105 dev_t journal_stream_dev
= 0;
3106 ino_t journal_stream_ino
= 0;
3107 bool userns_set_up
= false;
3108 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3109 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3110 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3111 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3113 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3114 bool use_selinux
= false;
3117 bool use_smack
= false;
3120 bool use_apparmor
= false;
3122 uid_t saved_uid
= getuid();
3123 gid_t saved_gid
= getgid();
3124 uid_t uid
= UID_INVALID
;
3125 gid_t gid
= GID_INVALID
;
3127 ExecDirectoryType dt
;
3129 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3130 int ngids_after_pam
= 0;
3136 assert(exit_status
);
3138 rename_process_from_path(command
->path
);
3140 /* We reset exactly these signals, since they are the
3141 * only ones we set to SIG_IGN in the main daemon. All
3142 * others we leave untouched because we set them to
3143 * SIG_DFL or a valid handler initially, both of which
3144 * will be demoted to SIG_DFL. */
3145 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3146 SIGNALS_IGNORE
, -1);
3148 if (context
->ignore_sigpipe
)
3149 (void) ignore_signals(SIGPIPE
, -1);
3151 r
= reset_signal_mask();
3153 *exit_status
= EXIT_SIGNAL_MASK
;
3154 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3157 if (params
->idle_pipe
)
3158 do_idle_pipe_dance(params
->idle_pipe
);
3160 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3161 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3162 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3163 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3166 log_set_open_when_needed(true);
3168 /* In case anything used libc syslog(), close this here, too */
3171 n_fds
= n_socket_fds
+ n_storage_fds
;
3172 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
3174 *exit_status
= EXIT_FDS
;
3175 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3178 if (!context
->same_pgrp
)
3180 *exit_status
= EXIT_SETSID
;
3181 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3184 exec_context_tty_reset(context
, params
);
3186 if (unit_shall_confirm_spawn(unit
)) {
3187 const char *vc
= params
->confirm_spawn
;
3188 _cleanup_free_
char *cmdline
= NULL
;
3190 cmdline
= exec_command_line(command
->argv
);
3192 *exit_status
= EXIT_MEMORY
;
3196 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3197 if (r
!= CONFIRM_EXECUTE
) {
3198 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3199 *exit_status
= EXIT_SUCCESS
;
3202 *exit_status
= EXIT_CONFIRM
;
3203 log_unit_error(unit
, "Execution cancelled by the user");
3208 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3209 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3210 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3211 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3212 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3213 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3214 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3215 *exit_status
= EXIT_MEMORY
;
3216 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3219 if (context
->dynamic_user
&& dcreds
) {
3220 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3222 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3223 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
3224 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3225 *exit_status
= EXIT_USER
;
3226 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3229 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3231 *exit_status
= EXIT_MEMORY
;
3235 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3237 *exit_status
= EXIT_USER
;
3239 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3242 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3245 if (!uid_is_valid(uid
)) {
3246 *exit_status
= EXIT_USER
;
3247 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3251 if (!gid_is_valid(gid
)) {
3252 *exit_status
= EXIT_USER
;
3253 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3258 username
= dcreds
->user
->name
;
3261 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3263 *exit_status
= EXIT_USER
;
3264 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3267 r
= get_fixed_group(context
, &groupname
, &gid
);
3269 *exit_status
= EXIT_GROUP
;
3270 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3274 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3275 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3276 &supplementary_gids
, &ngids
);
3278 *exit_status
= EXIT_GROUP
;
3279 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3282 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3284 *exit_status
= EXIT_USER
;
3285 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3288 user_lookup_fd
= safe_close(user_lookup_fd
);
3290 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3292 *exit_status
= EXIT_CHDIR
;
3293 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3296 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3297 * must sure to drop O_NONBLOCK */
3299 (void) fd_nonblock(socket_fd
, false);
3301 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3302 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3303 if (params
->cgroup_path
) {
3304 _cleanup_free_
char *p
= NULL
;
3306 r
= exec_parameters_get_cgroup_path(params
, &p
);
3308 *exit_status
= EXIT_CGROUP
;
3309 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3312 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3314 *exit_status
= EXIT_CGROUP
;
3315 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3319 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3320 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3322 *exit_status
= EXIT_NETWORK
;
3323 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3327 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3329 *exit_status
= EXIT_STDIN
;
3330 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3333 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3335 *exit_status
= EXIT_STDOUT
;
3336 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3339 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3341 *exit_status
= EXIT_STDERR
;
3342 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3345 if (context
->oom_score_adjust_set
) {
3346 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3347 * prohibit write access to this file, and we shouldn't trip up over that. */
3348 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3349 if (IN_SET(r
, -EPERM
, -EACCES
))
3350 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3352 *exit_status
= EXIT_OOM_ADJUST
;
3353 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3357 if (context
->coredump_filter_set
) {
3358 r
= set_coredump_filter(context
->coredump_filter
);
3359 if (ERRNO_IS_PRIVILEGE(r
))
3360 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
3362 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
3365 if (context
->nice_set
) {
3366 r
= setpriority_closest(context
->nice
);
3368 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3371 if (context
->cpu_sched_set
) {
3372 struct sched_param param
= {
3373 .sched_priority
= context
->cpu_sched_priority
,
3376 r
= sched_setscheduler(0,
3377 context
->cpu_sched_policy
|
3378 (context
->cpu_sched_reset_on_fork
?
