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"
42 #include "argv-util.h"
47 #include "capability-util.h"
48 #include "cgroup-setup.h"
49 #include "chase-symlinks.h"
50 #include "chown-recursive.h"
51 #include "cpu-set-util.h"
52 #include "creds-util.h"
53 #include "data-fd-util.h"
57 #include "errno-list.h"
60 #include "exit-status.h"
63 #include "format-util.h"
64 #include "glob-util.h"
65 #include "hexdecoct.h"
67 #include "ioprio-util.h"
72 #include "manager-dump.h"
73 #include "memory-util.h"
74 #include "missing_fs.h"
75 #include "missing_ioprio.h"
76 #include "mkdir-label.h"
77 #include "mount-util.h"
78 #include "mountpoint-util.h"
79 #include "namespace.h"
80 #include "parse-util.h"
81 #include "path-util.h"
82 #include "process-util.h"
83 #include "random-util.h"
84 #include "recurse-dir.h"
85 #include "rlimit-util.h"
88 #include "seccomp-util.h"
90 #include "securebits-util.h"
91 #include "selinux-util.h"
92 #include "signal-util.h"
93 #include "smack-util.h"
94 #include "socket-util.h"
95 #include "sort-util.h"
97 #include "stat-util.h"
98 #include "string-table.h"
99 #include "string-util.h"
101 #include "syslog-util.h"
102 #include "terminal-util.h"
103 #include "tmpfile-util.h"
104 #include "umask-util.h"
105 #include "unit-serialize.h"
106 #include "user-util.h"
107 #include "utmp-wtmp.h"
109 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
110 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
112 #define SNDBUF_SIZE (8*1024*1024)
114 static int shift_fds(int fds
[], size_t n_fds
) {
118 /* Modifies the fds array! (sorts it) */
122 for (int start
= 0;;) {
123 int restart_from
= -1;
125 for (int i
= start
; i
< (int) n_fds
; i
++) {
128 /* Already at right index? */
132 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
139 /* Hmm, the fd we wanted isn't free? Then
140 * let's remember that and try again from here */
141 if (nfd
!= i
+3 && restart_from
< 0)
145 if (restart_from
< 0)
148 start
= restart_from
;
154 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
158 n_fds
= n_socket_fds
+ n_storage_fds
;
164 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
165 * O_NONBLOCK only applies to socket activation though. */
167 for (size_t i
= 0; i
< n_fds
; i
++) {
169 if (i
< n_socket_fds
) {
170 r
= fd_nonblock(fds
[i
], nonblock
);
175 /* We unconditionally drop FD_CLOEXEC from the fds,
176 * since after all we want to pass these fds to our
179 r
= fd_cloexec(fds
[i
], false);
187 static const char *exec_context_tty_path(const ExecContext
*context
) {
190 if (context
->stdio_as_fds
)
193 if (context
->tty_path
)
194 return context
->tty_path
;
196 return "/dev/console";
199 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
204 path
= exec_context_tty_path(context
);
206 if (context
->tty_vhangup
) {
207 if (p
&& p
->stdin_fd
>= 0)
208 (void) terminal_vhangup_fd(p
->stdin_fd
);
210 (void) terminal_vhangup(path
);
213 if (context
->tty_reset
) {
214 if (p
&& p
->stdin_fd
>= 0)
215 (void) reset_terminal_fd(p
->stdin_fd
, true);
217 (void) reset_terminal(path
);
220 if (p
&& p
->stdin_fd
>= 0)
221 (void) terminal_set_size_fd(p
->stdin_fd
, path
, context
->tty_rows
, context
->tty_cols
);
223 if (context
->tty_vt_disallocate
&& path
)
224 (void) vt_disallocate(path
);
227 static bool is_terminal_input(ExecInput i
) {
230 EXEC_INPUT_TTY_FORCE
,
231 EXEC_INPUT_TTY_FAIL
);
234 static bool is_terminal_output(ExecOutput o
) {
237 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
238 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
241 static bool is_kmsg_output(ExecOutput o
) {
244 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
247 static bool exec_context_needs_term(const ExecContext
*c
) {
250 /* Return true if the execution context suggests we should set $TERM to something useful. */
252 if (is_terminal_input(c
->std_input
))
255 if (is_terminal_output(c
->std_output
))
258 if (is_terminal_output(c
->std_error
))
261 return !!c
->tty_path
;
264 static int open_null_as(int flags
, int nfd
) {
269 fd
= open("/dev/null", flags
|O_NOCTTY
);
