1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <sys/eventfd.h>
10 #include <sys/personality.h>
11 #include <sys/prctl.h>
13 #include <sys/types.h>
19 #include <security/pam_appl.h>
23 #include <selinux/selinux.h>
31 #include <sys/apparmor.h>
34 #include "sd-messages.h"
38 #include "alloc-util.h"
40 #include "apparmor-util.h"
45 #include "capability-util.h"
46 #include "cgroup-setup.h"
47 #include "chown-recursive.h"
48 #include "cpu-set-util.h"
52 #include "errno-list.h"
54 #include "exit-status.h"
57 #include "format-util.h"
59 #include "glob-util.h"
60 #include "hexdecoct.h"
67 #include "memory-util.h"
68 #include "missing_fs.h"
70 #include "mount-util.h"
71 #include "mountpoint-util.h"
72 #include "namespace.h"
73 #include "parse-util.h"
74 #include "path-util.h"
75 #include "process-util.h"
76 #include "random-util.h"
77 #include "rlimit-util.h"
80 #include "seccomp-util.h"
82 #include "securebits-util.h"
83 #include "selinux-util.h"
84 #include "signal-util.h"
85 #include "smack-util.h"
86 #include "socket-util.h"
88 #include "stat-util.h"
89 #include "string-table.h"
90 #include "string-util.h"
92 #include "syslog-util.h"
93 #include "terminal-util.h"
94 #include "tmpfile-util.h"
95 #include "umask-util.h"
97 #include "user-util.h"
98 #include "utmp-wtmp.h"
100 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
101 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
103 #define SNDBUF_SIZE (8*1024*1024)
105 static int shift_fds(int fds
[], size_t n_fds
) {
109 /* Modifies the fds array! (sorts it) */
113 for (int start
= 0;;) {
114 int restart_from
= -1;
116 for (int i
= start
; i
< (int) n_fds
; i
++) {
119 /* Already at right index? */
123 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
130 /* Hmm, the fd we wanted isn't free? Then
131 * let's remember that and try again from here */
132 if (nfd
!= i
+3 && restart_from
< 0)
136 if (restart_from
< 0)
139 start
= restart_from
;
145 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
149 n_fds
= n_socket_fds
+ n_storage_fds
;
155 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
156 * O_NONBLOCK only applies to socket activation though. */
158 for (size_t i
= 0; i
< n_fds
; i
++) {
160 if (i
< n_socket_fds
) {
161 r
= fd_nonblock(fds
[i
], nonblock
);
166 /* We unconditionally drop FD_CLOEXEC from the fds,
167 * since after all we want to pass these fds to our
170 r
= fd_cloexec(fds
[i
], false);
178 static const char *exec_context_tty_path(const ExecContext
*context
) {
181 if (context
->stdio_as_fds
)
184 if (context
->tty_path
)
185 return context
->tty_path
;
187 return "/dev/console";
190 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
195 path
= exec_context_tty_path(context
);
197 if (context
->tty_vhangup
) {
198 if (p
&& p
->stdin_fd
>= 0)
199 (void) terminal_vhangup_fd(p
->stdin_fd
);
201 (void) terminal_vhangup(path
);
204 if (context
->tty_reset
) {
205 if (p
&& p
->stdin_fd
>= 0)
206 (void) reset_terminal_fd(p
->stdin_fd
, true);
208 (void) reset_terminal(path
);
211 if (context
->tty_vt_disallocate
&& path
)
212 (void) vt_disallocate(path
);
215 static bool is_terminal_input(ExecInput i
) {
218 EXEC_INPUT_TTY_FORCE
,
219 EXEC_INPUT_TTY_FAIL
);
222 static bool is_terminal_output(ExecOutput o
) {
225 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
226 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
229 static bool is_kmsg_output(ExecOutput o
) {
232 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
235 static bool exec_context_needs_term(const ExecContext
*c
) {
238 /* Return true if the execution context suggests we should set $TERM to something useful. */
240 if (is_terminal_input(c
->std_input
))
243 if (is_terminal_output(c
->std_output
))
246 if (is_terminal_output(c
->std_error
))
249 return !!c
->tty_path
;
252 static int open_null_as(int flags
, int nfd
) {
257 fd
= open("/dev/null", flags
|O_NOCTTY
);
261 return move_fd(fd
, nfd
, false);
264 static int connect_journal_socket(
266 const char *log_namespace
,
270 union sockaddr_union sa
;
272 uid_t olduid
= UID_INVALID
;
273 gid_t oldgid
= GID_INVALID
;
278 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
279 "/run/systemd/journal/stdout";
280 r
= sockaddr_un_set_path(&sa
.un
, j
);
285 if (gid_is_valid(gid
)) {
288 if (setegid(gid
) < 0)
292 if (uid_is_valid(uid
)) {
295 if (seteuid(uid
) < 0) {
301 r
= connect(fd
, &sa
.sa
, sa_len
) < 0 ? -errno
: 0;
303 /* If we fail to restore the uid or gid, things will likely
304 fail later on. This should only happen if an LSM interferes. */
306 if (uid_is_valid(uid
))
307 (void) seteuid(olduid
);
310 if (gid_is_valid(gid
))
311 (void) setegid(oldgid
);
316 static int connect_logger_as(
318 const ExecContext
*context
,
319 const ExecParameters
*params
,
326 _cleanup_close_
int fd
= -1;
331 assert(output
< _EXEC_OUTPUT_MAX
);
335 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
339 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
343 if (shutdown(fd
, SHUT_RD
) < 0)
346 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
356 context
->syslog_identifier
?: ident
,
357 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
358 context
->syslog_priority
,
359 !!context
->syslog_level_prefix
,
361 is_kmsg_output(output
),
362 is_terminal_output(output
)) < 0)
365 return move_fd(TAKE_FD(fd
), nfd
, false);
368 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
374 fd
= open_terminal(path
, flags
| O_NOCTTY
);
378 return move_fd(fd
, nfd
, false);
381 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
382 union sockaddr_union sa
;
384 _cleanup_close_
int fd
= -1;
389 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
392 fd
= open(path
, flags
|O_NOCTTY
, mode
);
396 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
399 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
401 r
= sockaddr_un_set_path(&sa
.un
, path
);
403 return r
== -EINVAL
? -ENXIO
: r
;
406 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
410 if (connect(fd
, &sa
.sa
, sa_len
) < 0)
411 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
412 * indication that this wasn't an AF_UNIX socket after all */
414 if ((flags
& O_ACCMODE
) == O_RDONLY
)
415 r
= shutdown(fd
, SHUT_WR
);
416 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
417 r
= shutdown(fd
, SHUT_RD
);
426 static int fixup_input(
427 const ExecContext
*context
,
429 bool apply_tty_stdin
) {
435 std_input
= context
->std_input
;
437 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
438 return EXEC_INPUT_NULL
;
440 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
441 return EXEC_INPUT_NULL
;
443 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
444 return EXEC_INPUT_NULL
;
449 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
451 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
452 return EXEC_OUTPUT_INHERIT
;
457 static int setup_input(
458 const ExecContext
*context
,
459 const ExecParameters
*params
,
461 const int named_iofds
[static 3]) {
469 if (params
->stdin_fd
>= 0) {
470 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
473 /* Try to make this the controlling tty, if it is a tty, and reset it */
474 if (isatty(STDIN_FILENO
)) {
475 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
476 (void) reset_terminal_fd(STDIN_FILENO
, true);
482 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
486 case EXEC_INPUT_NULL
:
487 return open_null_as(O_RDONLY
, STDIN_FILENO
);
490 case EXEC_INPUT_TTY_FORCE
:
491 case EXEC_INPUT_TTY_FAIL
: {
494 fd
= acquire_terminal(exec_context_tty_path(context
),
495 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
496 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
497 ACQUIRE_TERMINAL_WAIT
,
502 return move_fd(fd
, STDIN_FILENO
, false);
505 case EXEC_INPUT_SOCKET
:
506 assert(socket_fd
>= 0);
508 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
510 case EXEC_INPUT_NAMED_FD
:
511 assert(named_iofds
[STDIN_FILENO
] >= 0);
513 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
514 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
516 case EXEC_INPUT_DATA
: {
519 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
523 return move_fd(fd
, STDIN_FILENO
, false);
526 case EXEC_INPUT_FILE
: {
530 assert(context
->stdio_file
[STDIN_FILENO
]);
532 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
533 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
535 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
539 return move_fd(fd
, STDIN_FILENO
, false);
543 assert_not_reached("Unknown input type");
547 static bool can_inherit_stderr_from_stdout(
548 const ExecContext
*context
,
554 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
557 if (e
== EXEC_OUTPUT_INHERIT
)
562 if (e
== EXEC_OUTPUT_NAMED_FD
)
563 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
565 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
, EXEC_OUTPUT_FILE_TRUNCATE
))
566 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
571 static int setup_output(
573 const ExecContext
*context
,
574 const ExecParameters
*params
,
577 const int named_iofds
[static 3],
581 dev_t
*journal_stream_dev
,
582 ino_t
*journal_stream_ino
) {
592 assert(journal_stream_dev
);
593 assert(journal_stream_ino
);
595 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
597 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
600 return STDOUT_FILENO
;
603 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
604 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
607 return STDERR_FILENO
;
610 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
611 o
= fixup_output(context
->std_output
, socket_fd
);
613 if (fileno
== STDERR_FILENO
) {
615 e
= fixup_output(context
->std_error
, socket_fd
);
617 /* This expects the input and output are already set up */
619 /* Don't change the stderr file descriptor if we inherit all
620 * the way and are not on a tty */
621 if (e
== EXEC_OUTPUT_INHERIT
&&
622 o
== EXEC_OUTPUT_INHERIT
&&
623 i
== EXEC_INPUT_NULL
&&
624 !is_terminal_input(context
->std_input
) &&
628 /* Duplicate from stdout if possible */
629 if (can_inherit_stderr_from_stdout(context
, o
, e
))
630 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
634 } else if (o
== EXEC_OUTPUT_INHERIT
) {
635 /* If input got downgraded, inherit the original value */
636 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
637 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
639 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
640 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
641 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
643 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
647 /* We need to open /dev/null here anew, to get the right access mode. */
648 return open_null_as(O_WRONLY
, fileno
);
653 case EXEC_OUTPUT_NULL
:
654 return open_null_as(O_WRONLY
, fileno
);
656 case EXEC_OUTPUT_TTY
:
657 if (is_terminal_input(i
))
658 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
660 /* We don't reset the terminal if this is just about output */
661 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
663 case EXEC_OUTPUT_KMSG
:
664 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
665 case EXEC_OUTPUT_JOURNAL
:
666 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
667 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
669 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
670 r
= open_null_as(O_WRONLY
, fileno
);
674 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
675 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
676 * services to detect whether they are connected to the journal or not.
678 * If both stdout and stderr are connected to a stream then let's make sure to store the data
679 * about STDERR as that's usually the best way to do logging. */
681 if (fstat(fileno
, &st
) >= 0 &&
682 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
683 *journal_stream_dev
= st
.st_dev
;
684 *journal_stream_ino
= st
.st_ino
;
689 case EXEC_OUTPUT_SOCKET
:
690 assert(socket_fd
>= 0);
692 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
694 case EXEC_OUTPUT_NAMED_FD
:
695 assert(named_iofds
[fileno
] >= 0);
697 (void) fd_nonblock(named_iofds
[fileno
], false);
698 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
700 case EXEC_OUTPUT_FILE
:
701 case EXEC_OUTPUT_FILE_APPEND
:
702 case EXEC_OUTPUT_FILE_TRUNCATE
: {
706 assert(context
->stdio_file
[fileno
]);
708 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
709 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
712 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
715 if (o
== EXEC_OUTPUT_FILE_APPEND
)
717 else if (o
== EXEC_OUTPUT_FILE_TRUNCATE
)
720 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
724 return move_fd(fd
, fileno
, 0);
728 assert_not_reached("Unknown error type");
732 static int chown_terminal(int fd
, uid_t uid
) {
737 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
738 if (isatty(fd
) < 1) {
739 if (IN_SET(errno
, EINVAL
, ENOTTY
))
740 return 0; /* not a tty */
745 /* This might fail. What matters are the results. */
746 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, -1);
753 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
754 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
757 assert(_saved_stdin
);
758 assert(_saved_stdout
);
760 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
764 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
765 if (saved_stdout
< 0)
768 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
772 r
= chown_terminal(fd
, getuid());
776 r
= reset_terminal_fd(fd
, true);
780 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
785 *_saved_stdin
= saved_stdin
;
786 *_saved_stdout
= saved_stdout
;
788 saved_stdin
= saved_stdout
= -1;
793 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
796 if (err
== -ETIMEDOUT
)
797 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
800 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
804 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
805 _cleanup_close_
int fd
= -1;
809 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
813 write_confirm_error_fd(err
, fd
, u
);
816 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
820 assert(saved_stdout
);
824 if (*saved_stdin
>= 0)
825 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
828 if (*saved_stdout
>= 0)
829 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
832 *saved_stdin
= safe_close(*saved_stdin
);
833 *saved_stdout
= safe_close(*saved_stdout
);
839 CONFIRM_PRETEND_FAILURE
= -1,
840 CONFIRM_PRETEND_SUCCESS
= 0,
844 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
845 int saved_stdout
= -1, saved_stdin
= -1, r
;
846 _cleanup_free_
char *e
= NULL
;
849 /* For any internal errors, assume a positive response. */
850 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
852 write_confirm_error(r
, vc
, u
);
853 return CONFIRM_EXECUTE
;
856 /* confirm_spawn might have been disabled while we were sleeping. */
857 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
862 e
= ellipsize(cmdline
, 60, 100);
870 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
872 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
879 printf("Resuming normal execution.\n");
880 manager_disable_confirm_spawn();
884 unit_dump(u
, stdout
, " ");
885 continue; /* ask again */
887 printf("Failing execution.\n");
888 r
= CONFIRM_PRETEND_FAILURE
;
891 printf(" c - continue, proceed without asking anymore\n"
892 " D - dump, show the state of the unit\n"
893 " f - fail, don't execute the command and pretend it failed\n"
895 " i - info, show a short summary of the unit\n"
896 " j - jobs, show jobs that are in progress\n"
897 " s - skip, don't execute the command and pretend it succeeded\n"
898 " y - yes, execute the command\n");
899 continue; /* ask again */
901 printf(" Description: %s\n"
904 u
->id
, u
->description
, cmdline
);
905 continue; /* ask again */
907 manager_dump_jobs(u
->manager
, stdout
, " ");
908 continue; /* ask again */
910 /* 'n' was removed in favor of 'f'. */
911 printf("Didn't understand 'n', did you mean 'f'?\n");
912 continue; /* ask again */
914 printf("Skipping execution.\n");
915 r
= CONFIRM_PRETEND_SUCCESS
;
921 assert_not_reached("Unhandled choice");
927 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
931 static int get_fixed_user(const ExecContext
*c
, const char **user
,
932 uid_t
*uid
, gid_t
*gid
,
933 const char **home
, const char **shell
) {
942 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
943 * (i.e. are "/" or "/bin/nologin"). */
946 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
954 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
964 r
= get_group_creds(&name
, gid
, 0);
972 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
973 const char *group
, gid_t gid
,
974 gid_t
**supplementary_gids
, int *ngids
) {
978 bool keep_groups
= false;
979 gid_t
*groups
= NULL
;
980 _cleanup_free_ gid_t
*l_gids
= NULL
;
985 * If user is given, then lookup GID and supplementary groups list.
