1 /* SPDX-License-Identifier: LGPL-2.1+ */
6 #include <sys/eventfd.h>
9 #include <sys/personality.h>
10 #include <sys/prctl.h>
12 #include <sys/types.h>
18 #include <security/pam_appl.h>
22 #include <selinux/selinux.h>
30 #include <sys/apparmor.h>
33 #include "sd-messages.h"
36 #include "alloc-util.h"
38 #include "apparmor-util.h"
43 #include "capability-util.h"
44 #include "chown-recursive.h"
45 #include "cgroup-setup.h"
46 #include "cpu-set-util.h"
50 #include "errno-list.h"
52 #include "exit-status.h"
54 #include "format-util.h"
56 #include "glob-util.h"
63 #include "memory-util.h"
64 #include "missing_fs.h"
66 #include "namespace.h"
67 #include "parse-util.h"
68 #include "path-util.h"
69 #include "process-util.h"
70 #include "rlimit-util.h"
73 #include "seccomp-util.h"
75 #include "securebits-util.h"
76 #include "selinux-util.h"
77 #include "signal-util.h"
78 #include "smack-util.h"
79 #include "socket-util.h"
81 #include "stat-util.h"
82 #include "string-table.h"
83 #include "string-util.h"
85 #include "syslog-util.h"
86 #include "terminal-util.h"
87 #include "umask-util.h"
89 #include "user-util.h"
90 #include "utmp-wtmp.h"
92 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
93 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
95 #define SNDBUF_SIZE (8*1024*1024)
97 static int shift_fds(int fds
[], size_t n_fds
) {
98 int start
, restart_from
;
103 /* Modifies the fds array! (sorts it) */
113 for (i
= start
; i
< (int) n_fds
; i
++) {
116 /* Already at right index? */
120 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
127 /* Hmm, the fd we wanted isn't free? Then
128 * let's remember that and try again from here */
129 if (nfd
!= i
+3 && restart_from
< 0)
133 if (restart_from
< 0)
136 start
= restart_from
;
142 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
146 n_fds
= n_socket_fds
+ n_storage_fds
;
152 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
153 * O_NONBLOCK only applies to socket activation though. */
155 for (i
= 0; i
< n_fds
; i
++) {
157 if (i
< n_socket_fds
) {
158 r
= fd_nonblock(fds
[i
], nonblock
);
163 /* We unconditionally drop FD_CLOEXEC from the fds,
164 * since after all we want to pass these fds to our
167 r
= fd_cloexec(fds
[i
], false);
175 static const char *exec_context_tty_path(const ExecContext
*context
) {
178 if (context
->stdio_as_fds
)
181 if (context
->tty_path
)
182 return context
->tty_path
;
184 return "/dev/console";
187 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
192 path
= exec_context_tty_path(context
);
194 if (context
->tty_vhangup
) {
195 if (p
&& p
->stdin_fd
>= 0)
196 (void) terminal_vhangup_fd(p
->stdin_fd
);
198 (void) terminal_vhangup(path
);
201 if (context
->tty_reset
) {
202 if (p
&& p
->stdin_fd
>= 0)
203 (void) reset_terminal_fd(p
->stdin_fd
, true);
205 (void) reset_terminal(path
);
208 if (context
->tty_vt_disallocate
&& path
)
209 (void) vt_disallocate(path
);
212 static bool is_terminal_input(ExecInput i
) {
215 EXEC_INPUT_TTY_FORCE
,
216 EXEC_INPUT_TTY_FAIL
);
219 static bool is_terminal_output(ExecOutput o
) {
222 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
223 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
224 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
227 static bool is_syslog_output(ExecOutput o
) {
230 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
233 static bool is_kmsg_output(ExecOutput o
) {
236 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
239 static bool exec_context_needs_term(const ExecContext
*c
) {
242 /* Return true if the execution context suggests we should set $TERM to something useful. */
244 if (is_terminal_input(c
->std_input
))
247 if (is_terminal_output(c
->std_output
))
250 if (is_terminal_output(c
->std_error
))
253 return !!c
->tty_path
;
256 static int open_null_as(int flags
, int nfd
) {
261 fd
= open("/dev/null", flags
|O_NOCTTY
);
265 return move_fd(fd
, nfd
, false);
268 static int connect_journal_socket(
270 const char *log_namespace
,
274 union sockaddr_union sa
;
276 uid_t olduid
= UID_INVALID
;
277 gid_t oldgid
= GID_INVALID
;
282 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
283 "/run/systemd/journal/stdout";
284 r
= sockaddr_un_set_path(&sa
.un
, j
);
289 if (gid_is_valid(gid
)) {
292 if (setegid(gid
) < 0)
296 if (uid_is_valid(uid
)) {
299 if (seteuid(uid
) < 0) {
305 r
= connect(fd
, &sa
.sa
, sa_len
) < 0 ? -errno
: 0;
307 /* If we fail to restore the uid or gid, things will likely
308 fail later on. This should only happen if an LSM interferes. */
310 if (uid_is_valid(uid
))
311 (void) seteuid(olduid
);
314 if (gid_is_valid(gid
))
315 (void) setegid(oldgid
);
320 static int connect_logger_as(
322 const ExecContext
*context
,
323 const ExecParameters
*params
,
330 _cleanup_close_
int fd
= -1;
335 assert(output
< _EXEC_OUTPUT_MAX
);
339 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
343 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
347 if (shutdown(fd
, SHUT_RD
) < 0)
350 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
360 context
->syslog_identifier
?: ident
,
361 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
362 context
->syslog_priority
,
363 !!context
->syslog_level_prefix
,
364 is_syslog_output(output
),
365 is_kmsg_output(output
),
366 is_terminal_output(output
)) < 0)
369 return move_fd(TAKE_FD(fd
), nfd
, false);
372 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
378 fd
= open_terminal(path
, flags
| O_NOCTTY
);
382 return move_fd(fd
, nfd
, false);
385 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
386 union sockaddr_union sa
;
388 _cleanup_close_
int fd
= -1;
393 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
396 fd
= open(path
, flags
|O_NOCTTY
, mode
);
400 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
403 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
405 r
= sockaddr_un_set_path(&sa
.un
, path
);
407 return r
== -EINVAL
? -ENXIO
: r
;
410 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
414 if (connect(fd
, &sa
.sa
, sa_len
) < 0)
415 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
416 * indication that his wasn't an AF_UNIX socket after all */
418 if ((flags
& O_ACCMODE
) == O_RDONLY
)
419 r
= shutdown(fd
, SHUT_WR
);
420 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
421 r
= shutdown(fd
, SHUT_RD
);
430 static int fixup_input(
431 const ExecContext
*context
,
433 bool apply_tty_stdin
) {
439 std_input
= context
->std_input
;
441 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
442 return EXEC_INPUT_NULL
;
444 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
445 return EXEC_INPUT_NULL
;
447 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
448 return EXEC_INPUT_NULL
;
453 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
455 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
456 return EXEC_OUTPUT_INHERIT
;
461 static int setup_input(
462 const ExecContext
*context
,
463 const ExecParameters
*params
,
465 const int named_iofds
[static 3]) {
473 if (params
->stdin_fd
>= 0) {
474 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
477 /* Try to make this the controlling tty, if it is a tty, and reset it */
478 if (isatty(STDIN_FILENO
)) {
479 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
480 (void) reset_terminal_fd(STDIN_FILENO
, true);
486 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
490 case EXEC_INPUT_NULL
:
491 return open_null_as(O_RDONLY
, STDIN_FILENO
);
494 case EXEC_INPUT_TTY_FORCE
:
495 case EXEC_INPUT_TTY_FAIL
: {
498 fd
= acquire_terminal(exec_context_tty_path(context
),
499 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
500 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
501 ACQUIRE_TERMINAL_WAIT
,
506 return move_fd(fd
, STDIN_FILENO
, false);
509 case EXEC_INPUT_SOCKET
:
510 assert(socket_fd
>= 0);
512 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
514 case EXEC_INPUT_NAMED_FD
:
515 assert(named_iofds
[STDIN_FILENO
] >= 0);
517 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
518 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
520 case EXEC_INPUT_DATA
: {
523 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
527 return move_fd(fd
, STDIN_FILENO
, false);
530 case EXEC_INPUT_FILE
: {
534 assert(context
->stdio_file
[STDIN_FILENO
]);
536 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
537 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
539 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
543 return move_fd(fd
, STDIN_FILENO
, false);
547 assert_not_reached("Unknown input type");
551 static bool can_inherit_stderr_from_stdout(
552 const ExecContext
*context
,
558 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
561 if (e
== EXEC_OUTPUT_INHERIT
)
566 if (e
== EXEC_OUTPUT_NAMED_FD
)
567 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
569 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
570 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
575 static int setup_output(
577 const ExecContext
*context
,
578 const ExecParameters
*params
,
581 const int named_iofds
[static 3],
585 dev_t
*journal_stream_dev
,
586 ino_t
*journal_stream_ino
) {
596 assert(journal_stream_dev
);
597 assert(journal_stream_ino
);
599 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
601 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
604 return STDOUT_FILENO
;
607 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
608 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
611 return STDERR_FILENO
;
614 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
615 o
= fixup_output(context
->std_output
, socket_fd
);
617 if (fileno
== STDERR_FILENO
) {
619 e
= fixup_output(context
->std_error
, socket_fd
);
621 /* This expects the input and output are already set up */
623 /* Don't change the stderr file descriptor if we inherit all
624 * the way and are not on a tty */
625 if (e
== EXEC_OUTPUT_INHERIT
&&
626 o
== EXEC_OUTPUT_INHERIT
&&
627 i
== EXEC_INPUT_NULL
&&
628 !is_terminal_input(context
->std_input
) &&
632 /* Duplicate from stdout if possible */
633 if (can_inherit_stderr_from_stdout(context
, o
, e
))
634 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
638 } else if (o
== EXEC_OUTPUT_INHERIT
) {
639 /* If input got downgraded, inherit the original value */
640 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
641 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
643 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
644 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
645 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
647 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
651 /* We need to open /dev/null here anew, to get the right access mode. */
652 return open_null_as(O_WRONLY
, fileno
);
657 case EXEC_OUTPUT_NULL
:
658 return open_null_as(O_WRONLY
, fileno
);
660 case EXEC_OUTPUT_TTY
:
661 if (is_terminal_input(i
))
662 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
664 /* We don't reset the terminal if this is just about output */
665 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
667 case EXEC_OUTPUT_SYSLOG
:
668 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
669 case EXEC_OUTPUT_KMSG
:
670 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
671 case EXEC_OUTPUT_JOURNAL
:
672 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
673 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
675 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
676 r
= open_null_as(O_WRONLY
, fileno
);
680 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
681 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
682 * services to detect whether they are connected to the journal or not.