3379 SCHED_RESET_ON_FORK
: 0),
3382 *exit_status
= EXIT_SETSCHEDULER
;
3383 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3387 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
3388 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
3389 const CPUSet
*cpu_set
;
3391 if (context
->cpu_affinity_from_numa
) {
3392 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
3394 *exit_status
= EXIT_CPUAFFINITY
;
3395 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
3398 cpu_set
= &converted_cpu_set
;
3400 cpu_set
= &context
->cpu_set
;
3402 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
3403 *exit_status
= EXIT_CPUAFFINITY
;
3404 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3408 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
3409 r
= apply_numa_policy(&context
->numa_policy
);
3410 if (r
== -EOPNOTSUPP
)
3411 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
3413 *exit_status
= EXIT_NUMA_POLICY
;
3414 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
3418 if (context
->ioprio_set
)
3419 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3420 *exit_status
= EXIT_IOPRIO
;
3421 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3424 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3425 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3426 *exit_status
= EXIT_TIMERSLACK
;
3427 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3430 if (context
->personality
!= PERSONALITY_INVALID
) {
3431 r
= safe_personality(context
->personality
);
3433 *exit_status
= EXIT_PERSONALITY
;
3434 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3438 if (context
->utmp_id
)
3439 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3441 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3442 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3446 if (uid_is_valid(uid
)) {
3447 r
= chown_terminal(STDIN_FILENO
, uid
);
3449 *exit_status
= EXIT_STDIN
;
3450 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3454 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3455 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3456 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3457 * touch a single hierarchy too. */
3458 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3459 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3461 *exit_status
= EXIT_CGROUP
;
3462 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3466 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3467 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3469 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3472 r
= build_environment(
3484 *exit_status
= EXIT_MEMORY
;
3488 r
= build_pass_environment(context
, &pass_env
);
3490 *exit_status
= EXIT_MEMORY
;
3494 accum_env
= strv_env_merge(5,
3495 params
->environment
,
3498 context
->environment
,
3501 *exit_status
= EXIT_MEMORY
;
3504 accum_env
= strv_env_clean(accum_env
);
3506 (void) umask(context
->umask
);
3508 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3510 *exit_status
= EXIT_KEYRING
;
3511 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3514 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3515 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3517 /* 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 */
3518 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3520 /* 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 */
3521 if (needs_ambient_hack
)
3522 needs_setuid
= false;
3524 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3526 if (needs_sandboxing
) {
3527 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3528 * present. The actual MAC context application will happen later, as late as possible, to avoid
3529 * impacting our own code paths. */
3532 use_selinux
= mac_selinux_use();
3535 use_smack
= mac_smack_use();
3538 use_apparmor
= mac_apparmor_use();
3542 if (needs_sandboxing
) {
3545 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3546 * is set here. (See below.) */
3548 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3550 *exit_status
= EXIT_LIMITS
;
3551 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3557 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3558 * wins here. (See above.) */
3560 if (context
->pam_name
&& username
) {
3561 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3563 *exit_status
= EXIT_PAM
;
3564 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3567 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
3568 if (ngids_after_pam
< 0) {
3569 *exit_status
= EXIT_MEMORY
;
3570 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
3575 if (needs_sandboxing
) {
3577 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3578 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3580 *exit_status
= EXIT_SELINUX_CONTEXT
;
3581 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3586 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
3587 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
3588 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
3589 if (context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
3590 userns_set_up
= true;
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 for unprivileged user: %m");
3599 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3601 if (ns_type_supported(NAMESPACE_NET
)) {
3602 r
= setup_netns(runtime
->netns_storage_socket
);
3604 log_unit_warning_errno(unit
, r
,
3605 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
3607 *exit_status
= EXIT_NETWORK
;
3608 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3610 } else if (context
->network_namespace_path
) {
3611 *exit_status
= EXIT_NETWORK
;
3612 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
3613 "NetworkNamespacePath= is not supported, refusing.");
3615 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3618 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3619 if (needs_mount_namespace
) {
3620 _cleanup_free_
char *error_path
= NULL
;
3622 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
, &error_path
);
3624 *exit_status
= EXIT_NAMESPACE
;
3625 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
3626 error_path
? ": " : "", strempty(error_path
));
3630 if (needs_sandboxing
) {
3631 r
= apply_protect_hostname(unit
, context
, exit_status
);
3636 /* Drop groups as early as possible.
3637 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
3638 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
3640 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
3641 int ngids_to_enforce
= 0;
3643 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
3648 if (ngids_to_enforce
< 0) {
3649 *exit_status
= EXIT_MEMORY
;
3650 return log_unit_error_errno(unit
,
3652 "Failed to merge group lists. Group membership might be incorrect: %m");
3655 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
3657 *exit_status
= EXIT_GROUP
;
3658 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3662 /* If the user namespace was not set up above, try to do it now.