986 * We avoid NSS lookups for gid=0. Also we have to initialize groups
987 * here and as early as possible so we keep the list of supplementary
988 * groups of the caller.
990 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
991 /* First step, initialize groups from /etc/groups */
992 if (initgroups(user
, gid
) < 0)
998 if (strv_isempty(c
->supplementary_groups
))
1002 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1003 * be positive, otherwise fail.
1006 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1007 if (ngroups_max
<= 0)
1008 return errno_or_else(EOPNOTSUPP
);
1010 l_gids
= new(gid_t
, ngroups_max
);
1016 * Lookup the list of groups that the user belongs to, we
1017 * avoid NSS lookups here too for gid=0.
1020 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1025 STRV_FOREACH(i
, c
->supplementary_groups
) {
1028 if (k
>= ngroups_max
)
1032 r
= get_group_creds(&g
, l_gids
+k
, 0);
1040 * Sets ngids to zero to drop all supplementary groups, happens
1041 * when we are under root and SupplementaryGroups= is empty.
1048 /* Otherwise get the final list of supplementary groups */
1049 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1053 *supplementary_gids
= groups
;
1061 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1064 /* Handle SupplementaryGroups= if it is not empty */
1066 r
= maybe_setgroups(ngids
, supplementary_gids
);
1071 if (gid_is_valid(gid
)) {
1072 /* Then set our gids */
1073 if (setresgid(gid
, gid
, gid
) < 0)
1080 static int set_securebits(int bits
, int mask
) {
1081 int current
, applied
;
1082 current
= prctl(PR_GET_SECUREBITS
);
1085 /* Clear all securebits defined in mask and set bits */
1086 applied
= (current
& ~mask
) | bits
;
1087 if (current
== applied
)
1089 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1094 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1098 if (!uid_is_valid(uid
))
1101 /* Sets (but doesn't look up) the uid and make sure we keep the
1102 * capabilities while doing so. For setting secure bits the capability CAP_SETPCAP is
1103 * required, so we also need keep-caps in this case.
1106 if (context
->capability_ambient_set
!= 0 || context
->secure_bits
!= 0) {
1108 /* First step: If we need to keep capabilities but
1109 * drop privileges we need to make sure we keep our
1110 * caps, while we drop privileges. */
1112 /* Add KEEP_CAPS to the securebits */
1113 r
= set_securebits(1<<SECURE_KEEP_CAPS
, 0);
1119 /* Second step: actually set the uids */
1120 if (setresuid(uid
, uid
, uid
) < 0)
1123 /* At this point we should have all necessary capabilities but
1124 are otherwise a normal user. However, the caps might got
1125 corrupted due to the setresuid() so we need clean them up
1126 later. This is done outside of this call. */
1133 static int null_conv(
1135 const struct pam_message
**msg
,
1136 struct pam_response
**resp
,
1137 void *appdata_ptr
) {
1139 /* We don't support conversations */
1141 return PAM_CONV_ERR
;
1146 static int setup_pam(
1153 const int fds
[], size_t n_fds
) {
1157 static const struct pam_conv conv
= {
1162 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1163 pam_handle_t
*handle
= NULL
;
1165 int pam_code
= PAM_SUCCESS
, r
;
1166 char **nv
, **e
= NULL
;
1167 bool close_session
= false;
1168 pid_t pam_pid
= 0, parent_pid
;
1175 /* We set up PAM in the parent process, then fork. The child
1176 * will then stay around until killed via PR_GET_PDEATHSIG or
1177 * systemd via the cgroup logic. It will then remove the PAM
1178 * session again. The parent process will exec() the actual
1179 * daemon. We do things this way to ensure that the main PID
1180 * of the daemon is the one we initially fork()ed. */
1182 r
= barrier_create(&barrier
);
1186 if (log_get_max_level() < LOG_DEBUG
)
1187 flags
|= PAM_SILENT
;
1189 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1190 if (pam_code
!= PAM_SUCCESS
) {
1196 _cleanup_free_
char *q
= NULL
;
1198 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1199 * out if that's the case, and read the TTY off it. */
1201 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1202 tty
= strjoina("/dev/", q
);
1206 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1207 if (pam_code
!= PAM_SUCCESS
)
1211 STRV_FOREACH(nv
, *env
) {
1212 pam_code
= pam_putenv(handle
, *nv
);
1213 if (pam_code
!= PAM_SUCCESS
)
1217 pam_code
= pam_acct_mgmt(handle
, flags
);
1218 if (pam_code
!= PAM_SUCCESS
)
1221 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1222 if (pam_code
!= PAM_SUCCESS
)
1223 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1225 pam_code
= pam_open_session(handle
, flags
);
1226 if (pam_code
!= PAM_SUCCESS
)
1229 close_session
= true;
1231 e
= pam_getenvlist(handle
);
1233 pam_code
= PAM_BUF_ERR
;
1237 /* Block SIGTERM, so that we know that it won't get lost in
1240 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1242 parent_pid
= getpid_cached();
1244 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1248 int sig
, ret
= EXIT_PAM
;
1250 /* The child's job is to reset the PAM session on
1252 barrier_set_role(&barrier
, BARRIER_CHILD
);
1254 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1255 * those fds are open here that have been opened by PAM. */
1256 (void) close_many(fds
, n_fds
);
1258 /* Drop privileges - we don't need any to pam_close_session
1259 * and this will make PR_SET_PDEATHSIG work in most cases.
1260 * If this fails, ignore the error - but expect sd-pam threads
1261 * to fail to exit normally */
1263 r
= maybe_setgroups(0, NULL
);
1265 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1266 if (setresgid(gid
, gid
, gid
) < 0)
1267 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1268 if (setresuid(uid
, uid
, uid
) < 0)
1269 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1271 (void) ignore_signals(SIGPIPE
, -1);
1273 /* Wait until our parent died. This will only work if
1274 * the above setresuid() succeeds, otherwise the kernel
1275 * will not allow unprivileged parents kill their privileged
1276 * children this way. We rely on the control groups kill logic
1277 * to do the rest for us. */
1278 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1281 /* Tell the parent that our setup is done. This is especially
1282 * important regarding dropping privileges. Otherwise, unit
1283 * setup might race against our setresuid(2) call.
1285 * If the parent aborted, we'll detect this below, hence ignore
1286 * return failure here. */
1287 (void) barrier_place(&barrier
);
1289 /* Check if our parent process might already have died? */
1290 if (getppid() == parent_pid
) {
1293 assert_se(sigemptyset(&ss
) >= 0);
1294 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1297 if (sigwait(&ss
, &sig
) < 0) {
1304 assert(sig
== SIGTERM
);
1309 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1310 if (pam_code
!= PAM_SUCCESS
)
1313 /* If our parent died we'll end the session */
1314 if (getppid() != parent_pid
) {
1315 pam_code
= pam_close_session(handle
, flags
);
1316 if (pam_code
!= PAM_SUCCESS
)
1323 pam_end(handle
, pam_code
| flags
);
1327 barrier_set_role(&barrier
, BARRIER_PARENT
);
1329 /* If the child was forked off successfully it will do all the
1330 * cleanups, so forget about the handle here. */
1333 /* Unblock SIGTERM again in the parent */
1334 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1336 /* We close the log explicitly here, since the PAM modules
1337 * might have opened it, but we don't want this fd around. */
1340 /* Synchronously wait for the child to initialize. We don't care for
1341 * errors as we cannot recover. However, warn loudly if it happens. */
1342 if (!barrier_place_and_sync(&barrier
))
1343 log_error("PAM initialization failed");
1345 return strv_free_and_replace(*env
, e
);
1348 if (pam_code
!= PAM_SUCCESS
) {
1349 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1350 r
= -EPERM
; /* PAM errors do not map to errno */
1352 log_error_errno(r
, "PAM failed: %m");
1356 pam_code
= pam_close_session(handle
, flags
);
1358 pam_end(handle
, pam_code
| flags
);
1370 static void rename_process_from_path(const char *path
) {
1371 char process_name
[11];
1375 /* This resulting string must fit in 10 chars (i.e. the length
1376 * of "/sbin/init") to look pretty in /bin/ps */
1380 rename_process("(...)");
1386 /* The end of the process name is usually more
1387 * interesting, since the first bit might just be
1393 process_name
[0] = '(';
1394 memcpy(process_name
+1, p
, l
);
1395 process_name
[1+l
] = ')';
1396 process_name
[1+l
+1] = 0;
1398 rename_process(process_name
);
1401 static bool context_has_address_families(const ExecContext
*c
) {
1404 return c
->address_families_allow_list
||
1405 !set_isempty(c
->address_families
);
1408 static bool context_has_syscall_filters(const ExecContext
*c
) {
1411 return c
->syscall_allow_list
||
1412 !hashmap_isempty(c
->syscall_filter
);
1415 static bool context_has_syscall_logs(const ExecContext
*c
) {
1418 return c
->syscall_log_allow_list
||
1419 !hashmap_isempty(c
->syscall_log
);
1422 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1425 if (c
->no_new_privileges
)
1428 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1431 /* We need NNP if we have any form of seccomp and are unprivileged */
1432 return context_has_address_families(c
) ||
1433 c
->memory_deny_write_execute
||
1434 c
->restrict_realtime
||
1435 c
->restrict_suid_sgid
||
1436 exec_context_restrict_namespaces_set(c
) ||
1438 c
->protect_kernel_tunables
||
1439 c
->protect_kernel_modules
||
1440 c
->protect_kernel_logs
||
1441 c
->private_devices
||
1442 context_has_syscall_filters(c
) ||
1443 context_has_syscall_logs(c
) ||
1444 !set_isempty(c
->syscall_archs
) ||
1445 c
->lock_personality
||
1446 c
->protect_hostname
;
1449 static bool exec_context_has_credentials(const ExecContext
*context
) {
1453 return !hashmap_isempty(context
->set_credentials
) ||
1454 context
->load_credentials
;
1459 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1461 if (is_seccomp_available())
1464 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1468 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1469 uint32_t negative_action
, default_action
, action
;
1475 if (!context_has_syscall_filters(c
))
1478 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1481 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1483 if (c
->syscall_allow_list
) {
1484 default_action
= negative_action
;
1485 action
= SCMP_ACT_ALLOW
;
1487 default_action
= SCMP_ACT_ALLOW
;
1488 action
= negative_action
;
1491 if (needs_ambient_hack
) {
1492 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1497 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1500 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1502 uint32_t default_action
, action
;
1508 if (!context_has_syscall_logs(c
))
1512 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1515 if (c
->syscall_log_allow_list
) {
1516 /* Log nothing but the ones listed */
1517 default_action
= SCMP_ACT_ALLOW
;
1518 action
= SCMP_ACT_LOG
;
1520 /* Log everything but the ones listed */
1521 default_action
= SCMP_ACT_LOG
;
1522 action
= SCMP_ACT_ALLOW
;
1525 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1527 /* old libseccomp */
1528 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1533 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1537 if (set_isempty(c
->syscall_archs
))
1540 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1543 return seccomp_restrict_archs(c
->syscall_archs
);
1546 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1550 if (!context_has_address_families(c
))
1553 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1556 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1559 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1563 if (!c
->memory_deny_write_execute
)
1566 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1569 return seccomp_memory_deny_write_execute();
1572 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1576 if (!c
->restrict_realtime
)
1579 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1582 return seccomp_restrict_realtime();
1585 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1589 if (!c
->restrict_suid_sgid
)
1592 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1595 return seccomp_restrict_suid_sgid();
1598 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1602 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1603 * let's protect even those systems where this is left on in the kernel. */
1605 if (!c
->protect_kernel_tunables
)
1608 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1611 return seccomp_protect_sysctl();
1614 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1618 /* Turn off module syscalls on ProtectKernelModules=yes */
1620 if (!c
->protect_kernel_modules
)
1623 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1626 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1629 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1633 if (!c
->protect_kernel_logs
)
1636 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1639 return seccomp_protect_syslog();
1642 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1646 if (!c
->protect_clock
)
1649 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1652 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1655 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1659 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1661 if (!c
->private_devices
)
1664 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1667 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1670 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1674 if (!exec_context_restrict_namespaces_set(c
))
1677 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1680 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1683 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1684 unsigned long personality
;
1690 if (!c
->lock_personality
)
1693 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1696 personality
= c
->personality
;
1698 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1699 if (personality
== PERSONALITY_INVALID
) {
1701 r
= opinionated_personality(&personality
);
1706 return seccomp_lock_personality(personality
);
1711 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1715 if (!c
->protect_hostname
)