684 * If both stdout and stderr are connected to a stream then let's make sure to store the data
685 * about STDERR as that's usually the best way to do logging. */
687 if (fstat(fileno
, &st
) >= 0 &&
688 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
689 *journal_stream_dev
= st
.st_dev
;
690 *journal_stream_ino
= st
.st_ino
;
695 case EXEC_OUTPUT_SOCKET
:
696 assert(socket_fd
>= 0);
698 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
700 case EXEC_OUTPUT_NAMED_FD
:
701 assert(named_iofds
[fileno
] >= 0);
703 (void) fd_nonblock(named_iofds
[fileno
], false);
704 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
706 case EXEC_OUTPUT_FILE
:
707 case EXEC_OUTPUT_FILE_APPEND
: {
711 assert(context
->stdio_file
[fileno
]);
713 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
714 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
717 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
720 if (o
== EXEC_OUTPUT_FILE_APPEND
)
723 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
727 return move_fd(fd
, fileno
, 0);
731 assert_not_reached("Unknown error type");
735 static int chown_terminal(int fd
, uid_t uid
) {
740 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
741 if (isatty(fd
) < 1) {
742 if (IN_SET(errno
, EINVAL
, ENOTTY
))
743 return 0; /* not a tty */
748 /* This might fail. What matters are the results. */
749 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, -1);
756 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
757 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
760 assert(_saved_stdin
);
761 assert(_saved_stdout
);
763 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
767 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
768 if (saved_stdout
< 0)
771 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
775 r
= chown_terminal(fd
, getuid());
779 r
= reset_terminal_fd(fd
, true);
783 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
788 *_saved_stdin
= saved_stdin
;
789 *_saved_stdout
= saved_stdout
;
791 saved_stdin
= saved_stdout
= -1;
796 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
799 if (err
== -ETIMEDOUT
)
800 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
803 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
807 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
808 _cleanup_close_
int fd
= -1;
812 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
816 write_confirm_error_fd(err
, fd
, u
);
819 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
823 assert(saved_stdout
);
827 if (*saved_stdin
>= 0)
828 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
831 if (*saved_stdout
>= 0)
832 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
835 *saved_stdin
= safe_close(*saved_stdin
);
836 *saved_stdout
= safe_close(*saved_stdout
);
842 CONFIRM_PRETEND_FAILURE
= -1,
843 CONFIRM_PRETEND_SUCCESS
= 0,
847 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
848 int saved_stdout
= -1, saved_stdin
= -1, r
;
849 _cleanup_free_
char *e
= NULL
;
852 /* For any internal errors, assume a positive response. */
853 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
855 write_confirm_error(r
, vc
, u
);
856 return CONFIRM_EXECUTE
;
859 /* confirm_spawn might have been disabled while we were sleeping. */
860 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
865 e
= ellipsize(cmdline
, 60, 100);
873 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
875 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
882 printf("Resuming normal execution.\n");
883 manager_disable_confirm_spawn();
887 unit_dump(u
, stdout
, " ");
888 continue; /* ask again */
890 printf("Failing execution.\n");
891 r
= CONFIRM_PRETEND_FAILURE
;
894 printf(" c - continue, proceed without asking anymore\n"
895 " D - dump, show the state of the unit\n"
896 " f - fail, don't execute the command and pretend it failed\n"
898 " i - info, show a short summary of the unit\n"
899 " j - jobs, show jobs that are in progress\n"
900 " s - skip, don't execute the command and pretend it succeeded\n"
901 " y - yes, execute the command\n");
902 continue; /* ask again */
904 printf(" Description: %s\n"
907 u
->id
, u
->description
, cmdline
);
908 continue; /* ask again */
910 manager_dump_jobs(u
->manager
, stdout
, " ");
911 continue; /* ask again */
913 /* 'n' was removed in favor of 'f'. */
914 printf("Didn't understand 'n', did you mean 'f'?\n");
915 continue; /* ask again */
917 printf("Skipping execution.\n");
918 r
= CONFIRM_PRETEND_SUCCESS
;
924 assert_not_reached("Unhandled choice");
930 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
934 static int get_fixed_user(const ExecContext
*c
, const char **user
,
935 uid_t
*uid
, gid_t
*gid
,
936 const char **home
, const char **shell
) {
945 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
946 * (i.e. are "/" or "/bin/nologin"). */
949 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
957 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
967 r
= get_group_creds(&name
, gid
, 0);
975 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
976 const char *group
, gid_t gid
,
977 gid_t
**supplementary_gids
, int *ngids
) {
981 bool keep_groups
= false;
982 gid_t
*groups
= NULL
;
983 _cleanup_free_ gid_t
*l_gids
= NULL
;
988 * If user is given, then lookup GID and supplementary groups list.
989 * We avoid NSS lookups for gid=0. Also we have to initialize groups
990 * here and as early as possible so we keep the list of supplementary
991 * groups of the caller.
993 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
994 /* First step, initialize groups from /etc/groups */
995 if (initgroups(user
, gid
) < 0)
1001 if (strv_isempty(c
->supplementary_groups
))
1005 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1006 * be positive, otherwise fail.
1009 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1010 if (ngroups_max
<= 0)
1011 return errno_or_else(EOPNOTSUPP
);
1013 l_gids
= new(gid_t
, ngroups_max
);
1019 * Lookup the list of groups that the user belongs to, we
1020 * avoid NSS lookups here too for gid=0.
1023 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1028 STRV_FOREACH(i
, c
->supplementary_groups
) {
1031 if (k
>= ngroups_max
)
1035 r
= get_group_creds(&g
, l_gids
+k
, 0);
1043 * Sets ngids to zero to drop all supplementary groups, happens
1044 * when we are under root and SupplementaryGroups= is empty.
1051 /* Otherwise get the final list of supplementary groups */
1052 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1056 *supplementary_gids
= groups
;
1064 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1067 /* Handle SupplementaryGroups= if it is not empty */
1069 r
= maybe_setgroups(ngids
, supplementary_gids
);
1074 if (gid_is_valid(gid
)) {
1075 /* Then set our gids */
1076 if (setresgid(gid
, gid
, gid
) < 0)
1083 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1086 if (!uid_is_valid(uid
))
1089 /* Sets (but doesn't look up) the uid and make sure we keep the
1090 * capabilities while doing so. */
1092 if (context
->capability_ambient_set
!= 0) {
1094 /* First step: If we need to keep capabilities but
1095 * drop privileges we need to make sure we keep our
1096 * caps, while we drop privileges. */
1098 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1100 if (prctl(PR_GET_SECUREBITS
) != sb
)
1101 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1106 /* Second step: actually set the uids */
1107 if (setresuid(uid
, uid
, uid
) < 0)
1110 /* At this point we should have all necessary capabilities but
1111 are otherwise a normal user. However, the caps might got
1112 corrupted due to the setresuid() so we need clean them up
1113 later. This is done outside of this call. */
1120 static int null_conv(
1122 const struct pam_message
**msg
,
1123 struct pam_response
**resp
,
1124 void *appdata_ptr
) {
1126 /* We don't support conversations */
1128 return PAM_CONV_ERR
;
1133 static int setup_pam(
1140 const int fds
[], size_t n_fds
) {
1144 static const struct pam_conv conv
= {
1149 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1150 pam_handle_t
*handle
= NULL
;
1152 int pam_code
= PAM_SUCCESS
, r
;
1153 char **nv
, **e
= NULL
;
1154 bool close_session
= false;
1155 pid_t pam_pid
= 0, parent_pid
;
1162 /* We set up PAM in the parent process, then fork. The child
1163 * will then stay around until killed via PR_GET_PDEATHSIG or
1164 * systemd via the cgroup logic. It will then remove the PAM
1165 * session again. The parent process will exec() the actual
1166 * daemon. We do things this way to ensure that the main PID
1167 * of the daemon is the one we initially fork()ed. */
1169 r
= barrier_create(&barrier
);
1173 if (log_get_max_level() < LOG_DEBUG
)
1174 flags
|= PAM_SILENT
;
1176 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1177 if (pam_code
!= PAM_SUCCESS
) {
1183 _cleanup_free_
char *q
= NULL
;
1185 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1186 * out if that's the case, and read the TTY off it. */
1188 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1189 tty
= strjoina("/dev/", q
);
1193 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1194 if (pam_code
!= PAM_SUCCESS
)
1198 STRV_FOREACH(nv
, *env
) {
1199 pam_code
= pam_putenv(handle
, *nv
);
1200 if (pam_code
!= PAM_SUCCESS
)
1204 pam_code
= pam_acct_mgmt(handle
, flags
);
1205 if (pam_code
!= PAM_SUCCESS
)
1208 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1209 if (pam_code
!= PAM_SUCCESS
)
1210 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1212 pam_code
= pam_open_session(handle
, flags
);
1213 if (pam_code
!= PAM_SUCCESS
)
1216 close_session
= true;
1218 e
= pam_getenvlist(handle
);
1220 pam_code
= PAM_BUF_ERR
;
1224 /* Block SIGTERM, so that we know that it won't get lost in
1227 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1229 parent_pid
= getpid_cached();
1231 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1235 int sig
, ret
= EXIT_PAM
;
1237 /* The child's job is to reset the PAM session on
1239 barrier_set_role(&barrier
, BARRIER_CHILD
);
1241 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1242 * are open here that have been opened by PAM. */
1243 (void) close_many(fds
, n_fds
);
1245 /* Drop privileges - we don't need any to pam_close_session
1246 * and this will make PR_SET_PDEATHSIG work in most cases.
1247 * If this fails, ignore the error - but expect sd-pam threads
1248 * to fail to exit normally */
1250 r
= maybe_setgroups(0, NULL
);
1252 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1253 if (setresgid(gid
, gid
, gid
) < 0)
1254 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1255 if (setresuid(uid
, uid
, uid
) < 0)
1256 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1258 (void) ignore_signals(SIGPIPE
, -1);
1260 /* Wait until our parent died. This will only work if
1261 * the above setresuid() succeeds, otherwise the kernel
1262 * will not allow unprivileged parents kill their privileged
1263 * children this way. We rely on the control groups kill logic
1264 * to do the rest for us. */
1265 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1268 /* Tell the parent that our setup is done. This is especially
1269 * important regarding dropping privileges. Otherwise, unit
1270 * setup might race against our setresuid(2) call.