3663 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
3664 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
3665 * case of mount namespaces being less privileged when the mount point list is copied from a
3666 * different user namespace). */
3668 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
3669 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
3671 *exit_status
= EXIT_USER
;
3672 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3676 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3677 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3678 * however if we have it as we want to keep it open until the final execve(). */
3680 if (params
->exec_fd
>= 0) {
3681 exec_fd
= params
->exec_fd
;
3683 if (exec_fd
< 3 + (int) n_fds
) {
3686 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3687 * process we are about to execute. */
3689 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3691 *exit_status
= EXIT_FDS
;
3692 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3695 safe_close(exec_fd
);
3698 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3699 r
= fd_cloexec(exec_fd
, true);
3701 *exit_status
= EXIT_FDS
;
3702 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3706 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3707 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3708 fds_with_exec_fd
[n_fds
] = exec_fd
;
3709 n_fds_with_exec_fd
= n_fds
+ 1;
3711 fds_with_exec_fd
= fds
;
3712 n_fds_with_exec_fd
= n_fds
;
3715 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3717 r
= shift_fds(fds
, n_fds
);
3719 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3721 *exit_status
= EXIT_FDS
;
3722 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3725 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3726 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3727 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3730 secure_bits
= context
->secure_bits
;
3732 if (needs_sandboxing
) {
3735 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3736 * requested. (Note this is placed after the general resource limit initialization, see
3737 * above, in order to take precedence.) */
3738 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3739 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3740 *exit_status
= EXIT_LIMITS
;
3741 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3746 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3747 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3749 r
= setup_smack(context
, command
);
3751 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3752 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3757 bset
= context
->capability_bounding_set
;
3758 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3759 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3760 * instead of us doing that */
3761 if (needs_ambient_hack
)
3762 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3763 (UINT64_C(1) << CAP_SETUID
) |
3764 (UINT64_C(1) << CAP_SETGID
);
3766 if (!cap_test_all(bset
)) {
3767 r
= capability_bounding_set_drop(bset
, false);
3769 *exit_status
= EXIT_CAPABILITIES
;
3770 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3774 /* This is done before enforce_user, but ambient set
3775 * does not survive over setresuid() if keep_caps is not set. */
3776 if (!needs_ambient_hack
) {
3777 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3779 *exit_status
= EXIT_CAPABILITIES
;
3780 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3785 /* chroot to root directory first, before we lose the ability to chroot */
3786 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
3788 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
3791 if (uid_is_valid(uid
)) {
3792 r
= enforce_user(context
, uid
);
3794 *exit_status
= EXIT_USER
;
3795 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3798 if (!needs_ambient_hack
&&
3799 context
->capability_ambient_set
!= 0) {
3801 /* Fix the ambient capabilities after user change. */
3802 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3804 *exit_status
= EXIT_CAPABILITIES
;
3805 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3808 /* If we were asked to change user and ambient capabilities
3809 * were requested, we had to add keep-caps to the securebits
3810 * so that we would maintain the inherited capability set
3811 * through the setresuid(). Make sure that the bit is added
3812 * also to the context secure_bits so that we don't try to
3813 * drop the bit away next. */
3815 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3820 /* Apply working directory here, because the working directory might be on NFS and only the user running
3821 * this service might have the correct privilege to change to the working directory */
3822 r
= apply_working_directory(context
, params
, home
, exit_status
);
3824 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3826 if (needs_sandboxing
) {
3827 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3828 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3829 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3830 * are restricted. */
3834 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3837 r
= setexeccon(exec_context
);
3839 *exit_status
= EXIT_SELINUX_CONTEXT
;
3840 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3847 if (use_apparmor
&& context
->apparmor_profile
) {
3848 r
= aa_change_onexec(context
->apparmor_profile
);
3849 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3850 *exit_status
= EXIT_APPARMOR_PROFILE
;
3851 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3856 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3857 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3858 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3859 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3860 *exit_status
= EXIT_SECUREBITS
;
3861 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3864 if (context_has_no_new_privileges(context
))
3865 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3866 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3867 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3871 r
= apply_address_families(unit
, context
);
3873 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3874 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3877 r
= apply_memory_deny_write_execute(unit
, context
);
3879 *exit_status
= EXIT_SECCOMP
;
3880 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3883 r
= apply_restrict_realtime(unit
, context
);
3885 *exit_status
= EXIT_SECCOMP
;
3886 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3889 r
= apply_restrict_suid_sgid(unit
, context
);
3891 *exit_status
= EXIT_SECCOMP
;
3892 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
3895 r
= apply_restrict_namespaces(unit
, context
);
3897 *exit_status
= EXIT_SECCOMP
;
3898 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3901 r
= apply_protect_sysctl(unit
, context
);
3903 *exit_status
= EXIT_SECCOMP
;
3904 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3907 r
= apply_protect_kernel_modules(unit
, context
);
3909 *exit_status
= EXIT_SECCOMP
;
3910 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3913 r
= apply_protect_kernel_logs(unit
, context
);
3915 *exit_status
= EXIT_SECCOMP
;
3916 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
3919 r
= apply_protect_clock(unit
, context
);
3921 *exit_status
= EXIT_SECCOMP
;
3922 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
3925 r
= apply_private_devices(unit
, context
);
3927 *exit_status
= EXIT_SECCOMP
;
3928 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3931 r
= apply_syscall_archs(unit
, context
);
3933 *exit_status
= EXIT_SECCOMP
;
3934 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3937 r
= apply_lock_personality(unit
, context
);
3939 *exit_status
= EXIT_SECCOMP
;
3940 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3943 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3944 * by the filter as little as possible. */
3945 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3947 *exit_status
= EXIT_SECCOMP
;
3948 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3953 if (!strv_isempty(context
->unset_environment
)) {