1718 if (ns_type_supported(NAMESPACE_UTS
)) {
1719 if (unshare(CLONE_NEWUTS
) < 0) {
1720 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1721 *ret_exit_status
= EXIT_NAMESPACE
;
1722 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1725 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1728 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1733 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1736 r
= seccomp_protect_hostname();
1738 *ret_exit_status
= EXIT_SECCOMP
;
1739 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1746 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1749 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1750 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1752 if (idle_pipe
[0] >= 0) {
1755 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1757 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1760 /* Signal systemd that we are bored and want to continue. */
1761 n
= write(idle_pipe
[3], "x", 1);
1763 /* Wait for systemd to react to the signal above. */
1764 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1767 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1771 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1774 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1776 static int build_environment(
1778 const ExecContext
*c
,
1779 const ExecParameters
*p
,
1782 const char *username
,
1784 dev_t journal_stream_dev
,
1785 ino_t journal_stream_ino
,
1788 _cleanup_strv_free_
char **our_env
= NULL
;
1797 #define N_ENV_VARS 17
1798 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1803 _cleanup_free_
char *joined
= NULL
;
1805 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1807 our_env
[n_env
++] = x
;
1809 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1811 our_env
[n_env
++] = x
;
1813 joined
= strv_join(p
->fd_names
, ":");
1817 x
= strjoin("LISTEN_FDNAMES=", joined
);
1820 our_env
[n_env
++] = x
;
1823 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1824 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1826 our_env
[n_env
++] = x
;
1828 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1830 our_env
[n_env
++] = x
;
1833 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1834 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1835 * check the database directly. */
1836 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1837 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1840 our_env
[n_env
++] = x
;
1844 x
= strjoin("HOME=", home
);
1848 path_simplify(x
+ 5, true);
1849 our_env
[n_env
++] = x
;
1853 x
= strjoin("LOGNAME=", username
);
1856 our_env
[n_env
++] = x
;
1858 x
= strjoin("USER=", username
);
1861 our_env
[n_env
++] = x
;
1865 x
= strjoin("SHELL=", shell
);
1869 path_simplify(x
+ 6, true);
1870 our_env
[n_env
++] = x
;
1873 if (!sd_id128_is_null(u
->invocation_id
)) {
1874 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1877 our_env
[n_env
++] = x
;
1880 if (exec_context_needs_term(c
)) {
1881 const char *tty_path
, *term
= NULL
;
1883 tty_path
= exec_context_tty_path(c
);
1885 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1886 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1887 * container manager passes to PID 1 ends up all the way in the console login shown. */
1889 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1890 term
= getenv("TERM");
1893 term
= default_term_for_tty(tty_path
);
1895 x
= strjoin("TERM=", term
);
1898 our_env
[n_env
++] = x
;
1901 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1902 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1905 our_env
[n_env
++] = x
;
1908 if (c
->log_namespace
) {
1909 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1913 our_env
[n_env
++] = x
;
1916 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1917 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1923 if (strv_isempty(c
->directories
[t
].paths
))
1926 n
= exec_directory_env_name_to_string(t
);
1930 pre
= strjoin(p
->prefix
[t
], "/");
1934 joined
= strv_join_full(c
->directories
[t
].paths
, ":", pre
, true);
1938 x
= strjoin(n
, "=", joined
);
1942 our_env
[n_env
++] = x
;
1945 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
1946 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
1950 our_env
[n_env
++] = x
;
1953 if (asprintf(&x
, "SYSTEMD_EXEC_PID=" PID_FMT
, getpid_cached()) < 0)
1956 our_env
[n_env
++] = x
;
1958 our_env
[n_env
++] = NULL
;
1959 assert(n_env
<= N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1962 *ret
= TAKE_PTR(our_env
);
1967 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1968 _cleanup_strv_free_
char **pass_env
= NULL
;
1969 size_t n_env
= 0, n_bufsize
= 0;
1972 STRV_FOREACH(i
, c
->pass_environment
) {
1973 _cleanup_free_
char *x
= NULL
;
1979 x
= strjoin(*i
, "=", v
);
1983 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1986 pass_env
[n_env
++] = TAKE_PTR(x
);
1987 pass_env
[n_env
] = NULL
;
1990 *ret
= TAKE_PTR(pass_env
);
1995 bool exec_needs_mount_namespace(
1996 const ExecContext
*context
,
1997 const ExecParameters
*params
,
1998 const ExecRuntime
*runtime
) {
2002 if (context
->root_image
)
2005 if (!strv_isempty(context
->read_write_paths
) ||
2006 !strv_isempty(context
->read_only_paths
) ||
2007 !strv_isempty(context
->inaccessible_paths
) ||
2008 !strv_isempty(context
->exec_paths
) ||
2009 !strv_isempty(context
->no_exec_paths
))
2012 if (context
->n_bind_mounts
> 0)
2015 if (context
->n_temporary_filesystems
> 0)
2018 if (context
->n_mount_images
> 0)
2021 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
2024 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
2027 if (context
->private_devices
||
2028 context
->private_mounts
||
2029 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2030 context
->protect_home
!= PROTECT_HOME_NO
||
2031 context
->protect_kernel_tunables
||
2032 context
->protect_kernel_modules
||
2033 context
->protect_kernel_logs
||
2034 context
->protect_control_groups
||
2035 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2036 context
->proc_subset
!= PROC_SUBSET_ALL
)
2039 if (context
->root_directory
) {
2040 if (exec_context_get_effective_mount_apivfs(context
))
2043 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2044 if (params
&& !params
->prefix
[t
])
2047 if (!strv_isempty(context
->directories
[t
].paths
))
2052 if (context
->dynamic_user
&&
2053 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
2054 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
2055 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
2058 if (context
->log_namespace
)
2064 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2065 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2066 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
2067 _cleanup_close_
int unshare_ready_fd
= -1;
2068 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2073 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2074 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2075 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2076 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2077 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2078 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2079 * continues execution normally.
2080 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2081 * does not need CAP_SETUID to write the single line mapping to itself. */
2083 /* Can only set up multiple mappings with CAP_SETUID. */
2084 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
2085 r
= asprintf(&uid_map
,
2086 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2087 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2088 ouid
, ouid
, uid
, uid
);
2090 r
= asprintf(&uid_map
,
2091 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2097 /* Can only set up multiple mappings with CAP_SETGID. */
2098 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2099 r
= asprintf(&gid_map
,
2100 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2101 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2102 ogid
, ogid
, gid
, gid
);
2104 r
= asprintf(&gid_map
,
2105 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2111 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2113 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2114 if (unshare_ready_fd
< 0)
2117 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2119 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2122 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2126 _cleanup_close_
int fd
= -1;
2130 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2131 * here, after the parent opened its own user namespace. */
2134 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2136 /* Wait until the parent unshared the user namespace */
2137 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2142 /* Disable the setgroups() system call in the child user namespace, for good. */
2143 a
= procfs_file_alloca(ppid
, "setgroups");
2144 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2146 if (errno
!= ENOENT
) {
2151 /* If the file is missing the kernel is too old, let's continue anyway. */
2153 if (write(fd
, "deny\n", 5) < 0) {
2158 fd
= safe_close(fd
);
2161 /* First write the GID map */
2162 a
= procfs_file_alloca(ppid
, "gid_map");
2163 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2168 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2172 fd
= safe_close(fd
);
2174 /* The write the UID map */
2175 a
= procfs_file_alloca(ppid
, "uid_map");
2176 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2181 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2186 _exit(EXIT_SUCCESS
);
2189 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2190 _exit(EXIT_FAILURE
);
2193 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2195 if (unshare(CLONE_NEWUSER
) < 0)
2198 /* Let the child know that the namespace is ready now */
2199 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2202 /* Try to read an error code from the child */
2203 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2206 if (n
== sizeof(r
)) { /* an error code was sent to us */
2211 if (n
!= 0) /* on success we should have read 0 bytes */
2214 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2218 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2224 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2225 if (!context
->dynamic_user
)
2228 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2231 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2237 static int setup_exec_directory(
2238 const ExecContext
*context
,
2239 const ExecParameters
*params
,
2242 ExecDirectoryType type
,
2245 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2246 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2247 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2248 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2249 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2250 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2257 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2258 assert(exit_status
);
2260 if (!params
->prefix
[type
])
2263 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2264 if (!uid_is_valid(uid
))
2266 if (!gid_is_valid(gid
))
2270 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2271 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2273 p
= path_join(params
->prefix
[type
], *rt
);
2279 r
= mkdir_parents_label(p
, 0755);
2283 if (exec_directory_is_private(context
, type
)) {
2284 _cleanup_free_
char *private_root
= NULL
;
2286 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2287 * case we want to avoid leaving a directory around fully accessible that is owned by
2288 * a dynamic user whose UID is later on reused. To lock this down we use the same
2289 * trick used by container managers to prohibit host users to get access to files of
2290 * the same UID in containers: we place everything inside a directory that has an
2291 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2292 * for unprivileged host code. We then use fs namespacing to make this directory
2293 * permeable for the service itself.
2295 * Specifically: for a service which wants a special directory "foo/" we first create
2296 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2297 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2298 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2299 * unprivileged host users can't look into it. Inside of the namespace of the unit
2300 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2301 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2302 * for the service and making sure it only gets access to the dirs it needs but no
2303 * others. Tricky? Yes, absolutely, but it works!
2305 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2306 * to be owned by the service itself.
2308 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2309 * for sharing files or sockets with other services. */
2311 private_root
= path_join(params
->prefix
[type
], "private");
2312 if (!private_root
) {
2317 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2318 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2322 pp
= path_join(private_root
, *rt
);
2328 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2329 r
= mkdir_parents_label(pp
, 0755);
2333 if (is_dir(p
, false) > 0 &&
2334 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2336 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2337 * it over. Most likely the service has been upgraded from one that didn't use
2338 * DynamicUser=1, to one that does. */
2340 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2341 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2342 exec_directory_type_to_string(type
), p
, pp
);
2344 if (rename(p
, pp
) < 0) {
2349 /* Otherwise, create the actual directory for the service */
2351 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2352 if (r
< 0 && r
!= -EEXIST
)
2356 /* And link it up from the original place */
2357 r
= symlink_idempotent(pp
, p
, true);
2362 _cleanup_free_
char *target
= NULL
;
2364 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2365 readlink_and_make_absolute(p
, &target
) >= 0) {
2366 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2368 /* This already exists and is a symlink? Interesting. Maybe it's one created
2369 * by DynamicUser=1 (see above)?