1272 * If the parent aborted, we'll detect this below, hence ignore
1273 * return failure here. */
1274 (void) barrier_place(&barrier
);
1276 /* Check if our parent process might already have died? */
1277 if (getppid() == parent_pid
) {
1280 assert_se(sigemptyset(&ss
) >= 0);
1281 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1284 if (sigwait(&ss
, &sig
) < 0) {
1291 assert(sig
== SIGTERM
);
1296 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1297 if (pam_code
!= PAM_SUCCESS
)
1300 /* If our parent died we'll end the session */
1301 if (getppid() != parent_pid
) {
1302 pam_code
= pam_close_session(handle
, flags
);
1303 if (pam_code
!= PAM_SUCCESS
)
1310 pam_end(handle
, pam_code
| flags
);
1314 barrier_set_role(&barrier
, BARRIER_PARENT
);
1316 /* If the child was forked off successfully it will do all the
1317 * cleanups, so forget about the handle here. */
1320 /* Unblock SIGTERM again in the parent */
1321 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1323 /* We close the log explicitly here, since the PAM modules
1324 * might have opened it, but we don't want this fd around. */
1327 /* Synchronously wait for the child to initialize. We don't care for
1328 * errors as we cannot recover. However, warn loudly if it happens. */
1329 if (!barrier_place_and_sync(&barrier
))
1330 log_error("PAM initialization failed");
1332 return strv_free_and_replace(*env
, e
);
1335 if (pam_code
!= PAM_SUCCESS
) {
1336 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1337 r
= -EPERM
; /* PAM errors do not map to errno */
1339 log_error_errno(r
, "PAM failed: %m");
1343 pam_code
= pam_close_session(handle
, flags
);
1345 pam_end(handle
, pam_code
| flags
);
1357 static void rename_process_from_path(const char *path
) {
1358 char process_name
[11];
1362 /* This resulting string must fit in 10 chars (i.e. the length
1363 * of "/sbin/init") to look pretty in /bin/ps */
1367 rename_process("(...)");
1373 /* The end of the process name is usually more
1374 * interesting, since the first bit might just be
1380 process_name
[0] = '(';
1381 memcpy(process_name
+1, p
, l
);
1382 process_name
[1+l
] = ')';
1383 process_name
[1+l
+1] = 0;
1385 rename_process(process_name
);
1388 static bool context_has_address_families(const ExecContext
*c
) {
1391 return c
->address_families_whitelist
||
1392 !set_isempty(c
->address_families
);
1395 static bool context_has_syscall_filters(const ExecContext
*c
) {
1398 return c
->syscall_whitelist
||
1399 !hashmap_isempty(c
->syscall_filter
);
1402 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1405 if (c
->no_new_privileges
)
1408 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1411 /* We need NNP if we have any form of seccomp and are unprivileged */
1412 return context_has_address_families(c
) ||
1413 c
->memory_deny_write_execute
||
1414 c
->restrict_realtime
||
1415 c
->restrict_suid_sgid
||
1416 exec_context_restrict_namespaces_set(c
) ||
1418 c
->protect_kernel_tunables
||
1419 c
->protect_kernel_modules
||
1420 c
->protect_kernel_logs
||
1421 c
->private_devices
||
1422 context_has_syscall_filters(c
) ||
1423 !set_isempty(c
->syscall_archs
) ||
1424 c
->lock_personality
||
1425 c
->protect_hostname
;
1430 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1432 if (is_seccomp_available())
1435 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1439 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1440 uint32_t negative_action
, default_action
, action
;
1446 if (!context_has_syscall_filters(c
))
1449 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1452 negative_action
= c
->syscall_errno
== 0 ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1454 if (c
->syscall_whitelist
) {
1455 default_action
= negative_action
;
1456 action
= SCMP_ACT_ALLOW
;
1458 default_action
= SCMP_ACT_ALLOW
;
1459 action
= negative_action
;
1462 if (needs_ambient_hack
) {
1463 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1468 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1471 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1475 if (set_isempty(c
->syscall_archs
))
1478 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1481 return seccomp_restrict_archs(c
->syscall_archs
);
1484 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1488 if (!context_has_address_families(c
))
1491 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1494 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1497 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1501 if (!c
->memory_deny_write_execute
)
1504 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1507 return seccomp_memory_deny_write_execute();
1510 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1514 if (!c
->restrict_realtime
)
1517 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1520 return seccomp_restrict_realtime();
1523 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1527 if (!c
->restrict_suid_sgid
)
1530 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1533 return seccomp_restrict_suid_sgid();
1536 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1540 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1541 * let's protect even those systems where this is left on in the kernel. */
1543 if (!c
->protect_kernel_tunables
)
1546 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1549 return seccomp_protect_sysctl();
1552 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1556 /* Turn off module syscalls on ProtectKernelModules=yes */
1558 if (!c
->protect_kernel_modules
)
1561 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1564 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1567 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1571 if (!c
->protect_kernel_logs
)
1574 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1577 return seccomp_protect_syslog();
1580 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1584 if (!c
->protect_clock
)
1587 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1590 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1593 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1597 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1599 if (!c
->private_devices
)
1602 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1605 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1608 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1612 if (!exec_context_restrict_namespaces_set(c
))
1615 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1618 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1621 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1622 unsigned long personality
;
1628 if (!c
->lock_personality
)
1631 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1634 personality
= c
->personality
;
1636 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1637 if (personality
== PERSONALITY_INVALID
) {
1639 r
= opinionated_personality(&personality
);
1644 return seccomp_lock_personality(personality
);
1649 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1653 if (!c
->protect_hostname
)
1656 if (ns_type_supported(NAMESPACE_UTS
)) {
1657 if (unshare(CLONE_NEWUTS
) < 0) {
1658 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1659 *ret_exit_status
= EXIT_NAMESPACE
;
1660 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1663 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1666 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1671 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1674 r
= seccomp_protect_hostname();
1676 *ret_exit_status
= EXIT_SECCOMP
;
1677 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1684 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1687 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1688 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1690 if (idle_pipe
[0] >= 0) {
1693 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1695 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1698 /* Signal systemd that we are bored and want to continue. */
1699 n
= write(idle_pipe
[3], "x", 1);
1701 /* Wait for systemd to react to the signal above. */
1702 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1705 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1709 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1712 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1714 static int build_environment(
1716 const ExecContext
*c
,
1717 const ExecParameters
*p
,
1720 const char *username
,
1722 dev_t journal_stream_dev
,
1723 ino_t journal_stream_ino
,
1726 _cleanup_strv_free_
char **our_env
= NULL
;
1727 ExecDirectoryType t
;
1736 our_env
= new0(char*, 15 + _EXEC_DIRECTORY_TYPE_MAX
);
1741 _cleanup_free_
char *joined
= NULL
;
1743 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1745 our_env
[n_env
++] = x
;
1747 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1749 our_env
[n_env
++] = x
;
1751 joined
= strv_join(p
->fd_names
, ":");
1755 x
= strjoin("LISTEN_FDNAMES=", joined
);
1758 our_env
[n_env
++] = x
;
1761 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1762 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1764 our_env
[n_env
++] = x
;
1766 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1768 our_env
[n_env
++] = x
;
1771 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1772 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1773 * check the database directly. */
1774 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1775 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1778 our_env
[n_env
++] = x
;
1782 x
= strjoin("HOME=", home
);
1786 path_simplify(x
+ 5, true);
1787 our_env
[n_env
++] = x
;
1791 x
= strjoin("LOGNAME=", username
);
1794 our_env
[n_env
++] = x
;
1796 x
= strjoin("USER=", username
);
1799 our_env
[n_env
++] = x
;
1803 x
= strjoin("SHELL=", shell
);
1807 path_simplify(x
+ 6, true);
1808 our_env
[n_env
++] = x
;
1811 if (!sd_id128_is_null(u
->invocation_id
)) {
1812 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1815 our_env
[n_env
++] = x
;
1818 if (exec_context_needs_term(c
)) {
1819 const char *tty_path
, *term
= NULL
;
1821 tty_path
= exec_context_tty_path(c
);
1823 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1824 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1825 * container manager passes to PID 1 ends up all the way in the console login shown. */
1827 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1828 term
= getenv("TERM");
1831 term
= default_term_for_tty(tty_path
);
1833 x
= strjoin("TERM=", term
);
1836 our_env
[n_env
++] = x
;
1839 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1840 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1843 our_env
[n_env
++] = x
;
1846 if (c
->log_namespace
) {
1847 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1851 our_env
[n_env
++] = x
;
1854 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1855 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1861 if (strv_isempty(c
->directories
[t
].paths
))
1864 n
= exec_directory_env_name_to_string(t
);
1868 pre
= strjoin(p
->prefix
[t
], "/");
1872 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1876 x
= strjoin(n
, "=", joined
);
1880 our_env
[n_env
++] = x
;
1883 our_env
[n_env
++] = NULL
;
1884 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1886 *ret
= TAKE_PTR(our_env
);
1891 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1892 _cleanup_strv_free_
char **pass_env
= NULL
;
1893 size_t n_env
= 0, n_bufsize
= 0;
1896 STRV_FOREACH(i
, c
->pass_environment
) {
1897 _cleanup_free_
char *x
= NULL
;
1903 x
= strjoin(*i
, "=", v
);
1907 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1910 pass_env
[n_env
++] = TAKE_PTR(x
);
1911 pass_env
[n_env
] = NULL
;
1914 *ret
= TAKE_PTR(pass_env
);
1919 static bool exec_needs_mount_namespace(
1920 const ExecContext
*context
,
1921 const ExecParameters
*params
,
1922 const ExecRuntime
*runtime
) {
1927 if (context
->root_image
)
1930 if (!strv_isempty(context
->read_write_paths
) ||
1931 !strv_isempty(context
->read_only_paths
) ||
1932 !strv_isempty(context
->inaccessible_paths
))
1935 if (context
->n_bind_mounts
> 0)
1938 if (context
->n_temporary_filesystems
> 0)
1941 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
1944 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1947 if (context
->private_devices
||
1948 context
->private_mounts
||
1949 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1950 context
->protect_home
!= PROTECT_HOME_NO
||
1951 context
->protect_kernel_tunables
||
1952 context
->protect_kernel_modules
||
1953 context
->protect_kernel_logs
||
1954 context
->protect_control_groups
)
1957 if (context
->root_directory
) {
1958 ExecDirectoryType t
;
1960 if (context
->mount_apivfs
)
1963 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1964 if (!params
->prefix
[t
])
1967 if (!strv_isempty(context
->directories
[t
].paths
))
1972 if (context
->dynamic_user
&&
1973 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1974 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1975 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1978 if (context
->log_namespace
)
1984 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
1985 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1986 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1987 _cleanup_close_
int unshare_ready_fd
= -1;
1988 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1993 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
1994 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
1995 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1996 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1997 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1998 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1999 * continues execution normally.
2000 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2001 * does not need CAP_SETUID to write the single line mapping to itself. */
2003 /* Can only set up multiple mappings with CAP_SETUID. */
2004 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
2005 r
= asprintf(&uid_map
,
2006 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2007 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2008 ouid
, ouid
, uid
, uid
);
2010 r
= asprintf(&uid_map
,
2011 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2017 /* Can only set up multiple mappings with CAP_SETGID. */
2018 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2019 r
= asprintf(&gid_map
,
2020 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2021 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2022 ogid
, ogid
, gid
, gid
);
2024 r
= asprintf(&gid_map
,
2025 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2031 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2033 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2034 if (unshare_ready_fd
< 0)
2037 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2039 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2042 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2046 _cleanup_close_
int fd
= -1;
2050 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2051 * here, after the parent opened its own user namespace. */
2054 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2056 /* Wait until the parent unshared the user namespace */
2057 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2062 /* Disable the setgroups() system call in the child user namespace, for good. */
2063 a
= procfs_file_alloca(ppid
, "setgroups");
2064 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2066 if (errno
!= ENOENT
) {
2071 /* If the file is missing the kernel is too old, let's continue anyway. */
2073 if (write(fd
, "deny\n", 5) < 0) {
2078 fd
= safe_close(fd
);
2081 /* First write the GID map */
2082 a
= procfs_file_alloca(ppid
, "gid_map");
2083 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2088 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2092 fd
= safe_close(fd
);
2094 /* The write the UID map */
2095 a
= procfs_file_alloca(ppid
, "uid_map");
2096 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2101 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2106 _exit(EXIT_SUCCESS
);
2109 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2110 _exit(EXIT_FAILURE
);
2113 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2115 if (unshare(CLONE_NEWUSER
) < 0)
2118 /* Let the child know that the namespace is ready now */
2119 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2122 /* Try to read an error code from the child */
2123 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2126 if (n
== sizeof(r
)) { /* an error code was sent to us */
2131 if (n
!= 0) /* on success we should have read 0 bytes */
2134 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2138 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2144 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2145 if (!context
->dynamic_user
)
2148 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2151 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2157 static int setup_exec_directory(
2158 const ExecContext
*context
,
2159 const ExecParameters
*params
,
2162 ExecDirectoryType type
,
2165 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2166 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2167 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2168 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2169 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2170 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2177 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2178 assert(exit_status
);
2180 if (!params
->prefix
[type
])
2183 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2184 if (!uid_is_valid(uid
))
2186 if (!gid_is_valid(gid
))
2190 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2191 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2193 p
= path_join(params
->prefix
[type
], *rt
);
2199 r
= mkdir_parents_label(p
, 0755);
2203 if (exec_directory_is_private(context
, type
)) {
2204 _cleanup_free_
char *private_root
= NULL
;
2206 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2207 * case we want to avoid leaving a directory around fully accessible that is owned by
2208 * a dynamic user whose UID is later on reused. To lock this down we use the same
2209 * trick used by container managers to prohibit host users to get access to files of
2210 * the same UID in containers: we place everything inside a directory that has an
2211 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2212 * for unprivileged host code. We then use fs namespacing to make this directory
2213 * permeable for the service itself.
2215 * Specifically: for a service which wants a special directory "foo/" we first create
2216 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2217 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2218 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2219 * unprivileged host users can't look into it. Inside of the namespace of the unit
2220 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2221 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2222 * for the service and making sure it only gets access to the dirs it needs but no
2223 * others. Tricky? Yes, absolutely, but it works!
2225 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2226 * to be owned by the service itself.
2228 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2229 * for sharing files or sockets with other services. */
2231 private_root
= path_join(params
->prefix
[type
], "private");
2232 if (!private_root
) {
2237 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2238 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2242 pp
= path_join(private_root
, *rt
);
2248 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2249 r
= mkdir_parents_label(pp
, 0755);
2253 if (is_dir(p
, false) > 0 &&
2254 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2256 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2257 * it over. Most likely the service has been upgraded from one that didn't use
2258 * DynamicUser=1, to one that does. */
2260 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2261 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2262 exec_directory_type_to_string(type
), p
, pp
);
2264 if (rename(p
, pp
) < 0) {
2269 /* Otherwise, create the actual directory for the service */
2271 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2272 if (r
< 0 && r
!= -EEXIST
)
2276 /* And link it up from the original place */
2277 r
= symlink_idempotent(pp
, p
, true);
2282 _cleanup_free_
char *target
= NULL
;
2284 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2285 readlink_and_make_absolute(p
, &target
) >= 0) {
2286 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2288 /* This already exists and is a symlink? Interesting. Maybe it's one created
2289 * by DynamicUser=1 (see above)?