3956 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3958 *exit_status
= EXIT_MEMORY
;
3962 strv_free_and_replace(accum_env
, ee
);
3965 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3966 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3967 if (!replaced_argv
) {
3968 *exit_status
= EXIT_MEMORY
;
3971 final_argv
= replaced_argv
;
3973 final_argv
= command
->argv
;
3975 if (DEBUG_LOGGING
) {
3976 _cleanup_free_
char *line
;
3978 line
= exec_command_line(final_argv
);
3980 log_struct(LOG_DEBUG
,
3981 "EXECUTABLE=%s", command
->path
,
3982 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3984 LOG_UNIT_INVOCATION_ID(unit
));
3990 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3991 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3993 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3994 *exit_status
= EXIT_EXEC
;
3995 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3999 execve(command
->path
, final_argv
, accum_env
);
4005 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4006 * that POLLHUP on it no longer means execve() succeeded. */
4008 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4009 *exit_status
= EXIT_EXEC
;
4010 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4014 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4015 log_struct_errno(LOG_INFO
, r
,
4016 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4018 LOG_UNIT_INVOCATION_ID(unit
),
4019 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4021 "EXECUTABLE=%s", command
->path
);
4025 *exit_status
= EXIT_EXEC
;
4026 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
4029 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4030 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4032 int exec_spawn(Unit
*unit
,
4033 ExecCommand
*command
,
4034 const ExecContext
*context
,
4035 const ExecParameters
*params
,
4036 ExecRuntime
*runtime
,
4037 DynamicCreds
*dcreds
,
4040 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4041 _cleanup_free_
char *subcgroup_path
= NULL
;
4042 _cleanup_strv_free_
char **files_env
= NULL
;
4043 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4044 _cleanup_free_
char *line
= NULL
;
4052 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4054 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4055 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4056 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4058 if (params
->n_socket_fds
> 1) {
4059 log_unit_error(unit
, "Got more than one socket.");
4063 if (params
->n_socket_fds
== 0) {
4064 log_unit_error(unit
, "Got no socket.");
4068 socket_fd
= params
->fds
[0];
4072 n_socket_fds
= params
->n_socket_fds
;
4073 n_storage_fds
= params
->n_storage_fds
;
4076 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4078 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4080 r
= exec_context_load_environment(unit
, context
, &files_env
);
4082 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4084 line
= exec_command_line(command
->argv
);
4088 log_struct(LOG_DEBUG
,
4089 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
4090 "EXECUTABLE=%s", command
->path
,
4092 LOG_UNIT_INVOCATION_ID(unit
));
4094 if (params
->cgroup_path
) {
4095 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4097 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4098 if (r
> 0) { /* We are using a child cgroup */
4099 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4101 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4107 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4110 int exit_status
= EXIT_SUCCESS
;
4112 r
= exec_child(unit
,
4124 unit
->manager
->user_lookup_fds
[1],
4128 const char *status
=
4129 exit_status_to_string(exit_status
,
4130 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4132 log_struct_errno(LOG_ERR
, r
,
4133 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4135 LOG_UNIT_INVOCATION_ID(unit
),
4136 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4137 status
, command
->path
),
4138 "EXECUTABLE=%s", command
->path
);
4144 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4146 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4147 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4148 * process will be killed too). */
4150 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4152 exec_status_start(&command
->exec_status
, pid
);
4158 void exec_context_init(ExecContext
*c
) {
4159 ExecDirectoryType i
;
4164 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4165 c
->cpu_sched_policy
= SCHED_OTHER
;
4166 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4167 c
->syslog_level_prefix
= true;
4168 c
->ignore_sigpipe
= true;
4169 c
->timer_slack_nsec
= NSEC_INFINITY
;
4170 c
->personality
= PERSONALITY_INVALID
;
4171 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
4172 c
->directories
[i
].mode
= 0755;
4173 c
->timeout_clean_usec
= USEC_INFINITY
;
4174 c
->capability_bounding_set
= CAP_ALL
;
4175 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4176 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4177 c
->log_level_max
= -1;
4178 numa_policy_reset(&c
->numa_policy
);
4181 void exec_context_done(ExecContext
*c
) {
4182 ExecDirectoryType i
;
4187 c
->environment
= strv_free(c
->environment
);
4188 c
->environment_files
= strv_free(c
->environment_files
);
4189 c
->pass_environment
= strv_free(c
->pass_environment
);
4190 c
->unset_environment
= strv_free(c
->unset_environment
);
4192 rlimit_free_all(c
->rlimit
);
4194 for (l
= 0; l
< 3; l
++) {
4195 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4196 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4199 c
->working_directory
= mfree(c
->working_directory
);
4200 c
->root_directory
= mfree(c
->root_directory
);
4201 c
->root_image
= mfree(c
->root_image
);
4202 c
->root_hash
= mfree(c
->root_hash
);
4203 c
->root_hash_size
= 0;
4204 c
->root_hash_path
= mfree(c
->root_hash_path
);
4205 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
4206 c
->root_hash_sig_size
= 0;
4207 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
4208 c
->root_verity
= mfree(c
->root_verity
);
4209 c
->tty_path
= mfree(c
->tty_path
);
4210 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4211 c
->user
= mfree(c
->user
);
4212 c
->group
= mfree(c
->group
);
4214 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4216 c
->pam_name
= mfree(c
->pam_name
);
4218 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4219 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4220 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4222 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4223 c
->bind_mounts
= NULL
;
4224 c
->n_bind_mounts
= 0;
4225 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4226 c
->temporary_filesystems
= NULL
;
4227 c
->n_temporary_filesystems
= 0;
4229 cpu_set_reset(&c
->cpu_set
);
4230 numa_policy_reset(&c
->numa_policy
);
4232 c
->utmp_id
= mfree(c
->utmp_id
);
4233 c
->selinux_context
= mfree(c
->selinux_context
);
4234 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4235 c
->smack_process_label
= mfree(c
->smack_process_label
);
4237 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4238 c
->syscall_archs
= set_free(c
->syscall_archs
);
4239 c
->address_families
= set_free(c
->address_families
);
4241 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
4242 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
4244 c
->log_level_max
= -1;
4246 exec_context_free_log_extra_fields(c
);
4248 c
->log_ratelimit_interval_usec
= 0;
4249 c
->log_ratelimit_burst
= 0;
4251 c
->stdin_data
= mfree(c
->stdin_data
);
4252 c
->stdin_data_size
= 0;
4254 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4256 c
->log_namespace
= mfree(c
->log_namespace
);
4259 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4264 if (!runtime_prefix
)
4267 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4268 _cleanup_free_
char *p
;
4270 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4271 p
= path_join(runtime_prefix
, "private", *i
);
4273 p
= path_join(runtime_prefix
, *i
);
4277 /* We execute this synchronously, since we need to be sure this is gone when we start the
4279 (void) rm_rf(p
, REMOVE_ROOT
);
4285 static void exec_command_done(ExecCommand
*c
) {
4288 c
->path
= mfree(c
->path
);
4289 c
->argv