2371 * We do this for all directory types except for ConfigurationDirectory=,
2372 * since they all support the private/ symlink logic at least in some
2373 * configurations, see above. */
2375 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2379 q
= path_join(params
->prefix
[type
], "private", *rt
);
2385 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2386 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2390 if (path_equal(q_resolved
, target_resolved
)) {
2392 /* Hmm, apparently DynamicUser= was once turned on for this service,
2393 * but is no longer. Let's move the directory back up. */
2395 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2396 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2397 exec_directory_type_to_string(type
), q
, p
);
2399 if (unlink(p
) < 0) {
2404 if (rename(q
, p
) < 0) {
2411 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2416 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2419 /* Don't change the owner/access mode of the configuration directory,
2420 * as in the common case it is not written to by a service, and shall
2421 * not be writable. */
2423 if (stat(p
, &st
) < 0) {
2428 /* Still complain if the access mode doesn't match */
2429 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2430 log_warning("%s \'%s\' already exists but the mode is different. "
2431 "(File system: %o %sMode: %o)",
2432 exec_directory_type_to_string(type
), *rt
,
2433 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2440 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2441 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2442 * current UID/GID ownership.) */
2443 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2447 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2448 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2449 * assignments to exist.*/
2450 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2458 *exit_status
= exit_status_table
[type
];
2462 static int write_credential(
2468 bool ownership_ok
) {
2470 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2471 _cleanup_close_
int fd
= -1;
2474 r
= tempfn_random_child("", "cred", &tmp
);
2478 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2484 r
= loop_write(fd
, data
, size
, /* do_pool = */ false);
2488 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2491 if (uid_is_valid(uid
) && uid
!= getuid()) {
2492 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2494 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2497 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2498 * to express: that the user gets read access and nothing
2499 * else. But if the backing fs can't support that (e.g. ramfs)
2500 * then we can use file ownership instead. But that's only safe if
2501 * we can then re-mount the whole thing read-only, so that the
2502 * user can no longer chmod() the file to gain write access. */
2505 if (fchown(fd
, uid
, (gid_t
) -1) < 0)
2510 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2517 #define CREDENTIALS_BYTES_MAX (1024LU * 1024LU) /* Refuse to pass more than 1M, after all this is unswappable memory */
2519 static int acquire_credentials(
2520 const ExecContext
*context
,
2521 const ExecParameters
*params
,
2525 bool ownership_ok
) {
2527 uint64_t left
= CREDENTIALS_BYTES_MAX
;
2528 _cleanup_close_
int dfd
= -1;
2529 ExecSetCredential
*sc
;
2536 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
2540 /* First we use the literally specified credentials. Note that they might be overridden again below,
2541 * and thus act as a "default" if the same credential is specified multiple times */
2542 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
2545 add
= strlen(sc
->id
) + sc
->size
;
2549 r
= write_credential(dfd
, sc
->id
, sc
->data
, sc
->size
, uid
, ownership_ok
);
2556 /* Then, load credential off disk (or acquire via AF_UNIX socket) */
2557 STRV_FOREACH_PAIR(id
, fn
, context
->load_credentials
) {
2558 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
2559 _cleanup_(erase_and_freep
) char *data
= NULL
;
2560 _cleanup_free_
char *j
= NULL
, *bindname
= NULL
;
2564 if (path_is_absolute(*fn
)) {
2565 /* If this is an absolute path, read the data directly from it, and support AF_UNIX sockets */
2567 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2569 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2570 * via the source socket address in case we read off an AF_UNIX socket. */
2571 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, *id
) < 0)
2574 } else if (params
->received_credentials
) {
2575 /* If this is a relative path, take it relative to the credentials we received
2576 * ourselves. We don't support the AF_UNIX stuff in this mode, since we are operating
2577 * on a credential store, i.e. this is guaranteed to be regular files. */
2578 j
= path_join(params
->received_credentials
, *fn
);
2588 r
= read_full_file_full(AT_FDCWD
, source
, UINT64_MAX
, SIZE_MAX
, flags
, bindname
, &data
, &size
);
2592 faccessat(dfd
, *id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0) /* If the source file doesn't exist, but we already acquired the key otherwise, then don't fail */
2597 add
= strlen(*id
) + size
;
2601 r
= write_credential(dfd
, *id
, data
, size
, uid
, ownership_ok
);
2608 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
2611 /* After we created all keys with the right perms, also make sure the credential store as a whole is
2614 if (uid_is_valid(uid
) && uid
!= getuid()) {
2615 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
2617 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2623 if (fchown(dfd
, uid
, (gid_t
) -1) < 0)
2631 static int setup_credentials_internal(
2632 const ExecContext
*context
,
2633 const ExecParameters
*params
,
2635 const char *final
, /* This is where the credential store shall eventually end up at */
2636 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
2637 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
2638 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
2641 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
2642 * if we mounted something; false if we definitely can't mount anything */
2650 if (reuse_workspace
) {
2651 r
= path_is_mount_point(workspace
, NULL
, 0);
2655 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
2657 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
2659 workspace_mounted
= -1; /* ditto */
2661 r
= path_is_mount_point(final
, NULL
, 0);
2665 /* If the final place already has something mounted, we use that. If the workspace also has
2666 * something mounted we assume it's actually the same mount (but with MS_RDONLY
2668 final_mounted
= true;
2670 if (workspace_mounted
< 0) {
2671 /* If the final place is mounted, but the workspace we isn't, then let's bind mount
2672 * the final version to the workspace, and make it writable, so that we can make
2675 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2679 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2683 workspace_mounted
= true;
2686 final_mounted
= false;
2688 if (workspace_mounted
< 0) {
2689 /* Nothing is mounted on the workspace yet, let's try to mount something now */
2690 for (int try = 0;; try++) {
2693 /* Try "ramfs" first, since it's not swap backed */
2694 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
2696 workspace_mounted
= true;
2700 } else if (try == 1) {
2701 _cleanup_free_
char *opts
= NULL
;
2703 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%lu", CREDENTIALS_BYTES_MAX
) < 0)
2706 /* Fall back to "tmpfs" otherwise */
2707 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
2709 workspace_mounted
= true;
2714 /* If that didn't work, try to make a bind mount from the final to the workspace, so that we can make it writable there. */
2715 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2717 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
2720 if (must_mount
) /* If we it's not OK to use the plain directory
2721 * fallback, propagate all errors too */
2724 /* If we lack privileges to bind mount stuff, then let's gracefully
2725 * proceed for compat with container envs, and just use the final dir
2728 workspace_mounted
= false;
2732 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
2733 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2737 workspace_mounted
= true;
2743 assert(!must_mount
|| workspace_mounted
> 0);
2744 where
= workspace_mounted
? workspace
: final
;
2746 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
2750 if (workspace_mounted
) {
2751 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
2752 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2756 /* And mount it to the final place, read-only */
2758 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
2760 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
2764 _cleanup_free_
char *parent
= NULL
;
2766 /* If we do not have our own mount put used the plain directory fallback, then we need to
2767 * open access to the top-level credential directory and the per-service directory now */
2769 parent
= dirname_malloc(final
);
2772 if (chmod(parent
, 0755) < 0)
2779 static int setup_credentials(
2780 const ExecContext
*context
,
2781 const ExecParameters
*params
,
2785 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
2792 if (!exec_context_has_credentials(context
))
2795 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
2798 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
2799 * and the subdir we mount over with a read-only file system readable by the service's user */
2800 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
2804 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
2805 if (r
< 0 && r
!= -EEXIST
)
2808 p
= path_join(q
, unit
);
2812 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
2813 if (r
< 0 && r
!= -EEXIST
)
2816 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
2818 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
2820 /* If this is not a privilege or support issue then propagate the error */
2821 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2824 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
2825 * it into place, so that users can't access half-initialized credential stores. */
2826 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
2830 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
2831 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
2832 * after it is fully set up */
2833 u
= path_join(t
, unit
);
2837 FOREACH_STRING(i
, t
, u
) {
2838 r
= mkdir_label(i
, 0700);
2839 if (r
< 0 && r
!= -EEXIST
)
2843 r
= setup_credentials_internal(
2847 p
, /* final mount point */
2848 u
, /* temporary workspace to overmount */
2849 true, /* reuse the workspace if it is already a mount */
2850 false, /* it's OK to fall back to a plain directory if we can't mount anything */
2853 (void) rmdir(u
); /* remove the workspace again if we can. */
2858 } else if (r
== 0) {
2860 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
2861 * we can use the same directory for all cases, after turning off propagation. Question
2862 * though is: where do we turn off propagation exactly, and where do we place the workspace
2863 * directory? We need some place that is guaranteed to be a mount point in the host, and
2864 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
2865 * since we ultimately want to move the resulting file system there, i.e. we need propagation
2866 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
2867 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
2868 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
2869 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
2870 * propagation on the former, and then overmount the latter.
2872 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
2873 * for this purpose, but there are few other candidates that work equally well for us, and
2874 * given that the we do this in a privately namespaced short-lived single-threaded process
2875 * that no one else sees this should be OK to do.*/
2877 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
2881 r
= setup_credentials_internal(
2885 p
, /* final mount point */
2886 "/dev/shm", /* temporary workspace to overmount */
2887 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
2888 true, /* insist that something is mounted, do not allow fallback to plain directory */
2893 _exit(EXIT_SUCCESS
);
2896 _exit(EXIT_FAILURE
);
2903 static int setup_smack(
2904 const ExecContext
*context
,
2905 int executable_fd
) {
2909 assert(executable_fd
>= 0);
2911 if (context
->smack_process_label
) {
2912 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2916 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2918 _cleanup_free_
char *exec_label
= NULL
;
2920 r
= mac_smack_read_fd(executable_fd
, SMACK_ATTR_EXEC
, &exec_label
);
2921 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2924 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2934 static int compile_bind_mounts(
2935 const ExecContext
*context
,
2936 const ExecParameters
*params
,
2937 BindMount
**ret_bind_mounts
,
2938 size_t *ret_n_bind_mounts
,
2939 char ***ret_empty_directories
) {
2941 _cleanup_strv_free_
char **empty_directories
= NULL
;
2942 BindMount
*bind_mounts
;
2948 assert(ret_bind_mounts
);
2949 assert(ret_n_bind_mounts
);
2950 assert(ret_empty_directories
);
2952 n
= context
->n_bind_mounts
;
2953 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2954 if (!params
->prefix
[t
])
2957 n
+= strv_length(context
->directories
[t
].paths
);
2961 *ret_bind_mounts
= NULL
;
2962 *ret_n_bind_mounts
= 0;
2963 *ret_empty_directories
= NULL
;
2967 bind_mounts
= new(BindMount
, n
);
2971 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
2972 BindMount
*item
= context
->bind_mounts
+ i
;
2975 s
= strdup(item
->source
);
2981 d
= strdup(item
->destination
);
2988 bind_mounts
[h
++] = (BindMount
) {
2991 .read_only
= item
->read_only
,
2992 .recursive
= item
->recursive
,
2993 .ignore_enoent
= item
->ignore_enoent
,
2997 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3000 if (!params
->prefix
[t
])
3003 if (strv_isempty(context
->directories
[t
].paths
))
3006 if (exec_directory_is_private(context
, t
) &&
3007 !exec_context_with_rootfs(context
)) {
3010 /* So this is for a dynamic user, and we need to make sure the process can access its own
3011 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3012 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3014 private_root
= path_join(params
->prefix
[t
], "private");
3015 if (!private_root
) {
3020 r
= strv_consume(&empty_directories
, private_root
);
3025 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
3028 if (exec_directory_is_private(context
, t
))
3029 s
= path_join(params
->prefix
[t
], "private", *suffix
);
3031 s
= path_join(params
->prefix
[t
], *suffix
);
3037 if (exec_directory_is_private(context
, t
) &&
3038 exec_context_with_rootfs(context
))
3039 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3040 * directory is not created on the root directory. So, let's bind-mount the directory
3041 * on the 'non-private' place. */
3042 d
= path_join(params
->prefix
[t
], *suffix
);
3051 bind_mounts
[h
++] = (BindMount
) {
3055 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3057 .ignore_enoent
= false,
3064 *ret_bind_mounts
= bind_mounts
;
3065 *ret_n_bind_mounts
= n
;
3066 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3071 bind_mount_free_many(bind_mounts
, h
);
3075 static bool insist_on_sandboxing(
3076 const ExecContext
*context
,
3077 const char *root_dir
,
3078 const char *root_image
,
3079 const BindMount
*bind_mounts
,
3080 size_t n_bind_mounts
) {
3083 assert(n_bind_mounts
== 0 || bind_mounts
);
3085 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3086 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3087 * rearrange stuff in a way we cannot ignore gracefully. */
3089 if (context
->n_temporary_filesystems
> 0)
3092 if (root_dir
|| root_image
)
3095 if (context
->n_mount_images
> 0)
3098 if (context
->dynamic_user
)
3101 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3103 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3104 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3107 if (context
->log_namespace
)
3113 static int apply_mount_namespace(
3115 ExecCommandFlags command_flags
,
3116 const ExecContext
*context
,
3117 const ExecParameters
*params
,
3118 const ExecRuntime
*runtime
,
3119 char **error_path
) {
3121 _cleanup_strv_free_
char **empty_directories
= NULL
;
3122 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3123 const char *root_dir
= NULL
, *root_image
= NULL
;
3124 _cleanup_free_
char *creds_path
= NULL
, *incoming_dir
= NULL
, *propagate_dir
= NULL
;
3125 NamespaceInfo ns_info
;
3126 bool needs_sandboxing
;
3127 BindMount
*bind_mounts
= NULL
;
3128 size_t n_bind_mounts
= 0;
3133 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3134 root_image
= context
->root_image
;
3137 root_dir
= context
->root_directory
;
3140 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3144 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3145 if (needs_sandboxing
) {
3146 /* The runtime struct only contains the parent of the private /tmp,
3147 * which is non-accessible to world users. Inside of it there's a /tmp
3148 * that is sticky, and that's the one we want to use here.
3149 * This does not apply when we are using /run/systemd/empty as fallback. */
3151 if (context
->private_tmp
&& runtime
) {
3152 if (streq_ptr(runtime
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3153 tmp_dir
= runtime
->tmp_dir
;
3154 else if (runtime
->tmp_dir
)
3155 tmp_dir
= strjoina(runtime
->tmp_dir
, "/tmp");
3157 if (streq_ptr(runtime
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3158 var_tmp_dir
= runtime
->var_tmp_dir
;
3159 else if (runtime
->var_tmp_dir
)
3160 var_tmp_dir
= strjoina(runtime
->var_tmp_dir
, "/tmp");
3163 ns_info
= (NamespaceInfo
) {
3164 .ignore_protect_paths
= false,
3165 .private_dev
= context
->private_devices
,
3166 .protect_control_groups
= context
->protect_control_groups
,
3167 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3168 .protect_kernel_modules
= context
->protect_kernel_modules
,
3169 .protect_kernel_logs
= context
->protect_kernel_logs
,
3170 .protect_hostname
= context
->protect_hostname
,
3171 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3172 .private_mounts
= context
->private_mounts
,
3173 .protect_home
= context
->protect_home
,
3174 .protect_system
= context
->protect_system
,
3175 .protect_proc
= context
->protect_proc
,
3176 .proc_subset
= context
->proc_subset
,
3178 } else if (!context
->dynamic_user
&& root_dir
)
3180 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3181 * sandbox info, otherwise enforce it, don't ignore protected paths and
3182 * fail if we are enable to apply the sandbox inside the mount namespace.