2291 * We do this for all directory types except for ConfigurationDirectory=,
2292 * since they all support the private/ symlink logic at least in some
2293 * configurations, see above. */
2295 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2299 q
= path_join(params
->prefix
[type
], "private", *rt
);
2305 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2306 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2310 if (path_equal(q_resolved
, target_resolved
)) {
2312 /* Hmm, apparently DynamicUser= was once turned on for this service,
2313 * but is no longer. Let's move the directory back up. */
2315 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2316 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2317 exec_directory_type_to_string(type
), q
, p
);
2319 if (unlink(p
) < 0) {
2324 if (rename(q
, p
) < 0) {
2331 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2336 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2339 /* Don't change the owner/access mode of the configuration directory,
2340 * as in the common case it is not written to by a service, and shall
2341 * not be writable. */
2343 if (stat(p
, &st
) < 0) {
2348 /* Still complain if the access mode doesn't match */
2349 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2350 log_warning("%s \'%s\' already exists but the mode is different. "
2351 "(File system: %o %sMode: %o)",
2352 exec_directory_type_to_string(type
), *rt
,
2353 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2360 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2361 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2362 * current UID/GID ownership.) */
2363 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2367 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2368 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2369 * assignments to exist.*/
2370 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2378 *exit_status
= exit_status_table
[type
];
2383 static int setup_smack(
2384 const ExecContext
*context
,
2385 const ExecCommand
*command
) {
2392 if (context
->smack_process_label
) {
2393 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2397 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2399 _cleanup_free_
char *exec_label
= NULL
;
2401 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2402 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2405 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2415 static int compile_bind_mounts(
2416 const ExecContext
*context
,
2417 const ExecParameters
*params
,
2418 BindMount
**ret_bind_mounts
,
2419 size_t *ret_n_bind_mounts
,
2420 char ***ret_empty_directories
) {
2422 _cleanup_strv_free_
char **empty_directories
= NULL
;
2423 BindMount
*bind_mounts
;
2425 ExecDirectoryType t
;
2430 assert(ret_bind_mounts
);
2431 assert(ret_n_bind_mounts
);
2432 assert(ret_empty_directories
);
2434 n
= context
->n_bind_mounts
;
2435 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2436 if (!params
->prefix
[t
])
2439 n
+= strv_length(context
->directories
[t
].paths
);
2443 *ret_bind_mounts
= NULL
;
2444 *ret_n_bind_mounts
= 0;
2445 *ret_empty_directories
= NULL
;
2449 bind_mounts
= new(BindMount
, n
);
2453 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2454 BindMount
*item
= context
->bind_mounts
+ i
;
2457 s
= strdup(item
->source
);
2463 d
= strdup(item
->destination
);
2470 bind_mounts
[h
++] = (BindMount
) {
2473 .read_only
= item
->read_only
,
2474 .recursive
= item
->recursive
,
2475 .ignore_enoent
= item
->ignore_enoent
,
2479 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2482 if (!params
->prefix
[t
])
2485 if (strv_isempty(context
->directories
[t
].paths
))
2488 if (exec_directory_is_private(context
, t
) &&
2489 !(context
->root_directory
|| context
->root_image
)) {
2492 /* So this is for a dynamic user, and we need to make sure the process can access its own
2493 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2494 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2496 private_root
= path_join(params
->prefix
[t
], "private");
2497 if (!private_root
) {
2502 r
= strv_consume(&empty_directories
, private_root
);
2507 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2510 if (exec_directory_is_private(context
, t
))
2511 s
= path_join(params
->prefix
[t
], "private", *suffix
);
2513 s
= path_join(params
->prefix
[t
], *suffix
);
2519 if (exec_directory_is_private(context
, t
) &&
2520 (context
->root_directory
|| context
->root_image
))
2521 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2522 * directory is not created on the root directory. So, let's bind-mount the directory
2523 * on the 'non-private' place. */
2524 d
= path_join(params
->prefix
[t
], *suffix
);
2533 bind_mounts
[h
++] = (BindMount
) {
2537 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
2539 .ignore_enoent
= false,
2546 *ret_bind_mounts
= bind_mounts
;
2547 *ret_n_bind_mounts
= n
;
2548 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2553 bind_mount_free_many(bind_mounts
, h
);
2557 static bool insist_on_sandboxing(
2558 const ExecContext
*context
,
2559 const char *root_dir
,
2560 const char *root_image
,
2561 const BindMount
*bind_mounts
,
2562 size_t n_bind_mounts
) {
2567 assert(n_bind_mounts
== 0 || bind_mounts
);
2569 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
2570 * would alter the view on the file system beyond making things read-only or invisble, i.e. would
2571 * rearrange stuff in a way we cannot ignore gracefully. */
2573 if (context
->n_temporary_filesystems
> 0)
2576 if (root_dir
|| root_image
)
2579 if (context
->dynamic_user
)
2582 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
2584 for (i
= 0; i
< n_bind_mounts
; i
++)
2585 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
2588 if (context
->log_namespace
)
2594 static int apply_mount_namespace(
2596 const ExecCommand
*command
,
2597 const ExecContext
*context
,
2598 const ExecParameters
*params
,
2599 const ExecRuntime
*runtime
,
2600 char **error_path
) {
2602 _cleanup_strv_free_
char **empty_directories
= NULL
;
2603 char *tmp
= NULL
, *var
= NULL
;
2604 const char *root_dir
= NULL
, *root_image
= NULL
;
2605 NamespaceInfo ns_info
;
2606 bool needs_sandboxing
;
2607 BindMount
*bind_mounts
= NULL
;
2608 size_t n_bind_mounts
= 0;
2613 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2614 root_image
= context
->root_image
;
2617 root_dir
= context
->root_directory
;
2620 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2624 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2625 if (needs_sandboxing
) {
2626 /* The runtime struct only contains the parent of the private /tmp,
2627 * which is non-accessible to world users. Inside of it there's a /tmp
2628 * that is sticky, and that's the one we want to use here. */
2630 if (context
->private_tmp
&& runtime
) {
2631 if (runtime
->tmp_dir
)
2632 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2633 if (runtime
->var_tmp_dir
)
2634 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2637 ns_info
= (NamespaceInfo
) {
2638 .ignore_protect_paths
= false,
2639 .private_dev
= context
->private_devices
,
2640 .protect_control_groups
= context
->protect_control_groups
,
2641 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2642 .protect_kernel_modules
= context
->protect_kernel_modules
,
2643 .protect_kernel_logs
= context
->protect_kernel_logs
,
2644 .protect_hostname
= context
->protect_hostname
,
2645 .mount_apivfs
= context
->mount_apivfs
,
2646 .private_mounts
= context
->private_mounts
,
2648 } else if (!context
->dynamic_user
&& root_dir
)
2650 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2651 * sandbox info, otherwise enforce it, don't ignore protected paths and
2652 * fail if we are enable to apply the sandbox inside the mount namespace.
2654 ns_info
= (NamespaceInfo
) {
2655 .ignore_protect_paths
= true,
2658 ns_info
= (NamespaceInfo
) {};
2660 if (context
->mount_flags
== MS_SHARED
)
2661 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2663 r
= setup_namespace(root_dir
, root_image
,
2664 &ns_info
, context
->read_write_paths
,
2665 needs_sandboxing
? context
->read_only_paths
: NULL
,
2666 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2670 context
->temporary_filesystems
,
2671 context
->n_temporary_filesystems
,
2674 context
->log_namespace
,
2675 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2676 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2677 context
->mount_flags
,
2678 DISSECT_IMAGE_DISCARD_ON_LOOP
|DISSECT_IMAGE_RELAX_VAR_CHECK
|DISSECT_IMAGE_FSCK
,
2681 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2682 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
2683 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2684 * completely different execution environment. */
2686 if (insist_on_sandboxing(
2688 root_dir
, root_image
,
2691 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2692 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2693 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2697 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2702 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2706 static int apply_working_directory(
2707 const ExecContext
*context
,
2708 const ExecParameters
*params
,
2715 assert(exit_status
);
2717 if (context
->working_directory_home
) {
2720 *exit_status
= EXIT_CHDIR
;
2726 } else if (context
->working_directory
)
2727 wd
= context
->working_directory
;
2731 if (params
->flags
& EXEC_APPLY_CHROOT
)
2734 d
= prefix_roota(context
->root_directory
, wd
);
2736 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2737 *exit_status
= EXIT_CHDIR
;
2744 static int apply_root_directory(
2745 const ExecContext
*context
,
2746 const ExecParameters
*params
,
2747 const bool needs_mount_ns
,
2751 assert(exit_status
);
2753 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2754 if (!needs_mount_ns
&& context
->root_directory
)
2755 if (chroot(context
->root_directory
) < 0) {
2756 *exit_status
= EXIT_CHROOT
;
2764 static int setup_keyring(
2766 const ExecContext
*context
,
2767 const ExecParameters
*p
,
2768 uid_t uid
, gid_t gid
) {
2770 key_serial_t keyring
;
2779 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2780 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2781 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2782 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2783 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2784 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2786 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2789 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2790 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2791 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2792 * & group is just as nasty as acquiring a reference to the user keyring. */
2794 saved_uid
= getuid();
2795 saved_gid
= getgid();
2797 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2798 if (setregid(gid
, -1) < 0)
2799 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2802 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2803 if (setreuid(uid
, -1) < 0) {
2804 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2809 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2810 if (keyring
== -1) {
2811 if (errno
== ENOSYS
)
2812 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2813 else if (IN_SET(errno
, EACCES
, EPERM
))
2814 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2815 else if (errno
== EDQUOT
)
2816 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2818 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2823 /* When requested link the user keyring into the session keyring. */
2824 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2826 if (keyctl(KEYCTL_LINK
,
2827 KEY_SPEC_USER_KEYRING
,
2828 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2829 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2834 /* Restore uid/gid back */
2835 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2836 if (setreuid(saved_uid
, -1) < 0) {
2837 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2842 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2843 if (setregid(saved_gid
, -1) < 0)
2844 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2847 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2848 if (!sd_id128_is_null(u
->invocation_id
)) {
2851 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2853 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2855 if (keyctl(KEYCTL_SETPERM
, key
,
2856 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2857 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2858 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2863 /* Revert back uid & gid for the the last time, and exit */
2864 /* no extra logging, as only the first already reported error matters */
2865 if (getuid() != saved_uid
)
2866 (void) setreuid(saved_uid
, -1);
2868 if (getgid() != saved_gid
)
2869 (void) setregid(saved_gid
, -1);
2874 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2880 array
[(*n
)++] = pair
[0];
2882 array
[(*n
)++] = pair
[1];
2885 static int close_remaining_fds(
2886 const ExecParameters
*params
,
2887 const ExecRuntime
*runtime
,
2888 const DynamicCreds
*dcreds
,
2892 const int *fds
, size_t n_fds
) {
2894 size_t n_dont_close
= 0;
2895 int dont_close
[n_fds
+ 12];
2899 if (params
->stdin_fd
>= 0)
2900 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2901 if (params
->stdout_fd
>= 0)
2902 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2903 if (params
->stderr_fd
>= 0)
2904 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2907 dont_close
[n_dont_close
++] = socket_fd
;
2909 dont_close
[n_dont_close
++] = exec_fd
;
2911 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2912 n_dont_close
+= n_fds
;
2916 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2920 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2922 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2925 if (user_lookup_fd
>= 0)
2926 dont_close
[n_dont_close
++] = user_lookup_fd
;
2928 return close_all_fds(dont_close
, n_dont_close
);
2931 static int send_user_lookup(
2939 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2940 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2943 if (user_lookup_fd
< 0)
2946 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2949 if (writev(user_lookup_fd
,
2951 IOVEC_INIT(&uid
, sizeof(uid
)),
2952 IOVEC_INIT(&gid
, sizeof(gid
)),
2953 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2959 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2966 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2971 if (!c
->working_directory_home
)
2974 r
= get_home_dir(buf
);
2982 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2983 _cleanup_strv_free_
char ** list
= NULL
;
2984 ExecDirectoryType t
;
2991 assert(c
->dynamic_user
);
2993 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2994 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2997 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3000 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3006 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3009 if (exec_directory_is_private(c
, t
))
3010 e
= path_join(p
->prefix
[t
], "private", *i
);
3012 e
= path_join(p
->prefix
[t
], *i
);
3016 r
= strv_consume(&list
, e
);
3022 *ret
= TAKE_PTR(list
);
3027 static char *exec_command_line(char **argv
);
3029 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3030 bool using_subcgroup
;
3036 if (!params
->cgroup_path
)