= strv_free(c
->argv
);
4292 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4295 for (i
= 0; i
< n
; i
++)
4296 exec_command_done(c
+i
);
4299 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4303 LIST_REMOVE(command
, c
, i
);
4304 exec_command_done(i
);
4311 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4314 for (i
= 0; i
< n
; i
++)
4315 c
[i
] = exec_command_free_list(c
[i
]);
4318 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4321 for (i
= 0; i
< n
; i
++)
4322 exec_status_reset(&c
[i
].exec_status
);
4325 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
4328 for (i
= 0; i
< n
; i
++) {
4331 LIST_FOREACH(command
, z
, c
[i
])
4332 exec_status_reset(&z
->exec_status
);
4336 typedef struct InvalidEnvInfo
{
4341 static void invalid_env(const char *p
, void *userdata
) {
4342 InvalidEnvInfo
*info
= userdata
;
4344 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4347 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4353 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4356 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4359 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4362 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4365 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4368 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4375 static int exec_context_named_iofds(
4376 const ExecContext
*c
,
4377 const ExecParameters
*p
,
4378 int named_iofds
[static 3]) {
4381 const char* stdio_fdname
[3];
4386 assert(named_iofds
);
4388 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4389 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4390 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4392 for (i
= 0; i
< 3; i
++)
4393 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4395 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4397 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4398 if (named_iofds
[STDIN_FILENO
] < 0 &&
4399 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4400 stdio_fdname
[STDIN_FILENO
] &&
4401 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4403 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4406 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4407 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4408 stdio_fdname
[STDOUT_FILENO
] &&
4409 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4411 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4414 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4415 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4416 stdio_fdname
[STDERR_FILENO
] &&
4417 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4419 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4423 return targets
== 0 ? 0 : -ENOENT
;
4426 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4427 char **i
, **r
= NULL
;
4432 STRV_FOREACH(i
, c
->environment_files
) {
4436 bool ignore
= false;
4438 _cleanup_globfree_ glob_t pglob
= {};
4447 if (!path_is_absolute(fn
)) {
4455 /* Filename supports globbing, take all matching files */
4456 k
= safe_glob(fn
, 0, &pglob
);
4465 /* When we don't match anything, -ENOENT should be returned */
4466 assert(pglob
.gl_pathc
> 0);
4468 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4469 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4477 /* Log invalid environment variables with filename */
4479 InvalidEnvInfo info
= {
4481 .path
= pglob
.gl_pathv
[n
]
4484 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4492 m
= strv_env_merge(2, r
, p
);
4508 static bool tty_may_match_dev_console(const char *tty
) {
4509 _cleanup_free_
char *resolved
= NULL
;
4514 tty
= skip_dev_prefix(tty
);
4516 /* trivial identity? */
4517 if (streq(tty
, "console"))
4520 if (resolve_dev_console(&resolved
) < 0)
4521 return true; /* if we could not resolve, assume it may */
4523 /* "tty0" means the active VC, so it may be the same sometimes */
4524 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4527 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
4530 return ec
->tty_reset
||
4532 ec
->tty_vt_disallocate
||
4533 is_terminal_input(ec
->std_input
) ||
4534 is_terminal_output(ec
->std_output
) ||
4535 is_terminal_output(ec
->std_error
);
4538 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4540 return exec_context_may_touch_tty(ec
) &&
4541 tty_may_match_dev_console(exec_context_tty_path(ec
));
4544 static void strv_fprintf(FILE *f
, char **l
) {
4550 fprintf(f
, " %s", *g
);
4553 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4554 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
4555 ExecDirectoryType dt
;
4562 prefix
= strempty(prefix
);
4566 "%sWorkingDirectory: %s\n"
4567 "%sRootDirectory: %s\n"
4568 "%sNonBlocking: %s\n"
4569 "%sPrivateTmp: %s\n"
4570 "%sPrivateDevices: %s\n"
4571 "%sProtectKernelTunables: %s\n"
4572 "%sProtectKernelModules: %s\n"
4573 "%sProtectKernelLogs: %s\n"
4574 "%sProtectClock: %s\n"
4575 "%sProtectControlGroups: %s\n"
4576 "%sPrivateNetwork: %s\n"
4577 "%sPrivateUsers: %s\n"
4578 "%sProtectHome: %s\n"
4579 "%sProtectSystem: %s\n"
4580 "%sMountAPIVFS: %s\n"
4581 "%sIgnoreSIGPIPE: %s\n"
4582 "%sMemoryDenyWriteExecute: %s\n"
4583 "%sRestrictRealtime: %s\n"
4584 "%sRestrictSUIDSGID: %s\n"
4585 "%sKeyringMode: %s\n"
4586 "%sProtectHostname: %s\n",
4588 prefix
, c
->working_directory
? c
->working_directory
: "/",
4589 prefix
, c
->root_directory
? c
->root_directory
: "/",
4590 prefix
, yes_no(c
->non_blocking
),
4591 prefix
, yes_no(c
->private_tmp
),
4592 prefix
, yes_no(c
->private_devices
),
4593 prefix
, yes_no(c
->protect_kernel_tunables
),
4594 prefix
, yes_no(c
->protect_kernel_modules
),
4595 prefix
, yes_no(c
->protect_kernel_logs
),
4596 prefix
, yes_no(c
->protect_clock
),
4597 prefix
, yes_no(c
->protect_control_groups
),
4598 prefix
, yes_no(c
->private_network
),
4599 prefix
, yes_no(c
->private_users
),
4600 prefix
, protect_home_to_string(c
->protect_home
),
4601 prefix
, protect_system_to_string(c
->protect_system
),
4602 prefix
, yes_no(c
->mount_apivfs
),
4603 prefix
, yes_no(c
->ignore_sigpipe
),
4604 prefix
, yes_no(c
->memory_deny_write_execute
),
4605 prefix
, yes_no(c
->restrict_realtime
),
4606 prefix
, yes_no(c
->restrict_suid_sgid
),
4607 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4608 prefix
, yes_no(c
->protect_hostname
));
4611 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4614 _cleanup_free_
char *encoded
= NULL
;
4615 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
4617 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
4620 if (c
->root_hash_path
)
4621 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
4623 if (c
->root_hash_sig
) {
4624 _cleanup_free_
char *encoded
= NULL
;
4626 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
4628 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
4631 if (c
->root_hash_sig_path
)
4632 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
4635 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
4637 STRV_FOREACH(e
, c
->environment
)
4638 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4640 STRV_FOREACH(e
, c
->environment_files
)
4641 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4643 STRV_FOREACH(e
, c
->pass_environment
)
4644 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4646 STRV_FOREACH(e
, c
->unset_environment
)
4647 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4649 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4651 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4652 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4654 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4655 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4659 "%sTimeoutCleanSec: %s\n",
4660 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
4667 if (c
->oom_score_adjust_set
)
4669 "%sOOMScoreAdjust: %i\n",
4670 prefix
, c
->oom_score_adjust
);
4672 if (c
->coredump_filter_set
)
4674 "%sCoredumpFilter: 0x%"PRIx64
"\n",
4675 prefix
, c
->coredump_filter
);
4677 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4679 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4680 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4681 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4682 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4685 if (c