3184 ns_info
= (NamespaceInfo
) {
3185 .ignore_protect_paths
= true,
3188 ns_info
= (NamespaceInfo
) {};
3190 if (context
->mount_flags
== MS_SHARED
)
3191 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3193 if (exec_context_has_credentials(context
) &&
3194 params
->prefix
[EXEC_DIRECTORY_RUNTIME
] &&
3195 FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
3196 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3203 if (MANAGER_IS_SYSTEM(u
->manager
)) {
3204 propagate_dir
= path_join("/run/systemd/propagate/", u
->id
);
3207 incoming_dir
= strdup("/run/systemd/incoming");
3212 r
= setup_namespace(root_dir
, root_image
, context
->root_image_options
,
3213 &ns_info
, context
->read_write_paths
,
3214 needs_sandboxing
? context
->read_only_paths
: NULL
,
3215 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3216 needs_sandboxing
? context
->exec_paths
: NULL
,
3217 needs_sandboxing
? context
->no_exec_paths
: NULL
,
3221 context
->temporary_filesystems
,
3222 context
->n_temporary_filesystems
,
3223 context
->mount_images
,
3224 context
->n_mount_images
,
3228 context
->log_namespace
,
3229 context
->mount_flags
,
3230 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3231 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3232 context
->root_verity
,
3235 root_dir
|| root_image
? params
->notify_socket
: NULL
,
3236 DISSECT_IMAGE_DISCARD_ON_LOOP
|DISSECT_IMAGE_RELAX_VAR_CHECK
|DISSECT_IMAGE_FSCK
,
3239 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3240 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3241 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3242 * completely different execution environment. */
3244 if (insist_on_sandboxing(
3246 root_dir
, root_image
,
3249 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3250 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3251 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
3255 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3261 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
3265 static int apply_working_directory(
3266 const ExecContext
*context
,
3267 const ExecParameters
*params
,
3274 assert(exit_status
);
3276 if (context
->working_directory_home
) {
3279 *exit_status
= EXIT_CHDIR
;
3286 wd
= empty_to_root(context
->working_directory
);
3288 if (params
->flags
& EXEC_APPLY_CHROOT
)
3291 d
= prefix_roota(context
->root_directory
, wd
);
3293 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3294 *exit_status
= EXIT_CHDIR
;
3301 static int apply_root_directory(
3302 const ExecContext
*context
,
3303 const ExecParameters
*params
,
3304 const bool needs_mount_ns
,
3308 assert(exit_status
);
3310 if (params
->flags
& EXEC_APPLY_CHROOT
)
3311 if (!needs_mount_ns
&& context
->root_directory
)
3312 if (chroot(context
->root_directory
) < 0) {
3313 *exit_status
= EXIT_CHROOT
;
3320 static int setup_keyring(
3322 const ExecContext
*context
,
3323 const ExecParameters
*p
,
3324 uid_t uid
, gid_t gid
) {
3326 key_serial_t keyring
;
3335 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3336 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3337 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3338 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3339 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3340 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3342 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3345 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3346 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3347 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3348 * & group is just as nasty as acquiring a reference to the user keyring. */
3350 saved_uid
= getuid();
3351 saved_gid
= getgid();
3353 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3354 if (setregid(gid
, -1) < 0)
3355 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3358 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3359 if (setreuid(uid
, -1) < 0) {
3360 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3365 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3366 if (keyring
== -1) {
3367 if (errno
== ENOSYS
)
3368 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3369 else if (ERRNO_IS_PRIVILEGE(errno
))
3370 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3371 else if (errno
== EDQUOT
)
3372 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3374 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3379 /* When requested link the user keyring into the session keyring. */
3380 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3382 if (keyctl(KEYCTL_LINK
,
3383 KEY_SPEC_USER_KEYRING
,
3384 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3385 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3390 /* Restore uid/gid back */
3391 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3392 if (setreuid(saved_uid
, -1) < 0) {
3393 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3398 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3399 if (setregid(saved_gid
, -1) < 0)
3400 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3403 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3404 if (!sd_id128_is_null(u
->invocation_id
)) {
3407 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3409 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3411 if (keyctl(KEYCTL_SETPERM
, key
,
3412 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3413 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3414 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3419 /* Revert back uid & gid for the last time, and exit */
3420 /* no extra logging, as only the first already reported error matters */
3421 if (getuid() != saved_uid
)
3422 (void) setreuid(saved_uid
, -1);
3424 if (getgid() != saved_gid
)
3425 (void) setregid(saved_gid
, -1);
3430 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3436 array
[(*n
)++] = pair
[0];
3438 array
[(*n
)++] = pair
[1];
3441 static int close_remaining_fds(
3442 const ExecParameters
*params
,
3443 const ExecRuntime
*runtime
,
3444 const DynamicCreds
*dcreds
,
3447 const int *fds
, size_t n_fds
) {
3449 size_t n_dont_close
= 0;
3450 int dont_close
[n_fds
+ 12];
3454 if (params
->stdin_fd
>= 0)
3455 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3456 if (params
->stdout_fd
>= 0)
3457 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3458 if (params
->stderr_fd
>= 0)
3459 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3462 dont_close
[n_dont_close
++] = socket_fd
;
3464 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3465 n_dont_close
+= n_fds
;
3469 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3473 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3475 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3478 if (user_lookup_fd
>= 0)
3479 dont_close
[n_dont_close
++] = user_lookup_fd
;
3481 return close_all_fds(dont_close
, n_dont_close
);
3484 static int send_user_lookup(
3492 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3493 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3496 if (user_lookup_fd
< 0)
3499 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3502 if (writev(user_lookup_fd
,
3504 IOVEC_INIT(&uid
, sizeof(uid
)),
3505 IOVEC_INIT(&gid
, sizeof(gid
)),
3506 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3512 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3519 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3524 if (!c
->working_directory_home
)
3527 r
= get_home_dir(buf
);
3535 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3536 _cleanup_strv_free_
char ** list
= NULL
;
3543 assert(c
->dynamic_user
);
3545 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3546 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3549 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3552 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3558 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3561 if (exec_directory_is_private(c
, t
))
3562 e
= path_join(p
->prefix
[t
], "private", *i
);
3564 e
= path_join(p
->prefix
[t
], *i
);
3568 r
= strv_consume(&list
, e
);
3574 *ret
= TAKE_PTR(list
);
3579 static char *exec_command_line(char **argv
);
3581 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3582 bool using_subcgroup
;
3588 if (!params
->cgroup_path
)
3591 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3592 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3593 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3594 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3595 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3596 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3597 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3598 * flag, which is only passed for the former statements, not for the latter. */
3600 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3601 if (using_subcgroup
)
3602 p
= path_join(params
->cgroup_path
, ".control");
3604 p
= strdup(params
->cgroup_path
);
3609 return using_subcgroup
;
3612 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3613 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3619 if (!c
->numa_policy
.nodes
.set
) {
3620 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3624 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3630 return cpu_set_add_all(ret
, &s
);
3633 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3636 return c
->cpu_affinity_from_numa
;
3639 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
3644 assert(*n_fds
< fds_size
);
3652 if (fd
< 3 + (int) *n_fds
) {
3653 /* Let's move the fd up, so that it's outside of the fd range we will use to store
3654 * the fds we pass to the process (or which are closed only during execve). */
3656 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
3660 CLOSE_AND_REPLACE(fd
, r
);
3663 *ret_fd
= fds
[*n_fds
] = fd
;
3668 static int exec_child(
3670 const ExecCommand
*command
,
3671 const ExecContext
*context
,
3672 const ExecParameters
*params
,
3673 ExecRuntime
*runtime
,
3674 DynamicCreds
*dcreds
,
3676 const int named_iofds
[static 3],
3678 size_t n_socket_fds
,
3679 size_t n_storage_fds
,
3684 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3685 int r
, ngids
= 0, exec_fd
;
3686 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3687 const char *username
= NULL
, *groupname
= NULL
;
3688 _cleanup_free_
char *home_buffer
= NULL
;
3689 const char *home
= NULL
, *shell
= NULL
;
3690 char **final_argv
= NULL
;
3691 dev_t journal_stream_dev
= 0;
3692 ino_t journal_stream_ino
= 0;
3693 bool userns_set_up
= false;
3694 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3695 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3696 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3697 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3699 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3700 bool use_selinux
= false;
3703 bool use_smack
= false;
3706 bool use_apparmor
= false;
3708 uid_t saved_uid
= getuid();
3709 gid_t saved_gid
= getgid();
3710 uid_t uid
= UID_INVALID
;
3711 gid_t gid
= GID_INVALID
;
3712 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
3713 n_keep_fds
; /* total number of fds not to close */
3715 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3716 int ngids_after_pam
= 0;
3722 assert(exit_status
);
3724 rename_process_from_path(command
->path
);
3726 /* We reset exactly these signals, since they are the
3727 * only ones we set to SIG_IGN in the main daemon. All
3728 * others we leave untouched because we set them to
3729 * SIG_DFL or a valid handler initially, both of which
3730 * will be demoted to SIG_DFL. */
3731 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3732 SIGNALS_IGNORE
, -1);
3734 if (context
->ignore_sigpipe
)
3735 (void) ignore_signals(SIGPIPE
, -1);
3737 r
= reset_signal_mask();
3739 *exit_status
= EXIT_SIGNAL_MASK
;
3740 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3743 if (params
->idle_pipe
)
3744 do_idle_pipe_dance(params
->idle_pipe
);
3746 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3747 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3748 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3749 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3752 log_set_open_when_needed(true);
3754 /* In case anything used libc syslog(), close this here, too */
3757 int keep_fds
[n_fds
+ 2];
3758 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
3761 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
3763 *exit_status
= EXIT_FDS
;
3764 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
3767 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
3769 *exit_status
= EXIT_FDS
;
3770 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3773 if (!context
->same_pgrp
&&
3775 *exit_status
= EXIT_SETSID
;
3776 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3779 exec_context_tty_reset(context
, params
);
3781 if (unit_shall_confirm_spawn(unit
)) {
3782 const char *vc
= params
->confirm_spawn
;
3783 _cleanup_free_
char *cmdline
= NULL
;
3785 cmdline
= exec_command_line(command
->argv
);
3787 *exit_status
= EXIT_MEMORY
;
3791 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3792 if (r
!= CONFIRM_EXECUTE
) {
3793 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3794 *exit_status
= EXIT_SUCCESS
;
3797 *exit_status
= EXIT_CONFIRM
;
3798 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
3799 "Execution cancelled by the user");
3803 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3804 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3805 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3806 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3807 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3808 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3809 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3810 *exit_status
= EXIT_MEMORY
;
3811 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3814 if (context
->dynamic_user
&& dcreds
) {
3815 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3817 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3818 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
3819 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3820 *exit_status
= EXIT_USER
;
3821 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3824 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3826 *exit_status
= EXIT_MEMORY
;
3830 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3832 *exit_status
= EXIT_USER
;
3834 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
3835 "Failed to update dynamic user credentials: User or group with specified name already exists.");
3836 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3839 if (!uid_is_valid(uid
)) {
3840 *exit_status
= EXIT_USER
;
3841 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "UID validation failed for \""UID_FMT
"\"", uid
);
3844 if (!gid_is_valid(gid
)) {
3845 *exit_status
= EXIT_USER
;
3846 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "GID validation failed for \""GID_FMT
"\"", gid
);
3850 username
= dcreds
->user
->name
;
3853 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3855 *exit_status
= EXIT_USER
;
3856 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3859 r
= get_fixed_group(context
, &groupname
, &gid
);
3861 *exit_status
= EXIT_GROUP
;
3862 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3866 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3867 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3868 &supplementary_gids
, &ngids
);
3870 *exit_status
= EXIT_GROUP
;
3871 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3874 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3876 *exit_status
= EXIT_USER
;
3877 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3880 user_lookup_fd
= safe_close(user_lookup_fd
);
3882 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3884 *exit_status
= EXIT_CHDIR
;
3885 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3888 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3889 * must sure to drop O_NONBLOCK */
3891 (void) fd_nonblock(socket_fd
, false);
3893 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3894 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3895 if (params
->cgroup_path
) {
3896 _cleanup_free_
char *p
= NULL
;
3898 r
= exec_parameters_get_cgroup_path(params
, &p
);
3900 *exit_status
= EXIT_CGROUP
;
3901 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3904 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3906 *exit_status
= EXIT_CGROUP
;
3907 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3911 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3912 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3914 *exit_status
= EXIT_NETWORK
;
3915 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3919 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3921 *exit_status
= EXIT_STDIN
;
3922 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3925 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3927 *exit_status
= EXIT_STDOUT
;
3928 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3931 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3933 *exit_status
= EXIT_STDERR
;
3934 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3937 if (context
->oom_score_adjust_set
) {
3938 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3939 * prohibit write access to this file, and we shouldn't trip up over that. */
3940 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3941 if (ERRNO_IS_PRIVILEGE(r
))
3942 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3944 *exit_status
= EXIT_OOM_ADJUST
;
3945 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3949 if (context
->coredump_filter_set
) {
3950 r
= set_coredump_filter(context
->coredump_filter
);
3951 if (ERRNO_IS_PRIVILEGE(r
))
3952 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
3954 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
3957 if (context
->nice_set
) {
3958 r
= setpriority_closest(context
->nice
);
3960 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3963 if (context
->cpu_sched_set
) {
3964 struct sched_param param
= {
3965 .sched_priority
= context
->cpu_sched_priority
,
3968 r
= sched_setscheduler(0,
3969 context
->cpu_sched_policy
|
3970 (context
->cpu_sched_reset_on_fork
?
3971 SCHED_RESET_ON_FORK
: 0),
3974 *exit_status
= EXIT_SETSCHEDULER
;
3975 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3979 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
3980 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
3981 const CPUSet
*cpu_set
;
3983 if (context
->cpu_affinity_from_numa
) {
3984 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
3986 *exit_status
= EXIT_CPUAFFINITY
;
3987 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
3990 cpu_set
= &converted_cpu_set
;
3992 cpu_set
= &context
->cpu_set
;
3994 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
3995 *exit_status
= EXIT_CPUAFFINITY
;
3996 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
4000 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
4001 r
= apply_numa_policy(&context
->numa_policy
);
4002 if (r
== -EOPNOTSUPP
)
4003 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
4005 *exit_status
= EXIT_NUMA_POLICY
;
4006 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
4010 if (context
->ioprio_set
)
4011 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
4012 *exit_status
= EXIT_IOPRIO
;
4013 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
4016 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
4017 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
4018 *exit_status
= EXIT_TIMERSLACK
;
4019 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
4022 if (context
->personality
!= PERSONALITY_INVALID
) {
4023 r
= safe_personality(context
->personality
);
4025 *exit_status
= EXIT_PERSONALITY
;
4026 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
4030 if (context
->utmp_id
)
4031 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
4033 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
4034 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
4038 if (uid_is_valid(uid
)) {
4039 r
= chown_terminal(STDIN_FILENO
, uid
);
4041 *exit_status
= EXIT_STDIN
;
4042 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
4046 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4047 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4048 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4049 * touch a single hierarchy too. */
4050 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
4051 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
4053 *exit_status
= EXIT_CGROUP
;
4054 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4058 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4059 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
4061 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4064 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4065 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4067 *exit_status
= EXIT_CREDENTIALS
;
4068 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4072 r
= build_environment(
4084 *exit_status
= EXIT_MEMORY
;
4088 r
= build_pass_environment(context
, &pass_env
);
4090 *exit_status
= EXIT_MEMORY
;
4094 accum_env
= strv_env_merge(5,
4095 params
->environment
,
4098 context
->environment
,
4101 *exit_status
= EXIT_MEMORY
;
4104 accum_env
= strv_env_clean(accum_env
);
4106 (void) umask(context
->umask
);
4108 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4110 *exit_status
= EXIT_KEYRING
;
4111 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4114 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
4115 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4117 /* 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 */
4118 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4120 /* 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 */
4121 if (needs_ambient_hack
)
4122 needs_setuid
= false;
4124 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4126 if (needs_sandboxing
) {
4127 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
4128 * present. The actual MAC context application will happen later, as late as possible, to avoid
4129 * impacting our own code paths. */
4132 use_selinux
= mac_selinux_use();
4135 use_smack
= mac_smack_use();
4138 use_apparmor
= mac_apparmor_use();
4142 if (needs_sandboxing
) {
4145 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4146 * is set here. (See below.) */
4148 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4150 *exit_status
= EXIT_LIMITS
;
4151 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4155 if (needs_setuid
&& context
->pam_name
&& username
) {
4156 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4157 * wins here. (See above.) */
4159 /* All fds passed in the fds array will be closed in the pam child process. */
4160 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4162 *exit_status
= EXIT_PAM
;
4163 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4166 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4167 if (ngids_after_pam
< 0) {
4168 *exit_status
= EXIT_MEMORY
;
4169 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4173 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4174 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4175 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4176 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4178 userns_set_up
= true;
4179 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4181 *exit_status
= EXIT_USER
;
4182 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4186 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4188 if (ns_type_supported(NAMESPACE_NET
)) {
4189 r
= setup_netns(runtime
->netns_storage_socket
);
4191 log_unit_warning_errno(unit
, r
,
4192 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4194 *exit_status
= EXIT_NETWORK
;
4195 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4197 } else if (context
->network_namespace_path
) {
4198 *exit_status
= EXIT_NETWORK
;
4199 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4200 "NetworkNamespacePath= is not supported, refusing.");
4202 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4205 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4206 if (needs_mount_namespace
) {
4207 _cleanup_free_
char *error_path
= NULL
;
4209 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4211 *exit_status
= EXIT_NAMESPACE
;
4212 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4213 error_path
? ": " : "", strempty(error_path
));
4217 if (needs_sandboxing
) {
4218 r
= apply_protect_hostname(unit
, context
, exit_status
);
4223 /* Drop groups as early as possible.