3039 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3040 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3041 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3042 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3043 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3044 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3045 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3046 * flag, which is only passed for the former statements, not for the latter. */
3048 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3049 if (using_subcgroup
)
3050 p
= path_join(params
->cgroup_path
, ".control");
3052 p
= strdup(params
->cgroup_path
);
3057 return using_subcgroup
;
3060 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3061 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3067 if (!c
->numa_policy
.nodes
.set
) {
3068 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3072 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3078 return cpu_set_add_all(ret
, &s
);
3081 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3084 return c
->cpu_affinity_from_numa
;
3087 static int exec_child(
3089 const ExecCommand
*command
,
3090 const ExecContext
*context
,
3091 const ExecParameters
*params
,
3092 ExecRuntime
*runtime
,
3093 DynamicCreds
*dcreds
,
3095 const int named_iofds
[static 3],
3097 size_t n_socket_fds
,
3098 size_t n_storage_fds
,
3103 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3104 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
3105 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3106 const char *username
= NULL
, *groupname
= NULL
;
3107 _cleanup_free_
char *home_buffer
= NULL
;
3108 const char *home
= NULL
, *shell
= NULL
;
3109 char **final_argv
= NULL
;
3110 dev_t journal_stream_dev
= 0;
3111 ino_t journal_stream_ino
= 0;
3112 bool userns_set_up
= false;
3113 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3114 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3115 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3116 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3118 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3119 bool use_selinux
= false;
3122 bool use_smack
= false;
3125 bool use_apparmor
= false;
3127 uid_t saved_uid
= getuid();
3128 gid_t saved_gid
= getgid();
3129 uid_t uid
= UID_INVALID
;
3130 gid_t gid
= GID_INVALID
;
3132 ExecDirectoryType dt
;
3134 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3135 int ngids_after_pam
= 0;
3141 assert(exit_status
);
3143 rename_process_from_path(command
->path
);
3145 /* We reset exactly these signals, since they are the
3146 * only ones we set to SIG_IGN in the main daemon. All
3147 * others we leave untouched because we set them to
3148 * SIG_DFL or a valid handler initially, both of which
3149 * will be demoted to SIG_DFL. */
3150 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3151 SIGNALS_IGNORE
, -1);
3153 if (context
->ignore_sigpipe
)
3154 (void) ignore_signals(SIGPIPE
, -1);
3156 r
= reset_signal_mask();
3158 *exit_status
= EXIT_SIGNAL_MASK
;
3159 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3162 if (params
->idle_pipe
)
3163 do_idle_pipe_dance(params
->idle_pipe
);
3165 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3166 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3167 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3168 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3171 log_set_open_when_needed(true);
3173 /* In case anything used libc syslog(), close this here, too */
3176 n_fds
= n_socket_fds
+ n_storage_fds
;
3177 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
3179 *exit_status
= EXIT_FDS
;
3180 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3183 if (!context
->same_pgrp
)
3185 *exit_status
= EXIT_SETSID
;
3186 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3189 exec_context_tty_reset(context
, params
);
3191 if (unit_shall_confirm_spawn(unit
)) {
3192 const char *vc
= params
->confirm_spawn
;
3193 _cleanup_free_
char *cmdline
= NULL
;
3195 cmdline
= exec_command_line(command
->argv
);
3197 *exit_status
= EXIT_MEMORY
;
3201 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3202 if (r
!= CONFIRM_EXECUTE
) {
3203 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3204 *exit_status
= EXIT_SUCCESS
;
3207 *exit_status
= EXIT_CONFIRM
;
3208 log_unit_error(unit
, "Execution cancelled by the user");
3213 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3214 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3215 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3216 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3217 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3218 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3219 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3220 *exit_status
= EXIT_MEMORY
;
3221 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3224 if (context
->dynamic_user
&& dcreds
) {
3225 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3227 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3228 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
3229 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3230 *exit_status
= EXIT_USER
;
3231 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3234 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3236 *exit_status
= EXIT_MEMORY
;
3240 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3242 *exit_status
= EXIT_USER
;
3244 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3247 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3250 if (!uid_is_valid(uid
)) {
3251 *exit_status
= EXIT_USER
;
3252 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3256 if (!gid_is_valid(gid
)) {
3257 *exit_status
= EXIT_USER
;
3258 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3263 username
= dcreds
->user
->name
;
3266 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3268 *exit_status
= EXIT_USER
;
3269 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3272 r
= get_fixed_group(context
, &groupname
, &gid
);
3274 *exit_status
= EXIT_GROUP
;
3275 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3279 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3280 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3281 &supplementary_gids
, &ngids
);
3283 *exit_status
= EXIT_GROUP
;
3284 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3287 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3289 *exit_status
= EXIT_USER
;
3290 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3293 user_lookup_fd
= safe_close(user_lookup_fd
);
3295 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3297 *exit_status
= EXIT_CHDIR
;
3298 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3301 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3302 * must sure to drop O_NONBLOCK */
3304 (void) fd_nonblock(socket_fd
, false);
3306 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3307 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3308 if (params
->cgroup_path
) {
3309 _cleanup_free_
char *p
= NULL
;
3311 r
= exec_parameters_get_cgroup_path(params
, &p
);
3313 *exit_status
= EXIT_CGROUP
;
3314 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3317 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3319 *exit_status
= EXIT_CGROUP
;
3320 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3324 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3325 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3327 *exit_status
= EXIT_NETWORK
;
3328 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3332 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3334 *exit_status
= EXIT_STDIN
;
3335 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3338 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3340 *exit_status
= EXIT_STDOUT
;
3341 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3344 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3346 *exit_status
= EXIT_STDERR
;
3347 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3350 if (context
->oom_score_adjust_set
) {
3351 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3352 * prohibit write access to this file, and we shouldn't trip up over that. */
3353 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3354 if (IN_SET(r
, -EPERM
, -EACCES
))
3355 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3357 *exit_status
= EXIT_OOM_ADJUST
;
3358 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3362 if (context
->nice_set
) {
3363 r
= setpriority_closest(context
->nice
);
3365 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3368 if (context
->cpu_sched_set
) {
3369 struct sched_param param
= {
3370 .sched_priority
= context
->cpu_sched_priority
,
3373 r
= sched_setscheduler(0,
3374 context
->cpu_sched_policy
|
3375 (context
->cpu_sched_reset_on_fork
?
3376 SCHED_RESET_ON_FORK
: 0),
3379 *exit_status
= EXIT_SETSCHEDULER
;
3380 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3384 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
3385 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
3386 const CPUSet
*cpu_set
;
3388 if (context
->cpu_affinity_from_numa
) {
3389 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
3391 *exit_status
= EXIT_CPUAFFINITY
;
3392 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
3395 cpu_set
= &converted_cpu_set
;
3397 cpu_set
= &context
->cpu_set
;
3399 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
3400 *exit_status
= EXIT_CPUAFFINITY
;
3401 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3405 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
3406 r
= apply_numa_policy(&context
->numa_policy
);
3407 if (r
== -EOPNOTSUPP
)
3408 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
3410 *exit_status
= EXIT_NUMA_POLICY
;
3411 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
3415 if (context
->ioprio_set
)
3416 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3417 *exit_status
= EXIT_IOPRIO
;
3418 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3421 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3422 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3423 *exit_status
= EXIT_TIMERSLACK
;
3424 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3427 if (context
->personality
!= PERSONALITY_INVALID
) {
3428 r
= safe_personality(context
->personality
);
3430 *exit_status
= EXIT_PERSONALITY
;
3431 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3435 if (context
->utmp_id
)
3436 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3438 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3439 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3443 if (uid_is_valid(uid
)) {
3444 r
= chown_terminal(STDIN_FILENO
, uid
);
3446 *exit_status
= EXIT_STDIN
;
3447 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3451 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3452 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3453 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3454 * touch a single hierarchy too. */
3455 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3456 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3458 *exit_status
= EXIT_CGROUP
;
3459 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3463 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3464 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3466 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3469 r
= build_environment(
3481 *exit_status
= EXIT_MEMORY
;
3485 r
= build_pass_environment(context
, &pass_env
);
3487 *exit_status
= EXIT_MEMORY
;
3491 accum_env
= strv_env_merge(5,
3492 params
->environment
,
3495 context
->environment
,
3498 *exit_status
= EXIT_MEMORY
;
3501 accum_env
= strv_env_clean(accum_env
);
3503 (void) umask(context
->umask
);
3505 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3507 *exit_status
= EXIT_KEYRING
;
3508 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3511 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3512 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3514 /* 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 */
3515 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3517 /* 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 */
3518 if (needs_ambient_hack
)
3519 needs_setuid
= false;
3521 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3523 if (needs_sandboxing
) {
3524 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3525 * present. The actual MAC context application will happen later, as late as possible, to avoid
3526 * impacting our own code paths. */
3529 use_selinux
= mac_selinux_use();
3532 use_smack
= mac_smack_use();
3535 use_apparmor
= mac_apparmor_use();
3539 if (needs_sandboxing
) {
3542 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3543 * is set here. (See below.) */
3545 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3547 *exit_status
= EXIT_LIMITS
;
3548 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3554 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3555 * wins here. (See above.) */
3557 if (context
->pam_name
&& username
) {
3558 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3560 *exit_status
= EXIT_PAM
;
3561 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3564 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
3565 if (ngids_after_pam
< 0) {
3566 *exit_status
= EXIT_MEMORY
;
3567 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
3572 if (needs_sandboxing
) {
3574 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3575 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3577 *exit_status
= EXIT_SELINUX_CONTEXT
;
3578 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3583 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
3584 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
3585 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
3586 if (context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
3587 userns_set_up
= true;
3588 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
3590 *exit_status
= EXIT_USER
;
3591 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
3596 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3598 if (ns_type_supported(NAMESPACE_NET
)) {
3599 r
= setup_netns(runtime
->netns_storage_socket
);
3601 log_unit_warning_errno(unit
, r
,
3602 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
3604 *exit_status
= EXIT_NETWORK
;
3605 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3607 } else if (context
->network_namespace_path
) {
3608 *exit_status
= EXIT_NETWORK
;
3609 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
3610 "NetworkNamespacePath= is not supported, refusing.");
3612 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3615 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3616 if (needs_mount_namespace
) {
3617 _cleanup_free_
char *error_path
= NULL
;
3619 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
, &error_path
);
3621 *exit_status
= EXIT_NAMESPACE
;
3622 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
3623 error_path
? ": " : "", strempty(error_path
));
3627 if (needs_sandboxing
) {
3628 r
= apply_protect_hostname(unit
, context
, exit_status
);
3633 /* Drop groups as early as possible.
3634 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
3635 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
3637 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
3638 int ngids_to_enforce
= 0;
3640 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
3645 if (ngids_to_enforce
< 0) {
3646 *exit_status
= EXIT_MEMORY
;
3647 return log_unit_error_errno(unit
,
3649 "Failed to merge group lists. Group membership might be incorrect: %m");
3652 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
3654 *exit_status
= EXIT_GROUP
;
3655 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3659 /* If the user namespace was not set up above, try to do it now.