->ioprio_set
) {
4686 _cleanup_free_
char *class_str
= NULL
;
4688 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4690 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4692 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4695 if (c
->cpu_sched_set
) {
4696 _cleanup_free_
char *policy_str
= NULL
;
4698 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4700 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4703 "%sCPUSchedulingPriority: %i\n"
4704 "%sCPUSchedulingResetOnFork: %s\n",
4705 prefix
, c
->cpu_sched_priority
,
4706 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4709 if (c
->cpu_set
.set
) {
4710 _cleanup_free_
char *affinity
= NULL
;
4712 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
4713 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
4716 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
4717 _cleanup_free_
char *nodes
= NULL
;
4719 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
4720 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
4721 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
4724 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4725 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4728 "%sStandardInput: %s\n"
4729 "%sStandardOutput: %s\n"
4730 "%sStandardError: %s\n",
4731 prefix
, exec_input_to_string(c
->std_input
),
4732 prefix
, exec_output_to_string(c
->std_output
),
4733 prefix
, exec_output_to_string(c
->std_error
));
4735 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4736 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4737 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4738 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4739 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4740 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4742 if (c
->std_input
== EXEC_INPUT_FILE
)
4743 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4744 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4745 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4746 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4747 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4748 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4749 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4750 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4751 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4757 "%sTTYVHangup: %s\n"
4758 "%sTTYVTDisallocate: %s\n",
4759 prefix
, c
->tty_path
,
4760 prefix
, yes_no(c
->tty_reset
),
4761 prefix
, yes_no(c
->tty_vhangup
),
4762 prefix
, yes_no(c
->tty_vt_disallocate
));
4764 if (IN_SET(c
->std_output
,
4766 EXEC_OUTPUT_JOURNAL
,
4767 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4768 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4769 IN_SET(c
->std_error
,
4771 EXEC_OUTPUT_JOURNAL
,
4772 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4773 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4775 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4777 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4779 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4781 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4783 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4786 if (c
->log_level_max
>= 0) {
4787 _cleanup_free_
char *t
= NULL
;
4789 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4791 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4794 if (c
->log_ratelimit_interval_usec
> 0) {
4795 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4798 "%sLogRateLimitIntervalSec: %s\n",
4799 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
4802 if (c
->log_ratelimit_burst
> 0)
4803 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
4805 if (c
->n_log_extra_fields
> 0) {
4808 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4809 fprintf(f
, "%sLogExtraFields: ", prefix
);
4810 fwrite(c
->log_extra_fields
[j
].iov_base
,
4811 1, c
->log_extra_fields
[j
].iov_len
,
4817 if (c
->log_namespace
)
4818 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
4820 if (c
->secure_bits
) {
4821 _cleanup_free_
char *str
= NULL
;
4823 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4825 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4828 if (c
->capability_bounding_set
!= CAP_ALL
) {
4829 _cleanup_free_
char *str
= NULL
;
4831 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4833 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4836 if (c
->capability_ambient_set
!= 0) {
4837 _cleanup_free_
char *str
= NULL
;
4839 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4841 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4845 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4847 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4849 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4851 if (!strv_isempty(c
->supplementary_groups
)) {
4852 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4853 strv_fprintf(f
, c
->supplementary_groups
);
4858 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4860 if (!strv_isempty(c
->read_write_paths
)) {
4861 fprintf(f
, "%sReadWritePaths:", prefix
);
4862 strv_fprintf(f
, c
->read_write_paths
);
4866 if (!strv_isempty(c
->read_only_paths
)) {
4867 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4868 strv_fprintf(f
, c
->read_only_paths
);
4872 if (!strv_isempty(c
->inaccessible_paths
)) {
4873 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4874 strv_fprintf(f
, c
->inaccessible_paths
);
4878 if (c
->n_bind_mounts
> 0)
4879 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4880 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4881 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4882 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4883 c
->bind_mounts
[i
].source
,
4884 c
->bind_mounts
[i
].destination
,
4885 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4887 if (c
->n_temporary_filesystems
> 0)
4888 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4889 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4891 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4893 isempty(t
->options
) ? "" : ":",
4894 strempty(t
->options
));
4899 "%sUtmpIdentifier: %s\n",
4900 prefix
, c
->utmp_id
);
4902 if (c
->selinux_context
)
4904 "%sSELinuxContext: %s%s\n",
4905 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4907 if (c
->apparmor_profile
)
4909 "%sAppArmorProfile: %s%s\n",
4910 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4912 if (c
->smack_process_label
)
4914 "%sSmackProcessLabel: %s%s\n",
4915 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4917 if (c
->personality
!= PERSONALITY_INVALID
)
4919 "%sPersonality: %s\n",
4920 prefix
, strna(personality_to_string(c
->personality
)));
4923 "%sLockPersonality: %s\n",
4924 prefix
, yes_no(c
->lock_personality
));
4926 if (c
->syscall_filter
) {
4934 "%sSystemCallFilter: ",
4937 if (!c
->syscall_allow_list
)
4941 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4942 _cleanup_free_
char *name
= NULL
;
4943 const char *errno_name
= NULL
;
4944 int num
= PTR_TO_INT(val
);
4951 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4952 fputs(strna(name
), f
);
4955 errno_name
= errno_to_name(num
);
4957 fprintf(f
, ":%s", errno_name
);
4959 fprintf(f
, ":%d", num
);
4967 if (c
->syscall_archs
) {
4974 "%sSystemCallArchitectures:",
4978 SET_FOREACH(id
, c
->syscall_archs
, j
)
4979 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4984 if (exec_context_restrict_namespaces_set(c
)) {
4985 _cleanup_free_
char *s
= NULL
;
4987 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4989 fprintf(f
, "%sRestrictNamespaces: %s\n",
4993 if (c
->network_namespace_path
)
4995 "%sNetworkNamespacePath: %s\n",
4996 prefix
, c
->network_namespace_path
);
4998 if (c
->syscall_errno
> 0) {
4999 const char *errno_name
;
5001 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
5003 errno_name
= errno_to_name(c
->syscall_errno
);
5005 fprintf(f
, "%s\n", errno_name
);
5007 fprintf(f
, "%d\n", c
->syscall_errno
);
5011 bool exec_context_maintains_privileges(const ExecContext
*c
) {
5014 /* Returns true if the process forked off would run under
5015 * an unchanged UID or as root. */
5020 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
5026 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