4224 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4225 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4227 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4228 int ngids_to_enforce
= 0;
4230 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4235 if (ngids_to_enforce
< 0) {
4236 *exit_status
= EXIT_MEMORY
;
4237 return log_unit_error_errno(unit
,
4239 "Failed to merge group lists. Group membership might be incorrect: %m");
4242 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4244 *exit_status
= EXIT_GROUP
;
4245 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4249 /* If the user namespace was not set up above, try to do it now.
4250 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4251 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4252 * case of mount namespaces being less privileged when the mount point list is copied from a
4253 * different user namespace). */
4255 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4256 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4258 *exit_status
= EXIT_USER
;
4259 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4263 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4266 _cleanup_free_
char *executable
= NULL
;
4267 _cleanup_close_
int executable_fd
= -1;
4268 r
= find_executable_full(command
->path
, false, &executable
, &executable_fd
);
4270 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4271 log_struct_errno(LOG_INFO
, r
,
4272 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4274 LOG_UNIT_INVOCATION_ID(unit
),
4275 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4277 "EXECUTABLE=%s", command
->path
);
4281 *exit_status
= EXIT_EXEC
;
4282 return log_struct_errno(LOG_INFO
, r
,
4283 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4285 LOG_UNIT_INVOCATION_ID(unit
),
4286 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4288 "EXECUTABLE=%s", command
->path
);
4291 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, executable_fd
, &executable_fd
);
4293 *exit_status
= EXIT_FDS
;
4294 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4298 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
4299 r
= mac_selinux_get_child_mls_label(socket_fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4301 *exit_status
= EXIT_SELINUX_CONTEXT
;
4302 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4307 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4308 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
4309 * however if we have it as we want to keep it open until the final execve(). */
4311 r
= close_all_fds(keep_fds
, n_keep_fds
);
4313 r
= shift_fds(fds
, n_fds
);
4315 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4317 *exit_status
= EXIT_FDS
;
4318 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4321 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4322 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4323 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4326 secure_bits
= context
->secure_bits
;
4328 if (needs_sandboxing
) {
4331 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4332 * requested. (Note this is placed after the general resource limit initialization, see
4333 * above, in order to take precedence.) */
4334 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4335 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4336 *exit_status
= EXIT_LIMITS
;
4337 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4342 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4343 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4345 r
= setup_smack(context
, executable_fd
);
4347 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4348 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4353 bset
= context
->capability_bounding_set
;
4354 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4355 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4356 * instead of us doing that */
4357 if (needs_ambient_hack
)
4358 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4359 (UINT64_C(1) << CAP_SETUID
) |
4360 (UINT64_C(1) << CAP_SETGID
);
4362 if (!cap_test_all(bset
)) {
4363 r
= capability_bounding_set_drop(bset
, false);
4365 *exit_status
= EXIT_CAPABILITIES
;
4366 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4370 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4372 * To be able to raise the ambient capabilities after setresuid() they have to be
4373 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4374 * After setresuid() the ambient capabilities can be raised as they are present in
4375 * the permitted and inhertiable set. However it is possible that someone wants to
4376 * set ambient capabilities without changing the user, so we also set the ambient
4377 * capabilities here.
4378 * The requested ambient capabilities are raised in the inheritable set if the
4379 * second argument is true. */
4380 if (!needs_ambient_hack
) {
4381 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4383 *exit_status
= EXIT_CAPABILITIES
;
4384 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4389 /* chroot to root directory first, before we lose the ability to chroot */
4390 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4392 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4395 if (uid_is_valid(uid
)) {
4396 r
= enforce_user(context
, uid
);
4398 *exit_status
= EXIT_USER
;
4399 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4402 if (!needs_ambient_hack
&&
4403 context
->capability_ambient_set
!= 0) {
4405 /* Raise the ambient capabilities after user change. */
4406 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4408 *exit_status
= EXIT_CAPABILITIES
;
4409 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4415 /* Apply working directory here, because the working directory might be on NFS and only the user running
4416 * this service might have the correct privilege to change to the working directory */
4417 r
= apply_working_directory(context
, params
, home
, exit_status
);
4419 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4421 if (needs_sandboxing
) {
4422 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4423 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4424 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4425 * are restricted. */
4429 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4432 r
= setexeccon(exec_context
);
4434 *exit_status
= EXIT_SELINUX_CONTEXT
;
4435 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4442 if (use_apparmor
&& context
->apparmor_profile
) {
4443 r
= aa_change_onexec(context
->apparmor_profile
);
4444 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4445 *exit_status
= EXIT_APPARMOR_PROFILE
;
4446 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4451 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4452 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4454 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4455 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4456 * effective set here.
4457 * The effective set is overwritten during execve with the following values:
4458 * - ambient set (for non-root processes)
4459 * - (inheritable | bounding) set for root processes)
4461 * Hence there is no security impact to raise it in the effective set before execve
4463 r
= capability_gain_cap_setpcap(NULL
);
4465 *exit_status
= EXIT_CAPABILITIES
;
4466 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
4468 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
4469 *exit_status
= EXIT_SECUREBITS
;
4470 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
4474 if (context_has_no_new_privileges(context
))
4475 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
4476 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
4477 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
4481 r
= apply_address_families(unit
, context
);
4483 *exit_status
= EXIT_ADDRESS_FAMILIES
;
4484 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
4487 r
= apply_memory_deny_write_execute(unit
, context
);
4489 *exit_status
= EXIT_SECCOMP
;
4490 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
4493 r
= apply_restrict_realtime(unit
, context
);
4495 *exit_status
= EXIT_SECCOMP
;
4496 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
4499 r
= apply_restrict_suid_sgid(unit
, context
);
4501 *exit_status
= EXIT_SECCOMP
;
4502 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
4505 r
= apply_restrict_namespaces(unit
, context
);
4507 *exit_status
= EXIT_SECCOMP
;
4508 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
4511 r
= apply_protect_sysctl(unit
, context
);
4513 *exit_status
= EXIT_SECCOMP
;
4514 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
4517 r
= apply_protect_kernel_modules(unit
, context
);
4519 *exit_status
= EXIT_SECCOMP
;
4520 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
4523 r
= apply_protect_kernel_logs(unit
, context
);
4525 *exit_status
= EXIT_SECCOMP
;
4526 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
4529 r
= apply_protect_clock(unit
, context
);
4531 *exit_status
= EXIT_SECCOMP
;
4532 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
4535 r
= apply_private_devices(unit
, context
);
4537 *exit_status
= EXIT_SECCOMP
;
4538 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
4541 r
= apply_syscall_archs(unit
, context
);
4543 *exit_status
= EXIT_SECCOMP
;
4544 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
4547 r
= apply_lock_personality(unit
, context
);
4549 *exit_status
= EXIT_SECCOMP
;
4550 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
4553 r
= apply_syscall_log(unit
, context
);
4555 *exit_status
= EXIT_SECCOMP
;
4556 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
4559 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
4560 * by the filter as little as possible. */
4561 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
4563 *exit_status
= EXIT_SECCOMP
;
4564 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
4569 if (!strv_isempty(context
->unset_environment
)) {
4572 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
4574 *exit_status
= EXIT_MEMORY
;
4578 strv_free_and_replace(accum_env
, ee
);
4581 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
4582 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
4583 if (!replaced_argv
) {
4584 *exit_status
= EXIT_MEMORY
;
4587 final_argv
= replaced_argv
;
4589 final_argv
= command
->argv
;
4591 if (DEBUG_LOGGING
) {
4592 _cleanup_free_
char *line
;
4594 line
= exec_command_line(final_argv
);
4596 log_struct(LOG_DEBUG
,
4597 "EXECUTABLE=%s", executable
,
4598 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
4600 LOG_UNIT_INVOCATION_ID(unit
));
4606 /* We have finished with all our initializations. Let's now let the manager know that. From this point
4607 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
4609 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4610 *exit_status
= EXIT_EXEC
;
4611 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
4615 r
= fexecve_or_execve(executable_fd
, executable
, final_argv
, accum_env
);
4620 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4621 * that POLLHUP on it no longer means execve() succeeded. */
4623 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4624 *exit_status
= EXIT_EXEC
;
4625 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4629 *exit_status
= EXIT_EXEC
;
4630 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
4633 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4634 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4636 int exec_spawn(Unit
*unit
,
4637 ExecCommand
*command
,
4638 const ExecContext
*context
,
4639 const ExecParameters
*params
,
4640 ExecRuntime
*runtime
,
4641 DynamicCreds
*dcreds
,
4644 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4645 _cleanup_free_
char *subcgroup_path
= NULL
;
4646 _cleanup_strv_free_
char **files_env
= NULL
;
4647 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4648 _cleanup_free_
char *line
= NULL
;
4656 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4658 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4659 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4660 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4662 if (params
->n_socket_fds
> 1)
4663 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got more than one socket.");
4665 if (params
->n_socket_fds
== 0)
4666 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got no socket.");
4668 socket_fd
= params
->fds
[0];
4672 n_socket_fds
= params
->n_socket_fds
;
4673 n_storage_fds
= params
->n_storage_fds
;
4676 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4678 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4680 r
= exec_context_load_environment(unit
, context
, &files_env
);
4682 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4684 line
= exec_command_line(command
->argv
);
4688 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
4689 and, until the next SELinux policy changes, we save further reloads in future children. */
4690 mac_selinux_maybe_reload();
4692 log_struct(LOG_DEBUG
,
4693 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
4694 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
4695 the mount namespace in the child, but we want to log
4696 from the parent, so we need to use the (possibly
4697 inaccurate) path here. */
4699 LOG_UNIT_INVOCATION_ID(unit
));
4701 if (params
->cgroup_path
) {
4702 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4704 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4705 if (r
> 0) { /* We are using a child cgroup */
4706 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4708 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4710 /* Normally we would not propagate the oomd xattrs to children but since we created this
4711 * sub-cgroup internally we should do it. */
4712 cgroup_oomd_xattr_apply(unit
, subcgroup_path
);
4718 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4721 int exit_status
= EXIT_SUCCESS
;
4723 r
= exec_child(unit
,
4735 unit
->manager
->user_lookup_fds
[1],
4739 const char *status
=
4740 exit_status_to_string(exit_status
,
4741 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4743 log_struct_errno(LOG_ERR
, r
,
4744 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4746 LOG_UNIT_INVOCATION_ID(unit
),
4747 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4748 status
, command
->path
),
4749 "EXECUTABLE=%s", command
->path
);
4755 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4757 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4758 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4759 * process will be killed too). */
4761 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4763 exec_status_start(&command
->exec_status
, pid
);
4769 void exec_context_init(ExecContext
*c
) {
4773 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4774 c
->cpu_sched_policy
= SCHED_OTHER
;
4775 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4776 c
->syslog_level_prefix
= true;
4777 c
->ignore_sigpipe
= true;
4778 c
->timer_slack_nsec
= NSEC_INFINITY
;
4779 c
->personality
= PERSONALITY_INVALID
;
4780 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4781 c
->directories
[t
].mode
= 0755;
4782 c
->timeout_clean_usec
= USEC_INFINITY
;
4783 c
->capability_bounding_set
= CAP_ALL
;
4784 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4785 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4786 c
->log_level_max
= -1;
4788 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
4790 numa_policy_reset(&c
->numa_policy
);
4793 void exec_context_done(ExecContext
*c
) {
4796 c
->environment
= strv_free(c
->environment
);
4797 c
->environment_files
= strv_free(c
->environment_files
);
4798 c
->pass_environment
= strv_free(c
->pass_environment
);
4799 c
->unset_environment
= strv_free(c
->unset_environment
);
4801 rlimit_free_all(c
->rlimit
);
4803 for (size_t l
= 0; l
< 3; l
++) {
4804 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4805 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4808 c
->working_directory
= mfree(c
->working_directory
);
4809 c
->root_directory
= mfree(c
->root_directory
);
4810 c
->root_image
= mfree(c
->root_image
);
4811 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
4812 c
->root_hash
= mfree(c
->root_hash
);
4813 c
->root_hash_size
= 0;
4814 c
->root_hash_path
= mfree(c
->root_hash_path
);
4815 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
4816 c
->root_hash_sig_size
= 0;
4817 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
4818 c
->root_verity
= mfree(c
->root_verity
);
4819 c
->tty_path
= mfree(c
->tty_path
);
4820 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4821 c
->user
= mfree(c
->user
);
4822 c
->group
= mfree(c
->group
);
4824 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4826 c
->pam_name
= mfree(c
->pam_name
);
4828 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4829 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4830 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4831 c
->exec_paths
= strv_free(c
->exec_paths
);
4832 c
->no_exec_paths
= strv_free(c
->no_exec_paths
);
4834 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4835 c
->bind_mounts
= NULL
;
4836 c
->n_bind_mounts
= 0;
4837 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4838 c
->temporary_filesystems
= NULL
;
4839 c
->n_temporary_filesystems
= 0;
4840 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
4842 cpu_set_reset(&c
->cpu_set
);
4843 numa_policy_reset(&c