3660 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
3661 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
3662 * case of mount namespaces being less privileged when the mount point list is copied from a
3663 * different user namespace). */
3665 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
3666 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
3668 *exit_status
= EXIT_USER
;
3669 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3673 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3674 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3675 * however if we have it as we want to keep it open until the final execve(). */
3677 if (params
->exec_fd
>= 0) {
3678 exec_fd
= params
->exec_fd
;
3680 if (exec_fd
< 3 + (int) n_fds
) {
3683 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3684 * process we are about to execute. */
3686 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3688 *exit_status
= EXIT_FDS
;
3689 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3692 safe_close(exec_fd
);
3695 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3696 r
= fd_cloexec(exec_fd
, true);
3698 *exit_status
= EXIT_FDS
;
3699 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3703 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3704 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3705 fds_with_exec_fd
[n_fds
] = exec_fd
;
3706 n_fds_with_exec_fd
= n_fds
+ 1;
3708 fds_with_exec_fd
= fds
;
3709 n_fds_with_exec_fd
= n_fds
;
3712 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3714 r
= shift_fds(fds
, n_fds
);
3716 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3718 *exit_status
= EXIT_FDS
;
3719 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3722 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3723 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3724 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3727 secure_bits
= context
->secure_bits
;
3729 if (needs_sandboxing
) {
3732 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3733 * requested. (Note this is placed after the general resource limit initialization, see
3734 * above, in order to take precedence.) */
3735 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3736 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3737 *exit_status
= EXIT_LIMITS
;
3738 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3743 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3744 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3746 r
= setup_smack(context
, command
);
3748 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3749 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3754 bset
= context
->capability_bounding_set
;
3755 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3756 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3757 * instead of us doing that */
3758 if (needs_ambient_hack
)
3759 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3760 (UINT64_C(1) << CAP_SETUID
) |
3761 (UINT64_C(1) << CAP_SETGID
);
3763 if (!cap_test_all(bset
)) {
3764 r
= capability_bounding_set_drop(bset
, false);
3766 *exit_status
= EXIT_CAPABILITIES
;
3767 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3771 /* This is done before enforce_user, but ambient set
3772 * does not survive over setresuid() if keep_caps is not set. */
3773 if (!needs_ambient_hack
) {
3774 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3776 *exit_status
= EXIT_CAPABILITIES
;
3777 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3782 /* chroot to root directory first, before we lose the ability to chroot */
3783 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
3785 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
3788 if (uid_is_valid(uid
)) {
3789 r
= enforce_user(context
, uid
);
3791 *exit_status
= EXIT_USER
;
3792 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3795 if (!needs_ambient_hack
&&
3796 context
->capability_ambient_set
!= 0) {
3798 /* Fix the ambient capabilities after user change. */
3799 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3801 *exit_status
= EXIT_CAPABILITIES
;
3802 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3805 /* If we were asked to change user and ambient capabilities
3806 * were requested, we had to add keep-caps to the securebits
3807 * so that we would maintain the inherited capability set
3808 * through the setresuid(). Make sure that the bit is added
3809 * also to the context secure_bits so that we don't try to
3810 * drop the bit away next. */
3812 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3817 /* Apply working directory here, because the working directory might be on NFS and only the user running
3818 * this service might have the correct privilege to change to the working directory */
3819 r
= apply_working_directory(context
, params
, home
, exit_status
);
3821 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3823 if (needs_sandboxing
) {
3824 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3825 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3826 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3827 * are restricted. */
3831 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3834 r
= setexeccon(exec_context
);
3836 *exit_status
= EXIT_SELINUX_CONTEXT
;
3837 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3844 if (use_apparmor
&& context
->apparmor_profile
) {
3845 r
= aa_change_onexec(context
->apparmor_profile
);
3846 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3847 *exit_status
= EXIT_APPARMOR_PROFILE
;
3848 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3853 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3854 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3855 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3856 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3857 *exit_status
= EXIT_SECUREBITS
;
3858 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3861 if (context_has_no_new_privileges(context
))
3862 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3863 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3864 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3868 r
= apply_address_families(unit
, context
);
3870 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3871 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3874 r
= apply_memory_deny_write_execute(unit
, context
);
3876 *exit_status
= EXIT_SECCOMP
;
3877 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3880 r
= apply_restrict_realtime(unit
, context
);
3882 *exit_status
= EXIT_SECCOMP
;
3883 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3886 r
= apply_restrict_suid_sgid(unit
, context
);
3888 *exit_status
= EXIT_SECCOMP
;
3889 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
3892 r
= apply_restrict_namespaces(unit
, context
);
3894 *exit_status
= EXIT_SECCOMP
;
3895 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3898 r
= apply_protect_sysctl(unit
, context
);
3900 *exit_status
= EXIT_SECCOMP
;
3901 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3904 r
= apply_protect_kernel_modules(unit
, context
);
3906 *exit_status
= EXIT_SECCOMP
;
3907 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3910 r
= apply_protect_kernel_logs(unit
, context
);
3912 *exit_status
= EXIT_SECCOMP
;
3913 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
3916 r
= apply_protect_clock(unit
, context
);
3918 *exit_status
= EXIT_SECCOMP
;
3919 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
3922 r
= apply_private_devices(unit
, context
);
3924 *exit_status
= EXIT_SECCOMP
;
3925 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3928 r
= apply_syscall_archs(unit
, context
);
3930 *exit_status
= EXIT_SECCOMP
;
3931 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3934 r
= apply_lock_personality(unit
, context
);
3936 *exit_status
= EXIT_SECCOMP
;
3937 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3940 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3941 * by the filter as little as possible. */
3942 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3944 *exit_status
= EXIT_SECCOMP
;
3945 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3950 if (!strv_isempty(context
->unset_environment
)) {
3953 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3955 *exit_status
= EXIT_MEMORY
;
3959 strv_free_and_replace(accum_env
, ee
);
3962 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3963 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3964 if (!replaced_argv
) {
3965 *exit_status
= EXIT_MEMORY
;
3968 final_argv
= replaced_argv
;
3970 final_argv
= command
->argv
;
3972 if (DEBUG_LOGGING
) {
3973 _cleanup_free_
char *line
;
3975 line
= exec_command_line(final_argv
);
3977 log_struct(LOG_DEBUG
,
3978 "EXECUTABLE=%s", command
->path
,
3979 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3981 LOG_UNIT_INVOCATION_ID(unit
));
3987 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3988 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3990 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3991 *exit_status
= EXIT_EXEC
;
3992 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3996 execve(command
->path
, final_argv
, accum_env
);
4002 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4003 * that POLLHUP on it no longer means execve() succeeded. */
4005 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4006 *exit_status
= EXIT_EXEC
;
4007 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4011 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4012 log_struct_errno(LOG_INFO
, r
,
4013 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4015 LOG_UNIT_INVOCATION_ID(unit
),
4016 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4018 "EXECUTABLE=%s", command
->path
);
4022 *exit_status
= EXIT_EXEC
;
4023 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
4026 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4027 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4029 int exec_spawn(Unit
*unit
,
4030 ExecCommand
*command
,
4031 const ExecContext
*context
,
4032 const ExecParameters
*params
,
4033 ExecRuntime
*runtime
,
4034 DynamicCreds
*dcreds
,
4037 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4038 _cleanup_free_
char *subcgroup_path
= NULL
;
4039 _cleanup_strv_free_
char **files_env
= NULL
;
4040 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4041 _cleanup_free_
char *line
= NULL
;
4049 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4051 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4052 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4053 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4055 if (params
->n_socket_fds
> 1) {
4056 log_unit_error(unit
, "Got more than one socket.");
4060 if (params
->n_socket_fds
== 0) {
4061 log_unit_error(unit
, "Got no socket.");
4065 socket_fd
= params
->fds
[0];
4069 n_socket_fds
= params
->n_socket_fds
;
4070 n_storage_fds
= params
->n_storage_fds
;
4073 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4075 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4077 r
= exec_context_load_environment(unit
, context
, &files_env
);
4079 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4081 line
= exec_command_line(command
->argv
);
4085 log_struct(LOG_DEBUG
,
4086 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
4087 "EXECUTABLE=%s", command
->path
,
4089 LOG_UNIT_INVOCATION_ID(unit
));
4091 if (params
->cgroup_path
) {
4092 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4094 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4095 if (r
> 0) { /* We are using a child cgroup */
4096 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4098 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4104 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4107 int exit_status
= EXIT_SUCCESS
;
4109 r
= exec_child(unit
,
4121 unit
->manager
->user_lookup_fds
[1],
4125 const char *status
=
4126 exit_status_to_string(exit_status
,
4127 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4129 log_struct_errno(LOG_ERR
, r
,
4130 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4132 LOG_UNIT_INVOCATION_ID(unit
),
4133 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4134 status
, command
->path
),
4135 "EXECUTABLE=%s", command
->path
);
4141 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4143 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4144 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4145 * process will be killed too). */
4147 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4149 exec_status_start(&command
->exec_status
, pid
);
4155 void exec_context_init(ExecContext
*c
) {
4156 ExecDirectoryType i
;
4161 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4162 c
->cpu_sched_policy
= SCHED_OTHER
;
4163 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4164 c
->syslog_level_prefix
= true;
4165 c
->ignore_sigpipe
= true;
4166 c
->timer_slack_nsec
= NSEC_INFINITY
;
4167 c
->personality
= PERSONALITY_INVALID
;
4168 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
4169 c
->directories
[i
].mode
= 0755;
4170 c
->timeout_clean_usec
= USEC_INFINITY
;
4171 c
->capability_bounding_set
= CAP_ALL
;
4172 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4173 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4174 c
->log_level_max
= -1;
4175 numa_policy_reset(&c
->numa_policy
);
4178 void exec_context_done(ExecContext
*c
) {
4179 ExecDirectoryType i
;
4184 c
->environment
= strv_free(c
->environment
);
4185 c
->environment_files
= strv_free(c
->environment_files
);
4186 c
->pass_environment
= strv_free(c
->pass_environment
);
4187 c
->unset_environment
= strv_free(c
->unset_environment
);
4189 rlimit_free_all(c
->rlimit
);
4191 for (l
= 0; l
< 3; l
++) {
4192 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4193 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4196 c
->working_directory
= mfree(c
->working_directory
);
4197 c
->root_directory
= mfree(c
->root_directory
);
4198 c
->root_image
= mfree(c
->root_image
);
4199 c
->tty_path
= mfree(c
->tty_path
);
4200 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4201 c
->user
= mfree(c
->user
);
4202 c
->group
= mfree(c
->group
);
4204 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4206 c
->pam_name
= mfree(c
->pam_name
);
4208 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4209 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4210 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4212 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4213 c
->bind_mounts
= NULL
;
4214 c
->n_bind_mounts
= 0;
4215 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4216 c
->temporary_filesystems
= NULL
;
4217 c
->n_temporary_filesystems
= 0;
4219 cpu_set_reset(&c
->cpu_set
);
4220 numa_policy_reset(&c
->numa_policy
);
4222 c
->utmp_id
= mfree(c
->utmp_id
);
4223 c
->selinux_context
= mfree(c
->selinux_context
);
4224 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4225 c
->smack_process_label
= mfree(c
->smack_process_label
);
4227 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4228 c
->syscall_archs
= set_free(c
->syscall_archs
);
4229 c
->address_families
= set_free(c
->address_families
);
4231 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
4232 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
4234 c
->log_level_max
= -1;
4236 exec_context_free_log_extra_fields(c
);
4238 c
->log_ratelimit_interval_usec
= 0;
4239 c
->log_ratelimit_burst
= 0;
4241 c
->stdin_data
= mfree(c
->stdin_data
);
4242 c
->stdin_data_size
= 0;