5034 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5036 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5041 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5046 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
5047 free(c
->log_extra_fields
[l
].iov_base
);
5048 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5049 c
->n_log_extra_fields
= 0;
5052 void exec_context_revert_tty(ExecContext
*c
) {
5057 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5058 exec_context_tty_reset(c
, NULL
);
5060 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5061 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5062 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5064 if (exec_context_may_touch_tty(c
)) {
5067 path
= exec_context_tty_path(c
);
5069 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
5070 if (r
< 0 && r
!= -ENOENT
)
5071 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5076 int exec_context_get_clean_directories(
5082 _cleanup_strv_free_
char **l
= NULL
;
5083 ExecDirectoryType t
;
5090 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5093 if (!FLAGS_SET(mask
, 1U << t
))
5099 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5102 j
= path_join(prefix
[t
], *i
);
5106 r
= strv_consume(&l
, j
);
5110 /* Also remove private directories unconditionally. */
5111 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5112 j
= path_join(prefix
[t
], "private", *i
);
5116 r
= strv_consume(&l
, j
);
5127 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5128 ExecCleanMask mask
= 0;
5133 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5134 if (!strv_isempty(c
->directories
[t
].paths
))
5141 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5148 dual_timestamp_get(&s
->start_timestamp
);
5151 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5154 if (s
->pid
!= pid
) {
5160 dual_timestamp_get(&s
->exit_timestamp
);
5165 if (context
&& context
->utmp_id
)
5166 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5169 void exec_status_reset(ExecStatus
*s
) {
5172 *s
= (ExecStatus
) {};
5175 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5176 char buf
[FORMAT_TIMESTAMP_MAX
];
5184 prefix
= strempty(prefix
);
5187 "%sPID: "PID_FMT
"\n",
5190 if (dual_timestamp_is_set(&s
->start_timestamp
))
5192 "%sStart Timestamp: %s\n",
5193 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5195 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5197 "%sExit Timestamp: %s\n"
5199 "%sExit Status: %i\n",
5200 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5201 prefix
, sigchld_code_to_string(s
->code
),
5205 static char *exec_command_line(char **argv
) {
5213 STRV_FOREACH(a
, argv
)
5221 STRV_FOREACH(a
, argv
) {
5228 if (strpbrk(*a
, WHITESPACE
)) {
5239 /* FIXME: this doesn't really handle arguments that have
5240 * spaces and ticks in them */
5245 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5246 _cleanup_free_
char *cmd
= NULL
;
5247 const char *prefix2
;
5252 prefix
= strempty(prefix
);
5253 prefix2
= strjoina(prefix
, "\t");
5255 cmd
= exec_command_line(c
->argv
);
5257 "%sCommand Line: %s\n",
5258 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
5260 exec_status_dump(&c
->exec_status
, f
, prefix2
);
5263 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5266 prefix
= strempty(prefix
);
5268 LIST_FOREACH(command
, c
, c
)
5269 exec_command_dump(c
, f
, prefix
);
5272 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
5279 /* It's kind of important, that we keep the order here */
5280 LIST_FIND_TAIL(command
, *l
, end
);
5281 LIST_INSERT_AFTER(command
, *l
, end
, e
);
5286 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
5294 l
= strv_new_ap(path
, ap
);
5306 free_and_replace(c
->path
, p
);
5308 return strv_free_and_replace(c
->argv
, l
);
5311 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
5312 _cleanup_strv_free_
char **l
= NULL
;
5320 l
= strv_new_ap(path
, ap
);
5326 r
= strv_extend_strv(&c
->argv
, l
, false);
5333 static void *remove_tmpdir_thread(void *p
) {
5334 _cleanup_free_
char *path
= p
;
5336 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
5340 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
5347 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
5349 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
5350 if (destroy
&& rt
->tmp_dir
) {
5351 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
5353 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
5355 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
5362 if (destroy
&& rt
->var_tmp_dir
) {
5363 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
5365 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
5367 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
5368 free(rt
->var_tmp_dir
);
5371 rt
->var_tmp_dir
= NULL
;
5374 rt
->id
= mfree(rt
->id
);
5375 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
5376 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
5377 safe_close_pair(rt
->netns_storage_socket
);
5381 static void exec_runtime_freep(ExecRuntime
**rt
) {
5382 (void) exec_runtime_free(*rt
, false);
5385 static int exec_runtime_allocate(ExecRuntime
**ret
) {
5390 n
= new(ExecRuntime
, 1);
5394 *n
= (ExecRuntime
) {
5395 .netns_storage_socket
= { -1, -1 },
5402 static int exec_runtime_add(
5405 const char *tmp_dir
,
5406 const char *var_tmp_dir
,
5407 const int netns_storage_socket
[2],
5408 ExecRuntime
**ret
) {
5410 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
5416 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
5420 r
= exec_runtime_allocate(&rt
);
5424 rt
->id
= strdup(id
);
5429 rt
->tmp_dir
= strdup(tmp_dir
);
5433 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
5434 assert(var_tmp_dir
);
5435 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
5436 if (!rt
->var_tmp_dir
)
5440 if (netns_storage_socket
) {
5441 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
5442 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
5445 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
5454 /* do not remove created ExecRuntime object when the operation succeeds. */
5459 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
5460 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5461 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
5468 /* It is not necessary to create ExecRuntime object. */
5469 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
5472 if (c
->private_tmp
&&
5473 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
5474 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
5475 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
5476 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
5481 if (c
->private_network
|| c
->network_namespace_path
) {
5482 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
5486 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
5491 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
5495 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
5503 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
5505 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5511 /* If not found, then create a new object. */
5512 r
= exec_runtime_make(m
, c
, id
, &rt
);
5514 /* When r == 0, it is not necessary to create ExecRuntime object. */
5518 /* increment reference counter. */
5524 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5528 assert(rt
->n_ref
> 0);
5534 return exec_runtime_free(rt
, destroy
);
5537 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5545 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5546 fprintf(f
, "exec-runtime=%s", rt
->id
);
5549 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5551 if (rt
->var_tmp_dir
)
5552 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5554 if (rt
->netns_storage_socket
[0] >= 0) {
5557 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5561 fprintf(f
, " netns-socket-0=%i", copy
);
5564 if (rt
->netns_storage_socket
[1] >= 0) {
5567 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5571 fprintf(f
, " netns-socket-1=%i", copy
);
5580 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5581 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5585 /* This is for the migration from old (v237 or earlier) deserialization text.