->numa_policy
);
4845 c
->utmp_id
= mfree(c
->utmp_id
);
4846 c
->selinux_context
= mfree(c
->selinux_context
);
4847 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4848 c
->smack_process_label
= mfree(c
->smack_process_label
);
4850 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4851 c
->syscall_archs
= set_free(c
->syscall_archs
);
4852 c
->address_families
= set_free(c
->address_families
);
4854 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4855 c
->directories
[t
].paths
= strv_free(c
->directories
[t
].paths
);
4857 c
->log_level_max
= -1;
4859 exec_context_free_log_extra_fields(c
);
4861 c
->log_ratelimit_interval_usec
= 0;
4862 c
->log_ratelimit_burst
= 0;
4864 c
->stdin_data
= mfree(c
->stdin_data
);
4865 c
->stdin_data_size
= 0;
4867 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4869 c
->log_namespace
= mfree(c
->log_namespace
);
4871 c
->load_credentials
= strv_free(c
->load_credentials
);
4872 c
->set_credentials
= hashmap_free(c
->set_credentials
);
4875 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4880 if (!runtime_prefix
)
4883 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4884 _cleanup_free_
char *p
;
4886 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4887 p
= path_join(runtime_prefix
, "private", *i
);
4889 p
= path_join(runtime_prefix
, *i
);
4893 /* We execute this synchronously, since we need to be sure this is gone when we start the
4895 (void) rm_rf(p
, REMOVE_ROOT
);
4901 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
4902 _cleanup_free_
char *p
= NULL
;
4906 if (!runtime_prefix
|| !unit
)
4909 p
= path_join(runtime_prefix
, "credentials", unit
);
4913 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
4914 * unmount it, and afterwards remove the mount point */
4915 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
4916 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
4921 static void exec_command_done(ExecCommand
*c
) {
4924 c
->path
= mfree(c
->path
);
4925 c
->argv
= strv_free(c
->argv
);
4928 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4929 for (size_t i
= 0; i
< n
; i
++)
4930 exec_command_done(c
+i
);
4933 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4937 LIST_REMOVE(command
, c
, i
);
4938 exec_command_done(i
);
4945 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4946 for (size_t i
= 0; i
< n
; i
++)
4947 c
[i
] = exec_command_free_list(c
[i
]);
4950 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4951 for (size_t i
= 0; i
< n
; i
++)
4952 exec_status_reset(&c
[i
].exec_status
);
4955 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
4956 for (size_t i
= 0; i
< n
; i
++) {
4959 LIST_FOREACH(command
, z
, c
[i
])
4960 exec_status_reset(&z
->exec_status
);
4964 typedef struct InvalidEnvInfo
{
4969 static void invalid_env(const char *p
, void *userdata
) {
4970 InvalidEnvInfo
*info
= userdata
;
4972 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4975 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4981 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4984 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4987 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4990 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4993 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4996 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
5003 static int exec_context_named_iofds(
5004 const ExecContext
*c
,
5005 const ExecParameters
*p
,
5006 int named_iofds
[static 3]) {
5009 const char* stdio_fdname
[3];
5014 assert(named_iofds
);
5016 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
5017 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
5018 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
5020 for (size_t i
= 0; i
< 3; i
++)
5021 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
5023 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
5025 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
5026 if (named_iofds
[STDIN_FILENO
] < 0 &&
5027 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
5028 stdio_fdname
[STDIN_FILENO
] &&
5029 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
5031 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
5034 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
5035 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
5036 stdio_fdname
[STDOUT_FILENO
] &&
5037 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
5039 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5042 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5043 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5044 stdio_fdname
[STDERR_FILENO
] &&
5045 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5047 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5051 return targets
== 0 ? 0 : -ENOENT
;
5054 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
5055 char **i
, **r
= NULL
;
5060 STRV_FOREACH(i
, c
->environment_files
) {
5063 bool ignore
= false;
5065 _cleanup_globfree_ glob_t pglob
= {};
5074 if (!path_is_absolute(fn
)) {
5082 /* Filename supports globbing, take all matching files */
5083 k
= safe_glob(fn
, 0, &pglob
);
5092 /* When we don't match anything, -ENOENT should be returned */
5093 assert(pglob
.gl_pathc
> 0);
5095 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5096 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5104 /* Log invalid environment variables with filename */
5106 InvalidEnvInfo info
= {
5108 .path
= pglob
.gl_pathv
[n
]
5111 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5119 m
= strv_env_merge(2, r
, p
);
5135 static bool tty_may_match_dev_console(const char *tty
) {
5136 _cleanup_free_
char *resolved
= NULL
;
5141 tty
= skip_dev_prefix(tty
);
5143 /* trivial identity? */
5144 if (streq(tty
, "console"))
5147 if (resolve_dev_console(&resolved
) < 0)
5148 return true; /* if we could not resolve, assume it may */
5150 /* "tty0" means the active VC, so it may be the same sometimes */
5151 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5154 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5157 return ec
->tty_reset
||
5159 ec
->tty_vt_disallocate
||
5160 is_terminal_input(ec
->std_input
) ||
5161 is_terminal_output(ec
->std_output
) ||
5162 is_terminal_output(ec
->std_error
);
5165 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5167 return exec_context_may_touch_tty(ec
) &&
5168 tty_may_match_dev_console(exec_context_tty_path(ec
));
5171 static void strv_fprintf(FILE *f
, char **l
) {
5177 fprintf(f
, " %s", *g
);
5180 static void strv_dump(FILE* f
, const char *prefix
, const char *name
, char **strv
) {
5185 if (!strv_isempty(strv
)) {
5186 fprintf(f
, "%s%s:", name
, prefix
);
5187 strv_fprintf(f
, strv
);
5192 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5193 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
5199 prefix
= strempty(prefix
);
5203 "%sWorkingDirectory: %s\n"
5204 "%sRootDirectory: %s\n"
5205 "%sNonBlocking: %s\n"
5206 "%sPrivateTmp: %s\n"
5207 "%sPrivateDevices: %s\n"
5208 "%sProtectKernelTunables: %s\n"
5209 "%sProtectKernelModules: %s\n"
5210 "%sProtectKernelLogs: %s\n"
5211 "%sProtectClock: %s\n"
5212 "%sProtectControlGroups: %s\n"
5213 "%sPrivateNetwork: %s\n"
5214 "%sPrivateUsers: %s\n"
5215 "%sProtectHome: %s\n"
5216 "%sProtectSystem: %s\n"
5217 "%sMountAPIVFS: %s\n"
5218 "%sIgnoreSIGPIPE: %s\n"
5219 "%sMemoryDenyWriteExecute: %s\n"
5220 "%sRestrictRealtime: %s\n"
5221 "%sRestrictSUIDSGID: %s\n"
5222 "%sKeyringMode: %s\n"
5223 "%sProtectHostname: %s\n"
5224 "%sProtectProc: %s\n"
5225 "%sProcSubset: %s\n",
5227 prefix
, empty_to_root(c
->working_directory
),
5228 prefix
, empty_to_root(c
->root_directory
),
5229 prefix
, yes_no(c
->non_blocking
),
5230 prefix
, yes_no(c
->private_tmp
),
5231 prefix
, yes_no(c
->private_devices
),
5232 prefix
, yes_no(c
->protect_kernel_tunables
),
5233 prefix
, yes_no(c
->protect_kernel_modules
),
5234 prefix
, yes_no(c
->protect_kernel_logs
),
5235 prefix
, yes_no(c
->protect_clock
),
5236 prefix
, yes_no(c
->protect_control_groups
),
5237 prefix
, yes_no(c
->private_network
),
5238 prefix
, yes_no(c
->private_users
),
5239 prefix
, protect_home_to_string(c
->protect_home
),
5240 prefix
, protect_system_to_string(c
->protect_system
),
5241 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
5242 prefix
, yes_no(c
->ignore_sigpipe
),
5243 prefix
, yes_no(c
->memory_deny_write_execute
),
5244 prefix
, yes_no(c
->restrict_realtime
),
5245 prefix
, yes_no(c
->restrict_suid_sgid
),
5246 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5247 prefix
, yes_no(c
->protect_hostname
),
5248 prefix
, protect_proc_to_string(c
->protect_proc
),
5249 prefix
, proc_subset_to_string(c
->proc_subset
));
5252 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5254 if (c
->root_image_options
) {
5257 fprintf(f
, "%sRootImageOptions:", prefix
);
5258 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5259 if (!isempty(o
->options
))
5260 fprintf(f
, " %s:%s",
5261 partition_designator_to_string(o
->partition_designator
),
5267 _cleanup_free_
char *encoded
= NULL
;
5268 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5270 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5273 if (c
->root_hash_path
)
5274 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5276 if (c
->root_hash_sig
) {
5277 _cleanup_free_
char *encoded
= NULL
;
5279 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5281 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5284 if (c
->root_hash_sig_path
)
5285 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5288 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5290 STRV_FOREACH(e
, c
->environment
)
5291 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5293 STRV_FOREACH(e
, c
->environment_files
)
5294 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5296 STRV_FOREACH(e
, c
->pass_environment
)
5297 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5299 STRV_FOREACH(e
, c
->unset_environment
)
5300 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5302 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5304 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5305 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5307 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
5308 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
5312 "%sTimeoutCleanSec: %s\n",
5313 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
5320 if (c
->oom_score_adjust_set
)
5322 "%sOOMScoreAdjust: %i\n",
5323 prefix
, c
->oom_score_adjust
);
5325 if (c
->coredump_filter_set
)
5327 "%sCoredumpFilter: 0x%"PRIx64
"\n",
5328 prefix
, c
->coredump_filter
);
5330 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5332 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5333 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5334 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5335 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5338 if (c
->ioprio_set
) {
5339 _cleanup_free_
char *class_str
= NULL
;
5341 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
5343 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5345 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
5348 if (c
->cpu_sched_set
) {
5349 _cleanup_free_
char *policy_str
= NULL
;
5351 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5353 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5356 "%sCPUSchedulingPriority: %i\n"
5357 "%sCPUSchedulingResetOnFork: %s\n",
5358 prefix
, c
->cpu_sched_priority
,
5359 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5362 if (c
->cpu_set
.set
) {
5363 _cleanup_free_
char *affinity
= NULL
;
5365 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5366 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5369 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5370 _cleanup_free_
char *nodes
= NULL
;
5372 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5373 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5374 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5377 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5378 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5381 "%sStandardInput: %s\n"
5382 "%sStandardOutput: %s\n"
5383 "%sStandardError: %s\n",
5384 prefix
, exec_input_to_string(c
->std_input
),
5385 prefix
, exec_output_to_string(c
->std_output
),
5386 prefix
, exec_output_to_string(c
->std_error
));
5388 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5389 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5390 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5391 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5392 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5393 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5395 if (c
->std_input
== EXEC_INPUT_FILE
)
5396 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5397 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5398 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5399 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5400 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5401 if (c
->std_output
== EXEC_OUTPUT_FILE_TRUNCATE
)
5402 fprintf(f
, "%sStandardOutputFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5403 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5404 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5405 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5406 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5407 if (c
->std_error
== EXEC_OUTPUT_FILE_TRUNCATE
)
5408 fprintf(f
, "%sStandardErrorFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5414 "%sTTYVHangup: %s\n"
5415 "%sTTYVTDisallocate: %s\n",
5416 prefix
, c
->tty_path
,
5417 prefix
, yes_no(c
->tty_reset
),
5418 prefix
, yes_no(c
->tty_vhangup
),
5419 prefix
, yes_no(c
->tty_vt_disallocate
));
5421 if (IN_SET(c
->std_output
,
5423 EXEC_OUTPUT_JOURNAL
,
5424 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5425 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5426 IN_SET(c
->std_error
,
5428 EXEC_OUTPUT_JOURNAL
,
5429 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5430 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5432 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5434 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5436 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5438 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5440 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5443 if (c
->log_level_max
>= 0) {
5444 _cleanup_free_
char *t
= NULL
;
5446 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5448 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5451 if (c
->log_ratelimit_interval_usec
> 0) {
5452 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
5455 "%sLogRateLimitIntervalSec: %s\n",
5456 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
5459 if (c
->log_ratelimit_burst
> 0)
5460 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
5462 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
5463 fprintf(f
, "%sLogExtraFields: ", prefix
);
5464 fwrite(c
->log_extra_fields
[j
].iov_base
,
5465 1, c
->log_extra_fields
[j
].iov_len
,
5470 if (c
->log_namespace
)
5471 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
5473 if (c
->secure_bits
) {
5474 _cleanup_free_
char *str
= NULL
;
5476 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
5478 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
5481 if (c
->capability_bounding_set
!= CAP_ALL
) {
5482 _cleanup_free_
char *str
= NULL
;
5484 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
5486 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
5489 if (c
->capability_ambient_set
!= 0) {
5490 _cleanup_free_
char *str
= NULL
;
5492 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
5494 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
5498 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
5500 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
5502 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
5504 strv_dump(f
, prefix
, "SupplementaryGroups", c
->supplementary_groups
);
5507 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
5509 strv_dump(f
, prefix
, "ReadWritePaths", c
->read_write_paths
);
5510 strv_dump(f
, prefix
, "ReadOnlyPaths", c
->read_only_paths
);
5511 strv_dump(f
, prefix
, "InaccessiblePaths", c
->inaccessible_paths
);
5512 strv_dump(f
, prefix
, "ExecPaths", c
->exec_paths
);
5513 strv_dump(f
, prefix
, "NoExecPaths", c
->no_exec_paths
);
5515 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
5516 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
5517 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
5518 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
5519 c
->bind_mounts
[i
].source
,
5520 c
->bind_mounts
[i
].destination
,
5521 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
5523 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
5524 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
5526 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
5528 isempty(t
->options
) ? "" : ":",
5529 strempty(t
->options
));
5534 "%sUtmpIdentifier: %s\n",
5535 prefix
, c
->utmp_id
);
5537 if (c
->selinux_context
)
5539 "%sSELinuxContext: %s%s\n",
5540 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
5542 if (c
->apparmor_profile
)
5544 "%sAppArmorProfile: %s%s\n",
5545 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
5547 if (c
->smack_process_label
)
5549 "%sSmackProcessLabel: %s%s\n",
5550 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
5552 if (c
->personality
!= PERSONALITY_INVALID
)