4244 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4246 c
->log_namespace
= mfree(c
->log_namespace
);
4249 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4254 if (!runtime_prefix
)
4257 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4258 _cleanup_free_
char *p
;
4260 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4261 p
= path_join(runtime_prefix
, "private", *i
);
4263 p
= path_join(runtime_prefix
, *i
);
4267 /* We execute this synchronously, since we need to be sure this is gone when we start the
4269 (void) rm_rf(p
, REMOVE_ROOT
);
4275 static void exec_command_done(ExecCommand
*c
) {
4278 c
->path
= mfree(c
->path
);
4279 c
->argv
= strv_free(c
->argv
);
4282 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4285 for (i
= 0; i
< n
; i
++)
4286 exec_command_done(c
+i
);
4289 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4293 LIST_REMOVE(command
, c
, i
);
4294 exec_command_done(i
);
4301 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4304 for (i
= 0; i
< n
; i
++)
4305 c
[i
] = exec_command_free_list(c
[i
]);
4308 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4311 for (i
= 0; i
< n
; i
++)
4312 exec_status_reset(&c
[i
].exec_status
);
4315 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
4318 for (i
= 0; i
< n
; i
++) {
4321 LIST_FOREACH(command
, z
, c
[i
])
4322 exec_status_reset(&z
->exec_status
);
4326 typedef struct InvalidEnvInfo
{
4331 static void invalid_env(const char *p
, void *userdata
) {
4332 InvalidEnvInfo
*info
= userdata
;
4334 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4337 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4343 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4346 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4349 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4352 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4355 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4358 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4365 static int exec_context_named_iofds(
4366 const ExecContext
*c
,
4367 const ExecParameters
*p
,
4368 int named_iofds
[static 3]) {
4371 const char* stdio_fdname
[3];
4376 assert(named_iofds
);
4378 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4379 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4380 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4382 for (i
= 0; i
< 3; i
++)
4383 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4385 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4387 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4388 if (named_iofds
[STDIN_FILENO
] < 0 &&
4389 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4390 stdio_fdname
[STDIN_FILENO
] &&
4391 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4393 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4396 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4397 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4398 stdio_fdname
[STDOUT_FILENO
] &&
4399 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4401 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4404 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4405 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4406 stdio_fdname
[STDERR_FILENO
] &&
4407 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4409 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4413 return targets
== 0 ? 0 : -ENOENT
;
4416 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4417 char **i
, **r
= NULL
;
4422 STRV_FOREACH(i
, c
->environment_files
) {
4426 bool ignore
= false;
4428 _cleanup_globfree_ glob_t pglob
= {};
4437 if (!path_is_absolute(fn
)) {
4445 /* Filename supports globbing, take all matching files */
4446 k
= safe_glob(fn
, 0, &pglob
);
4455 /* When we don't match anything, -ENOENT should be returned */
4456 assert(pglob
.gl_pathc
> 0);
4458 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4459 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4467 /* Log invalid environment variables with filename */
4469 InvalidEnvInfo info
= {
4471 .path
= pglob
.gl_pathv
[n
]
4474 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4482 m
= strv_env_merge(2, r
, p
);
4498 static bool tty_may_match_dev_console(const char *tty
) {
4499 _cleanup_free_
char *resolved
= NULL
;
4504 tty
= skip_dev_prefix(tty
);
4506 /* trivial identity? */
4507 if (streq(tty
, "console"))
4510 if (resolve_dev_console(&resolved
) < 0)
4511 return true; /* if we could not resolve, assume it may */
4513 /* "tty0" means the active VC, so it may be the same sometimes */
4514 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4517 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
4520 return ec
->tty_reset
||
4522 ec
->tty_vt_disallocate
||
4523 is_terminal_input(ec
->std_input
) ||
4524 is_terminal_output(ec
->std_output
) ||
4525 is_terminal_output(ec
->std_error
);
4528 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4530 return exec_context_may_touch_tty(ec
) &&
4531 tty_may_match_dev_console(exec_context_tty_path(ec
));
4534 static void strv_fprintf(FILE *f
, char **l
) {
4540 fprintf(f
, " %s", *g
);
4543 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4544 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
4545 ExecDirectoryType dt
;
4552 prefix
= strempty(prefix
);
4556 "%sWorkingDirectory: %s\n"
4557 "%sRootDirectory: %s\n"
4558 "%sNonBlocking: %s\n"
4559 "%sPrivateTmp: %s\n"
4560 "%sPrivateDevices: %s\n"
4561 "%sProtectKernelTunables: %s\n"
4562 "%sProtectKernelModules: %s\n"
4563 "%sProtectKernelLogs: %s\n"
4564 "%sProtectClock: %s\n"
4565 "%sProtectControlGroups: %s\n"
4566 "%sPrivateNetwork: %s\n"
4567 "%sPrivateUsers: %s\n"
4568 "%sProtectHome: %s\n"
4569 "%sProtectSystem: %s\n"
4570 "%sMountAPIVFS: %s\n"
4571 "%sIgnoreSIGPIPE: %s\n"
4572 "%sMemoryDenyWriteExecute: %s\n"
4573 "%sRestrictRealtime: %s\n"
4574 "%sRestrictSUIDSGID: %s\n"
4575 "%sKeyringMode: %s\n"
4576 "%sProtectHostname: %s\n",
4578 prefix
, c
->working_directory
? c
->working_directory
: "/",
4579 prefix
, c
->root_directory
? c
->root_directory
: "/",
4580 prefix
, yes_no(c
->non_blocking
),
4581 prefix
, yes_no(c
->private_tmp
),
4582 prefix
, yes_no(c
->private_devices
),
4583 prefix
, yes_no(c
->protect_kernel_tunables
),
4584 prefix
, yes_no(c
->protect_kernel_modules
),
4585 prefix
, yes_no(c
->protect_kernel_logs
),
4586 prefix
, yes_no(c
->protect_clock
),
4587 prefix
, yes_no(c
->protect_control_groups
),
4588 prefix
, yes_no(c
->private_network
),
4589 prefix
, yes_no(c
->private_users
),
4590 prefix
, protect_home_to_string(c
->protect_home
),
4591 prefix
, protect_system_to_string(c
->protect_system
),
4592 prefix
, yes_no(c
->mount_apivfs
),
4593 prefix
, yes_no(c
->ignore_sigpipe
),
4594 prefix
, yes_no(c
->memory_deny_write_execute
),
4595 prefix
, yes_no(c
->restrict_realtime
),
4596 prefix
, yes_no(c
->restrict_suid_sgid
),
4597 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4598 prefix
, yes_no(c
->protect_hostname
));
4601 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4603 STRV_FOREACH(e
, c
->environment
)
4604 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4606 STRV_FOREACH(e
, c
->environment_files
)
4607 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4609 STRV_FOREACH(e
, c
->pass_environment
)
4610 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4612 STRV_FOREACH(e
, c
->unset_environment
)
4613 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4615 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4617 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4618 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4620 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4621 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4625 "%sTimeoutCleanSec: %s\n",
4626 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
4633 if (c
->oom_score_adjust_set
)
4635 "%sOOMScoreAdjust: %i\n",
4636 prefix
, c
->oom_score_adjust
);
4638 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4640 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4641 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4642 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4643 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4646 if (c
->ioprio_set
) {
4647 _cleanup_free_
char *class_str
= NULL
;
4649 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4651 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4653 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4656 if (c
->cpu_sched_set
) {
4657 _cleanup_free_
char *policy_str
= NULL
;
4659 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4661 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4664 "%sCPUSchedulingPriority: %i\n"
4665 "%sCPUSchedulingResetOnFork: %s\n",
4666 prefix
, c
->cpu_sched_priority
,
4667 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4670 if (c
->cpu_set
.set
) {
4671 _cleanup_free_
char *affinity
= NULL
;
4673 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
4674 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
4677 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
4678 _cleanup_free_
char *nodes
= NULL
;
4680 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
4681 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
4682 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
4685 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4686 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4689 "%sStandardInput: %s\n"
4690 "%sStandardOutput: %s\n"
4691 "%sStandardError: %s\n",
4692 prefix
, exec_input_to_string(c
->std_input
),
4693 prefix
, exec_output_to_string(c
->std_output
),
4694 prefix
, exec_output_to_string(c
->std_error
));
4696 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4697 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4698 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4699 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4700 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4701 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4703 if (c
->std_input
== EXEC_INPUT_FILE
)
4704 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4705 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4706 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4707 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4708 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4709 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4710 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4711 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4712 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4718 "%sTTYVHangup: %s\n"
4719 "%sTTYVTDisallocate: %s\n",
4720 prefix
, c
->tty_path
,
4721 prefix
, yes_no(c
->tty_reset
),
4722 prefix
, yes_no(c
->tty_vhangup
),
4723 prefix
, yes_no(c
->tty_vt_disallocate
));
4725 if (IN_SET(c
->std_output
,
4728 EXEC_OUTPUT_JOURNAL
,
4729 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4730 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4731 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4732 IN_SET(c
->std_error
,
4735 EXEC_OUTPUT_JOURNAL
,
4736 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4737 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4738 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4740 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4742 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4744 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4746 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4748 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4751 if (c
->log_level_max
>= 0) {
4752 _cleanup_free_
char *t
= NULL
;
4754 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4756 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4759 if (c
->log_ratelimit_interval_usec
> 0) {
4760 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4763 "%sLogRateLimitIntervalSec: %s\n",
4764 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
4767 if (c
->log_ratelimit_burst
> 0)
4768 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
4770 if (c
->n_log_extra_fields
> 0) {
4773 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4774 fprintf(f
, "%sLogExtraFields: ", prefix
);
4775 fwrite(c
->log_extra_fields
[j
].iov_base
,
4776 1, c
->log_extra_fields
[j
].iov_len
,
4782 if (c
->log_namespace
)
4783 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
4785 if (c
->secure_bits
) {
4786 _cleanup_free_
char *str
= NULL
;
4788 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4790 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4793 if (c
->capability_bounding_set
!= CAP_ALL
) {
4794 _cleanup_free_
char *str
= NULL
;
4796 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4798 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4801 if (c
->capability_ambient_set
!= 0) {
4802 _cleanup_free_
char *str
= NULL
;
4804 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4806 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4810 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4812 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4814 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4816 if (!strv_isempty(c
->supplementary_groups
)) {
4817 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4818 strv_fprintf(f
, c
->supplementary_groups
);
4823 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4825 if (!strv_isempty(c
->read_write_paths
)) {
4826 fprintf(f
, "%sReadWritePaths:", prefix
);
4827 strv_fprintf(f
, c
->read_write_paths
);
4831 if (!strv_isempty(c
->read_only_paths
)) {
4832 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4833 strv_fprintf(f
, c
->read_only_paths
);
4837 if (!strv_isempty(c
->inaccessible_paths
)) {
4838 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4839 strv_fprintf(f
, c
->inaccessible_paths
);
4843 if (c
->n_bind_mounts
> 0)
4844 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4845 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4846 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4847 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4848 c
->bind_mounts
[i
].source
,
4849 c
->bind_mounts
[i
].destination
,
4850 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4852 if (c
->n_temporary_filesystems
> 0)
4853 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4854 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4856 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4858 isempty(t
->options
) ? "" : ":",
4859 strempty(t
->options
));
4864 "%sUtmpIdentifier: %s\n",
4865 prefix
, c
->utmp_id
);
4867 if (c
->selinux_context
)
4869 "%sSELinuxContext: %s%s\n",
4870 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4872 if (c
->apparmor_profile
)
4874 "%sAppArmorProfile: %s%s\n",
4875 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4877 if (c
->smack_process_label
)
4879 "%sSmackProcessLabel: %s%s\n",
4880 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4882 if (c
->personality
!= PERSONALITY_INVALID
)
4884 "%sPersonality: %s\n",
4885 prefix
, strna(personality_to_string(c
->personality
)));
4888 "%sLockPersonality: %s\n",
4889 prefix
, yes_no(c
->lock_personality
));
4891 if (c
->syscall_filter
) {
4899 "%sSystemCallFilter: ",
4902 if (!c
->syscall_whitelist
)
4906 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4907 _cleanup_free_
char *name
= NULL
;
4908 const char *errno_name
= NULL
;
4909 int num
= PTR_TO_INT(val
);
4916 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4917 fputs(strna(name
), f
);
4920 errno_name
= errno_to_name(num
);
4922 fprintf(f
, ":%s", errno_name
);