5586 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5587 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5588 * so or not from the serialized text, then we always creates a new object owned by this. */
5594 /* Manager manages ExecRuntime objects by the unit id.
5595 * So, we omit the serialized text when the unit does not have id (yet?)... */
5596 if (isempty(u
->id
)) {
5597 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5601 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5603 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5607 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5609 r
= exec_runtime_allocate(&rt_create
);
5613 rt_create
->id
= strdup(u
->id
);
5620 if (streq(key
, "tmp-dir")) {
5623 copy
= strdup(value
);
5627 free_and_replace(rt
->tmp_dir
, copy
);
5629 } else if (streq(key
, "var-tmp-dir")) {
5632 copy
= strdup(value
);
5636 free_and_replace(rt
->var_tmp_dir
, copy
);
5638 } else if (streq(key
, "netns-socket-0")) {
5641 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5642 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5646 safe_close(rt
->netns_storage_socket
[0]);
5647 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5649 } else if (streq(key
, "netns-socket-1")) {
5652 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5653 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5657 safe_close(rt
->netns_storage_socket
[1]);
5658 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5662 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5664 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5666 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5670 rt_create
->manager
= u
->manager
;
5679 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5680 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5681 int r
, fd0
= -1, fd1
= -1;
5682 const char *p
, *v
= value
;
5689 n
= strcspn(v
, " ");
5690 id
= strndupa(v
, n
);
5695 v
= startswith(p
, "tmp-dir=");
5697 n
= strcspn(v
, " ");
5698 tmp_dir
= strndupa(v
, n
);
5704 v
= startswith(p
, "var-tmp-dir=");
5706 n
= strcspn(v
, " ");
5707 var_tmp_dir
= strndupa(v
, n
);
5713 v
= startswith(p
, "netns-socket-0=");
5717 n
= strcspn(v
, " ");
5718 buf
= strndupa(v
, n
);
5719 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5720 log_debug("Unable to process exec-runtime netns fd specification.");
5723 fd0
= fdset_remove(fds
, fd0
);
5729 v
= startswith(p
, "netns-socket-1=");
5733 n
= strcspn(v
, " ");
5734 buf
= strndupa(v
, n
);
5735 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5736 log_debug("Unable to process exec-runtime netns fd specification.");
5739 fd1
= fdset_remove(fds
, fd1
);
5744 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5746 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5749 void exec_runtime_vacuum(Manager
*m
) {
5755 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5757 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5761 (void) exec_runtime_free(rt
, false);
5765 void exec_params_clear(ExecParameters
*p
) {
5769 strv_free(p
->environment
);
5772 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5773 [EXEC_INPUT_NULL
] = "null",
5774 [EXEC_INPUT_TTY
] = "tty",
5775 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5776 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5777 [EXEC_INPUT_SOCKET
] = "socket",
5778 [EXEC_INPUT_NAMED_FD
] = "fd",
5779 [EXEC_INPUT_DATA
] = "data",
5780 [EXEC_INPUT_FILE
] = "file",
5783 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5785 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5786 [EXEC_OUTPUT_INHERIT
] = "inherit",
5787 [EXEC_OUTPUT_NULL
] = "null",
5788 [EXEC_OUTPUT_TTY
] = "tty",
5789 [EXEC_OUTPUT_KMSG
] = "kmsg",
5790 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5791 [EXEC_OUTPUT_JOURNAL
] = "journal",
5792 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5793 [EXEC_OUTPUT_SOCKET
] = "socket",
5794 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5795 [EXEC_OUTPUT_FILE
] = "file",
5796 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5799 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5801 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5802 [EXEC_UTMP_INIT
] = "init",
5803 [EXEC_UTMP_LOGIN
] = "login",
5804 [EXEC_UTMP_USER
] = "user",
5807 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5809 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5810 [EXEC_PRESERVE_NO
] = "no",
5811 [EXEC_PRESERVE_YES
] = "yes",
5812 [EXEC_PRESERVE_RESTART
] = "restart",
5815 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5817 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
5818 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5819 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5820 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5821 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5822 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5823 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5826 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5828 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
5829 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
5830 * directories, specifically .timer units with their timestamp touch file. */
5831 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5832 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
5833 [EXEC_DIRECTORY_STATE
] = "state",
5834 [EXEC_DIRECTORY_CACHE
] = "cache",
5835 [EXEC_DIRECTORY_LOGS
] = "logs",
5836 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
5839 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
5841 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
5842 * the service payload in. */
5843 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5844 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5845 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5846 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5847 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5848 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5851 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5853 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5854 [EXEC_KEYRING_INHERIT
] = "inherit",
5855 [EXEC_KEYRING_PRIVATE
] = "private",
5856 [EXEC_KEYRING_SHARED
] = "shared",
5859 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);