5554 "%sPersonality: %s\n",
5555 prefix
, strna(personality_to_string(c
->personality
)));
5558 "%sLockPersonality: %s\n",
5559 prefix
, yes_no(c
->lock_personality
));
5561 if (c
->syscall_filter
) {
5568 "%sSystemCallFilter: ",
5571 if (!c
->syscall_allow_list
)
5575 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
5576 _cleanup_free_
char *name
= NULL
;
5577 const char *errno_name
= NULL
;
5578 int num
= PTR_TO_INT(val
);
5585 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
5586 fputs(strna(name
), f
);
5589 errno_name
= seccomp_errno_or_action_to_string(num
);
5591 fprintf(f
, ":%s", errno_name
);
5593 fprintf(f
, ":%d", num
);
5601 if (c
->syscall_archs
) {
5607 "%sSystemCallArchitectures:",
5611 SET_FOREACH(id
, c
->syscall_archs
)
5612 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
5617 if (exec_context_restrict_namespaces_set(c
)) {
5618 _cleanup_free_
char *s
= NULL
;
5620 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
5622 fprintf(f
, "%sRestrictNamespaces: %s\n",
5626 if (c
->network_namespace_path
)
5628 "%sNetworkNamespacePath: %s\n",
5629 prefix
, c
->network_namespace_path
);
5631 if (c
->syscall_errno
> 0) {
5633 const char *errno_name
;
5636 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
5639 errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
5641 fputs(errno_name
, f
);
5643 fprintf(f
, "%d", c
->syscall_errno
);
5648 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
5651 fprintf(f
, "%sMountImages: %s%s:%s%s", prefix
,
5652 c
->mount_images
[i
].ignore_enoent
? "-": "",
5653 c
->mount_images
[i
].source
,
5654 c
->mount_images
[i
].destination
,
5655 LIST_IS_EMPTY(c
->mount_images
[i
].mount_options
) ? "": ":");
5656 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
5658 partition_designator_to_string(o
->partition_designator
),
5664 bool exec_context_maintains_privileges(const ExecContext
*c
) {
5667 /* Returns true if the process forked off would run under
5668 * an unchanged UID or as root. */
5673 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
5679 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
5687 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5689 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5694 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
5697 /* Explicit setting wins */
5698 if (c
->mount_apivfs_set
)
5699 return c
->mount_apivfs
;
5701 /* Default to "yes" if root directory or image are specified */
5702 if (exec_context_with_rootfs(c
))
5708 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5711 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
5712 free(c
->log_extra_fields
[l
].iov_base
);
5713 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5714 c
->n_log_extra_fields
= 0;
5717 void exec_context_revert_tty(ExecContext
*c
) {
5722 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5723 exec_context_tty_reset(c
, NULL
);
5725 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5726 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5727 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5729 if (exec_context_may_touch_tty(c
)) {
5732 path
= exec_context_tty_path(c
);
5734 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
5735 if (r
< 0 && r
!= -ENOENT
)
5736 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5741 int exec_context_get_clean_directories(
5747 _cleanup_strv_free_
char **l
= NULL
;
5754 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5757 if (!FLAGS_SET(mask
, 1U << t
))
5763 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5766 j
= path_join(prefix
[t
], *i
);
5770 r
= strv_consume(&l
, j
);
5774 /* Also remove private directories unconditionally. */
5775 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5776 j
= path_join(prefix
[t
], "private", *i
);
5780 r
= strv_consume(&l
, j
);
5791 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5792 ExecCleanMask mask
= 0;
5797 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5798 if (!strv_isempty(c
->directories
[t
].paths
))
5805 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5812 dual_timestamp_get(&s
->start_timestamp
);
5815 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5823 dual_timestamp_get(&s
->exit_timestamp
);
5828 if (context
&& context
->utmp_id
)
5829 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5832 void exec_status_reset(ExecStatus
*s
) {
5835 *s
= (ExecStatus
) {};
5838 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5839 char buf
[FORMAT_TIMESTAMP_MAX
];
5847 prefix
= strempty(prefix
);
5850 "%sPID: "PID_FMT
"\n",
5853 if (dual_timestamp_is_set(&s
->start_timestamp
))
5855 "%sStart Timestamp: %s\n",
5856 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5858 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5860 "%sExit Timestamp: %s\n"
5862 "%sExit Status: %i\n",
5863 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5864 prefix
, sigchld_code_to_string(s
->code
),
5868 static char *exec_command_line(char **argv
) {
5876 STRV_FOREACH(a
, argv
)
5884 STRV_FOREACH(a
, argv
) {
5891 if (strpbrk(*a
, WHITESPACE
)) {
5902 /* FIXME: this doesn't really handle arguments that have
5903 * spaces and ticks in them */
5908 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5909 _cleanup_free_
char *cmd
= NULL
;
5910 const char *prefix2
;
5915 prefix
= strempty(prefix
);
5916 prefix2
= strjoina(prefix
, "\t");
5918 cmd
= exec_command_line(c
->argv
);
5920 "%sCommand Line: %s\n",
5921 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
5923 exec_status_dump(&c
->exec_status
, f
, prefix2
);
5926 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5929 prefix
= strempty(prefix
);
5931 LIST_FOREACH(command
, c
, c
)
5932 exec_command_dump(c
, f
, prefix
);
5935 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
5942 /* It's kind of important, that we keep the order here */
5943 LIST_FIND_TAIL(command
, *l
, end
);
5944 LIST_INSERT_AFTER(command
, *l
, end
, e
);
5949 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
5957 l
= strv_new_ap(path
, ap
);
5969 free_and_replace(c
->path
, p
);
5971 return strv_free_and_replace(c
->argv
, l
);
5974 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
5975 _cleanup_strv_free_
char **l
= NULL
;
5983 l
= strv_new_ap(path
, ap
);
5989 r
= strv_extend_strv(&c
->argv
, l
, false);
5996 static void *remove_tmpdir_thread(void *p
) {
5997 _cleanup_free_
char *path
= p
;
5999 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
6003 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
6010 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
6012 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
6014 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6015 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
6017 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
6019 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
6024 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6025 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
6027 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
6029 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
6031 rt
->var_tmp_dir
= NULL
;
6034 rt
->id
= mfree(rt
->id
);
6035 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
6036 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
6037 safe_close_pair(rt
->netns_storage_socket
);
6041 static void exec_runtime_freep(ExecRuntime
**rt
) {
6042 (void) exec_runtime_free(*rt
, false);
6045 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
6046 _cleanup_free_
char *id_copy
= NULL
;
6051 id_copy
= strdup(id
);
6055 n
= new(ExecRuntime
, 1);
6059 *n
= (ExecRuntime
) {
6060 .id
= TAKE_PTR(id_copy
),
6061 .netns_storage_socket
= { -1, -1 },
6068 static int exec_runtime_add(
6073 int netns_storage_socket
[2],
6074 ExecRuntime
**ret
) {
6076 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6082 /* tmp_dir, var_tmp_dir, netns_storage_socket fds are donated on success */
6084 r
= exec_runtime_allocate(&rt
, id
);
6088 r
= hashmap_ensure_put(&m
->exec_runtime_by_id
, &string_hash_ops
, rt
->id
, rt
);
6092 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6093 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6094 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6096 if (netns_storage_socket
) {
6097 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6098 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6105 /* do not remove created ExecRuntime object when the operation succeeds. */
6110 static int exec_runtime_make(
6112 const ExecContext
*c
,
6114 ExecRuntime
**ret
) {
6116 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6117 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
6124 /* It is not necessary to create ExecRuntime object. */
6125 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
) {
6130 if (c
->private_tmp
&&
6131 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6132 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6133 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6134 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6139 if (c
->private_network
|| c
->network_namespace_path
) {
6140 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6144 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ret
);
6151 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6159 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6161 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
6169 /* If not found, then create a new object. */
6170 r
= exec_runtime_make(m
, c
, id
, &rt
);
6174 /* When r == 0, it is not necessary to create ExecRuntime object. */
6180 /* increment reference counter. */
6186 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6190 assert(rt
->n_ref
> 0);
6196 return exec_runtime_free(rt
, destroy
);
6199 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6206 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6207 fprintf(f
, "exec-runtime=%s", rt
->id
);
6210 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6212 if (rt
->var_tmp_dir
)
6213 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6215 if (rt
->netns_storage_socket
[0] >= 0) {
6218 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6222 fprintf(f
, " netns-socket-0=%i", copy
);
6225 if (rt
->netns_storage_socket
[1] >= 0) {
6228 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6232 fprintf(f
, " netns-socket-1=%i", copy
);
6241 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6242 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6246 /* This is for the migration from old (v237 or earlier) deserialization text.
6247 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6248 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6249 * so or not from the serialized text, then we always creates a new object owned by this. */
6255 /* Manager manages ExecRuntime objects by the unit id.
6256 * So, we omit the serialized text when the unit does not have id (yet?)... */
6257 if (isempty(u
->id
)) {
6258 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6262 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
6264 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
6268 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6270 r
= exec_runtime_allocate(&rt_create
, u
->id
);
6277 if (streq(key
, "tmp-dir")) {
6280 copy
= strdup(value
);
6284 free_and_replace(rt
->tmp_dir
, copy
);
6286 } else if (streq(key
, "var-tmp-dir")) {
6289 copy
= strdup(value
);
6293 free_and_replace(rt
->var_tmp_dir
, copy
);
6295 } else if (streq(key
, "netns-socket-0")) {
6298 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6299 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6303 safe_close(rt
->netns_storage_socket
[0]);
6304 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6306 } else if (streq(key
, "netns-socket-1")) {
6309 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6310 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6314 safe_close(rt
->netns_storage_socket
[1]);
6315 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6319 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6321 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6323 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6327 rt_create
->manager
= u
->manager
;
6330 TAKE_PTR(rt_create
);
6336 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6337 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6339 int r
, fdpair
[] = {-1, -1};
6340 const char *p
, *v
= value
;
6347 n
= strcspn(v
, " ");
6348 id
= strndupa(v
, n
);
6353 v
= startswith(p
, "tmp-dir=");
6355 n
= strcspn(v
, " ");
6356 tmp_dir
= strndup(v
, n
);
6364 v
= startswith(p
, "var-tmp-dir=");
6366 n
= strcspn(v
, " ");
6367 var_tmp_dir
= strndup(v
, n
);
6375 v
= startswith(p
, "netns-socket-0=");
6379 n
= strcspn(v
, " ");
6380 buf
= strndupa(v
, n
);
6382 r
= safe_atoi(buf
, &fdpair
[0]);
6384 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6385 if (!fdset_contains(fds
, fdpair
[0]))
6386 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6387 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", fdpair
[0]);
6388 fdpair
[0] = fdset_remove(fds
, fdpair
[0]);
6394 v
= startswith(p
, "netns-socket-1=");
6398 n
= strcspn(v
, " ");
6399 buf
= strndupa(v
, n
);
6400 r
= safe_atoi(buf
, &fdpair
[1]);
6402 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6403 if (!fdset_contains(fds
, fdpair
[1]))
6404 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6405 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", fdpair
[1]);
6406 fdpair
[1] = fdset_remove(fds
, fdpair
[1]);
6410 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, fdpair
, NULL
);
6412 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
6416 void exec_runtime_vacuum(Manager
*m
) {
6421 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
6423 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6427 (void) exec_runtime_free(rt
, false);
6431 void exec_params_clear(ExecParameters
*p
) {
6435 p
->environment
= strv_free(p
->environment
);
6436 p
->fd_names
= strv_free(p
->fd_names
);
6437 p
->fds
= mfree(p
->fds
);
6438 p
->exec_fd
= safe_close(p
->exec_fd
);
6441 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
6450 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
6452 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
6453 [EXEC_INPUT_NULL
] = "null",
6454 [EXEC_INPUT_TTY
] = "tty",
6455 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
6456 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
6457 [EXEC_INPUT_SOCKET
] = "socket",
6458 [EXEC_INPUT_NAMED_FD
] = "fd",
6459 [EXEC_INPUT_DATA
] = "data",
6460 [EXEC_INPUT_FILE
] = "file",
6463 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
6465 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
6466 [EXEC_OUTPUT_INHERIT
] = "inherit",
6467 [EXEC_OUTPUT_NULL
] = "null",
6468 [EXEC_OUTPUT_TTY
] = "tty",
6469 [EXEC_OUTPUT_KMSG
] = "kmsg",
6470 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
6471 [EXEC_OUTPUT_JOURNAL
] = "journal",
6472 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
6473 [EXEC_OUTPUT_SOCKET
] = "socket",
6474 [EXEC_OUTPUT_NAMED_FD
] = "fd",
6475 [EXEC_OUTPUT_FILE
] = "file",
6476 [EXEC_OUTPUT_FILE_APPEND
] = "append",
6477 [EXEC_OUTPUT_FILE_TRUNCATE
] = "truncate",
6480 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
6482 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
6483 [EXEC_UTMP_INIT
] = "init",
6484 [EXEC_UTMP_LOGIN
] = "login",
6485 [EXEC_UTMP_USER
] = "user",
6488 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
6490 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
6491 [EXEC_PRESERVE_NO
] = "no",
6492 [EXEC_PRESERVE_YES
] = "yes",
6493 [EXEC_PRESERVE_RESTART
] = "restart",
6496 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
6498 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
6499 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6500 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
6501 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
6502 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
6503 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
6504 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
6507 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
6509 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
6510 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
6511 * directories, specifically .timer units with their timestamp touch file. */
6512 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6513 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
6514 [EXEC_DIRECTORY_STATE
] = "state",
6515 [EXEC_DIRECTORY_CACHE
] = "cache",
6516 [EXEC_DIRECTORY_LOGS
] = "logs",
6517 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
6520 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
6522 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
6523 * the service payload in. */
6524 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6525 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
6526 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
6527 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
6528 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
6529 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
6532 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
6534 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
6535 [EXEC_KEYRING_INHERIT
] = "inherit",
6536 [EXEC_KEYRING_PRIVATE
] = "private",
6537 [EXEC_KEYRING_SHARED
] = "shared",
6540 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);