4924 fprintf(f
, ":%d", num
);
4932 if (c
->syscall_archs
) {
4939 "%sSystemCallArchitectures:",
4943 SET_FOREACH(id
, c
->syscall_archs
, j
)
4944 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4949 if (exec_context_restrict_namespaces_set(c
)) {
4950 _cleanup_free_
char *s
= NULL
;
4952 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4954 fprintf(f
, "%sRestrictNamespaces: %s\n",
4958 if (c
->network_namespace_path
)
4960 "%sNetworkNamespacePath: %s\n",
4961 prefix
, c
->network_namespace_path
);
4963 if (c
->syscall_errno
> 0) {
4964 const char *errno_name
;
4966 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4968 errno_name
= errno_to_name(c
->syscall_errno
);
4970 fprintf(f
, "%s\n", errno_name
);
4972 fprintf(f
, "%d\n", c
->syscall_errno
);
4976 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4979 /* Returns true if the process forked off would run under
4980 * an unchanged UID or as root. */
4985 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4991 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4999 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5001 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5006 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5011 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
5012 free(c
->log_extra_fields
[l
].iov_base
);
5013 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5014 c
->n_log_extra_fields
= 0;
5017 void exec_context_revert_tty(ExecContext
*c
) {
5022 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5023 exec_context_tty_reset(c
, NULL
);
5025 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5026 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5027 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5029 if (exec_context_may_touch_tty(c
)) {
5032 path
= exec_context_tty_path(c
);
5034 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
5035 if (r
< 0 && r
!= -ENOENT
)
5036 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5041 int exec_context_get_clean_directories(
5047 _cleanup_strv_free_
char **l
= NULL
;
5048 ExecDirectoryType t
;
5055 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5058 if (!FLAGS_SET(mask
, 1U << t
))
5064 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5067 j
= path_join(prefix
[t
], *i
);
5071 r
= strv_consume(&l
, j
);
5075 /* Also remove private directories unconditionally. */
5076 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5077 j
= path_join(prefix
[t
], "private", *i
);
5081 r
= strv_consume(&l
, j
);
5092 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5093 ExecCleanMask mask
= 0;
5098 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5099 if (!strv_isempty(c
->directories
[t
].paths
))
5106 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5113 dual_timestamp_get(&s
->start_timestamp
);
5116 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5119 if (s
->pid
!= pid
) {
5125 dual_timestamp_get(&s
->exit_timestamp
);
5130 if (context
&& context
->utmp_id
)
5131 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5134 void exec_status_reset(ExecStatus
*s
) {
5137 *s
= (ExecStatus
) {};
5140 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5141 char buf
[FORMAT_TIMESTAMP_MAX
];
5149 prefix
= strempty(prefix
);
5152 "%sPID: "PID_FMT
"\n",
5155 if (dual_timestamp_is_set(&s
->start_timestamp
))
5157 "%sStart Timestamp: %s\n",
5158 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5160 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5162 "%sExit Timestamp: %s\n"
5164 "%sExit Status: %i\n",
5165 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5166 prefix
, sigchld_code_to_string(s
->code
),
5170 static char *exec_command_line(char **argv
) {
5178 STRV_FOREACH(a
, argv
)
5186 STRV_FOREACH(a
, argv
) {
5193 if (strpbrk(*a
, WHITESPACE
)) {
5204 /* FIXME: this doesn't really handle arguments that have
5205 * spaces and ticks in them */
5210 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5211 _cleanup_free_
char *cmd
= NULL
;
5212 const char *prefix2
;
5217 prefix
= strempty(prefix
);
5218 prefix2
= strjoina(prefix
, "\t");
5220 cmd
= exec_command_line(c
->argv
);
5222 "%sCommand Line: %s\n",
5223 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
5225 exec_status_dump(&c
->exec_status
, f
, prefix2
);
5228 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5231 prefix
= strempty(prefix
);
5233 LIST_FOREACH(command
, c
, c
)
5234 exec_command_dump(c
, f
, prefix
);
5237 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
5244 /* It's kind of important, that we keep the order here */
5245 LIST_FIND_TAIL(command
, *l
, end
);
5246 LIST_INSERT_AFTER(command
, *l
, end
, e
);
5251 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
5259 l
= strv_new_ap(path
, ap
);
5271 free_and_replace(c
->path
, p
);
5273 return strv_free_and_replace(c
->argv
, l
);
5276 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
5277 _cleanup_strv_free_
char **l
= NULL
;
5285 l
= strv_new_ap(path
, ap
);
5291 r
= strv_extend_strv(&c
->argv
, l
, false);
5298 static void *remove_tmpdir_thread(void *p
) {
5299 _cleanup_free_
char *path
= p
;
5301 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
5305 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
5312 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
5314 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
5315 if (destroy
&& rt
->tmp_dir
) {
5316 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
5318 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
5320 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
5327 if (destroy
&& rt
->var_tmp_dir
) {
5328 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
5330 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
5332 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
5333 free(rt
->var_tmp_dir
);
5336 rt
->var_tmp_dir
= NULL
;
5339 rt
->id
= mfree(rt
->id
);
5340 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
5341 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
5342 safe_close_pair(rt
->netns_storage_socket
);
5346 static void exec_runtime_freep(ExecRuntime
**rt
) {
5347 (void) exec_runtime_free(*rt
, false);
5350 static int exec_runtime_allocate(ExecRuntime
**ret
) {
5355 n
= new(ExecRuntime
, 1);
5359 *n
= (ExecRuntime
) {
5360 .netns_storage_socket
= { -1, -1 },
5367 static int exec_runtime_add(
5370 const char *tmp_dir
,
5371 const char *var_tmp_dir
,
5372 const int netns_storage_socket
[2],
5373 ExecRuntime
**ret
) {
5375 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
5381 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
5385 r
= exec_runtime_allocate(&rt
);
5389 rt
->id
= strdup(id
);
5394 rt
->tmp_dir
= strdup(tmp_dir
);
5398 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
5399 assert(var_tmp_dir
);
5400 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
5401 if (!rt
->var_tmp_dir
)
5405 if (netns_storage_socket
) {
5406 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
5407 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
5410 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
5419 /* do not remove created ExecRuntime object when the operation succeeds. */
5424 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
5425 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5426 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
5433 /* It is not necessary to create ExecRuntime object. */
5434 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
5437 if (c
->private_tmp
) {
5438 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
5443 if (c
->private_network
|| c
->network_namespace_path
) {
5444 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
5448 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
5453 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
5457 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
5465 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
5467 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5473 /* If not found, then create a new object. */
5474 r
= exec_runtime_make(m
, c
, id
, &rt
);
5476 /* When r == 0, it is not necessary to create ExecRuntime object. */
5480 /* increment reference counter. */
5486 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5490 assert(rt
->n_ref
> 0);
5496 return exec_runtime_free(rt
, destroy
);
5499 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5507 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5508 fprintf(f
, "exec-runtime=%s", rt
->id
);
5511 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5513 if (rt
->var_tmp_dir
)
5514 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5516 if (rt
->netns_storage_socket
[0] >= 0) {
5519 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5523 fprintf(f
, " netns-socket-0=%i", copy
);
5526 if (rt
->netns_storage_socket
[1] >= 0) {
5529 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5533 fprintf(f
, " netns-socket-1=%i", copy
);
5542 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5543 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5547 /* This is for the migration from old (v237 or earlier) deserialization text.
5548 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5549 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5550 * so or not from the serialized text, then we always creates a new object owned by this. */
5556 /* Manager manages ExecRuntime objects by the unit id.
5557 * So, we omit the serialized text when the unit does not have id (yet?)... */
5558 if (isempty(u
->id
)) {
5559 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5563 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5565 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5569 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5571 r
= exec_runtime_allocate(&rt_create
);
5575 rt_create
->id
= strdup(u
->id
);
5582 if (streq(key
, "tmp-dir")) {
5585 copy
= strdup(value
);
5589 free_and_replace(rt
->tmp_dir
, copy
);
5591 } else if (streq(key
, "var-tmp-dir")) {
5594 copy
= strdup(value
);
5598 free_and_replace(rt
->var_tmp_dir
, copy
);
5600 } else if (streq(key
, "netns-socket-0")) {
5603 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5604 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5608 safe_close(rt
->netns_storage_socket
[0]);
5609 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5611 } else if (streq(key
, "netns-socket-1")) {
5614 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5615 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5619 safe_close(rt
->netns_storage_socket
[1]);
5620 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5624 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5626 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5628 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5632 rt_create
->manager
= u
->manager
;
5641 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5642 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5643 int r
, fd0
= -1, fd1
= -1;
5644 const char *p
, *v
= value
;
5651 n
= strcspn(v
, " ");
5652 id
= strndupa(v
, n
);
5657 v
= startswith(p
, "tmp-dir=");
5659 n
= strcspn(v
, " ");
5660 tmp_dir
= strndupa(v
, n
);
5666 v
= startswith(p
, "var-tmp-dir=");
5668 n
= strcspn(v
, " ");
5669 var_tmp_dir
= strndupa(v
, n
);
5675 v
= startswith(p
, "netns-socket-0=");
5679 n
= strcspn(v
, " ");
5680 buf
= strndupa(v
, n
);
5681 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5682 log_debug("Unable to process exec-runtime netns fd specification.");
5685 fd0
= fdset_remove(fds
, fd0
);
5691 v
= startswith(p
, "netns-socket-1=");
5695 n
= strcspn(v
, " ");
5696 buf
= strndupa(v
, n
);
5697 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5698 log_debug("Unable to process exec-runtime netns fd specification.");
5701 fd1
= fdset_remove(fds
, fd1
);
5706 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5708 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5711 void exec_runtime_vacuum(Manager
*m
) {
5717 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5719 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5723 (void) exec_runtime_free(rt
, false);
5727 void exec_params_clear(ExecParameters
*p
) {
5731 strv_free(p
->environment
);
5734 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5735 [EXEC_INPUT_NULL
] = "null",
5736 [EXEC_INPUT_TTY
] = "tty",
5737 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5738 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5739 [EXEC_INPUT_SOCKET
] = "socket",
5740 [EXEC_INPUT_NAMED_FD
] = "fd",
5741 [EXEC_INPUT_DATA
] = "data",
5742 [EXEC_INPUT_FILE
] = "file",
5745 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5747 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5748 [EXEC_OUTPUT_INHERIT
] = "inherit",
5749 [EXEC_OUTPUT_NULL
] = "null",
5750 [EXEC_OUTPUT_TTY
] = "tty",
5751 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5752 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5753 [EXEC_OUTPUT_KMSG
] = "kmsg",
5754 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5755 [EXEC_OUTPUT_JOURNAL
] = "journal",
5756 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5757 [EXEC_OUTPUT_SOCKET
] = "socket",
5758 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5759 [EXEC_OUTPUT_FILE
] = "file",
5760 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5763 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5765 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5766 [EXEC_UTMP_INIT
] = "init",
5767 [EXEC_UTMP_LOGIN
] = "login",
5768 [EXEC_UTMP_USER
] = "user",
5771 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5773 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5774 [EXEC_PRESERVE_NO
] = "no",
5775 [EXEC_PRESERVE_YES
] = "yes",
5776 [EXEC_PRESERVE_RESTART
] = "restart",
5779 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5781 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
5782 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5783 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5784 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5785 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5786 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5787 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5790 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5792 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
5793 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
5794 * directories, specifically .timer units with their timestamp touch file. */
5795 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5796 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
5797 [EXEC_DIRECTORY_STATE
] = "state",
5798 [EXEC_DIRECTORY_CACHE
] = "cache",
5799 [EXEC_DIRECTORY_LOGS
] = "logs",
5800 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
5803 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
5805 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
5806 * the service payload in. */
5807 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5808 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5809 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5810 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5811 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5812 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5815 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5817 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5818 [EXEC_KEYRING_INHERIT
] = "inherit",
5819 [EXEC_KEYRING_PRIVATE
] = "private",
5820 [EXEC_KEYRING_SHARED
] = "shared",
5823 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);