273 return move_fd(fd
, nfd
, false);
276 static int connect_journal_socket(
278 const char *log_namespace
,
282 uid_t olduid
= UID_INVALID
;
283 gid_t oldgid
= GID_INVALID
;
288 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
289 "/run/systemd/journal/stdout";
291 if (gid_is_valid(gid
)) {
294 if (setegid(gid
) < 0)
298 if (uid_is_valid(uid
)) {
301 if (seteuid(uid
) < 0) {
307 r
= connect_unix_path(fd
, AT_FDCWD
, j
);
309 /* If we fail to restore the uid or gid, things will likely fail later on. This should only happen if
310 an LSM interferes. */
312 if (uid_is_valid(uid
))
313 (void) seteuid(olduid
);
316 if (gid_is_valid(gid
))
317 (void) setegid(oldgid
);
322 static int connect_logger_as(
324 const ExecContext
*context
,
325 const ExecParameters
*params
,
332 _cleanup_close_
int fd
= -1;
337 assert(output
< _EXEC_OUTPUT_MAX
);
341 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
345 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
349 if (shutdown(fd
, SHUT_RD
) < 0)
352 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
362 context
->syslog_identifier
?: ident
,
363 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
364 context
->syslog_priority
,
365 !!context
->syslog_level_prefix
,
367 is_kmsg_output(output
),
368 is_terminal_output(output
)) < 0)
371 return move_fd(TAKE_FD(fd
), nfd
, false);
374 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
380 fd
= open_terminal(path
, flags
| O_NOCTTY
);
384 return move_fd(fd
, nfd
, false);
387 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
388 _cleanup_close_
int fd
= -1;
393 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
396 fd
= open(path
, flags
|O_NOCTTY
, mode
);
400 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
403 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
405 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
409 r
= connect_unix_path(fd
, AT_FDCWD
, path
);
410 if (IN_SET(r
, -ENOTSOCK
, -EINVAL
))
411 /* Propagate initial error if we get ENOTSOCK or EINVAL, i.e. we have indication that this
412 * wasn't an AF_UNIX socket after all */
417 if ((flags
& O_ACCMODE
) == O_RDONLY
)
418 r
= shutdown(fd
, SHUT_WR
);
419 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
420 r
= shutdown(fd
, SHUT_RD
);
429 static int fixup_input(
430 const ExecContext
*context
,
432 bool apply_tty_stdin
) {
438 std_input
= context
->std_input
;
440 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
441 return EXEC_INPUT_NULL
;
443 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
444 return EXEC_INPUT_NULL
;
446 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
447 return EXEC_INPUT_NULL
;
452 static int fixup_output(ExecOutput output
, int socket_fd
) {
454 if (output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
455 return EXEC_OUTPUT_INHERIT
;
460 static int setup_input(
461 const ExecContext
*context
,
462 const ExecParameters
*params
,
464 const int named_iofds
[static 3]) {
473 if (params
->stdin_fd
>= 0) {
474 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
477 /* Try to make this the controlling tty, if it is a tty, and reset it */
478 if (isatty(STDIN_FILENO
)) {
479 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
480 (void) reset_terminal_fd(STDIN_FILENO
, true);
481 (void) terminal_set_size_fd(STDIN_FILENO
, NULL
, context
->tty_rows
, context
->tty_cols
);
487 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
491 case EXEC_INPUT_NULL
:
492 return open_null_as(O_RDONLY
, STDIN_FILENO
);
495 case EXEC_INPUT_TTY_FORCE
:
496 case EXEC_INPUT_TTY_FAIL
: {
499 fd
= acquire_terminal(exec_context_tty_path(context
),
500 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
501 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
502 ACQUIRE_TERMINAL_WAIT
,
507 r
= terminal_set_size_fd(fd
, exec_context_tty_path(context
), context
->tty_rows
, context
->tty_cols
);
511 return move_fd(fd
, STDIN_FILENO
, false);
514 case EXEC_INPUT_SOCKET
:
515 assert(socket_fd
>= 0);
517 return RET_NERRNO(dup2(socket_fd
, STDIN_FILENO
));
519 case EXEC_INPUT_NAMED_FD
:
520 assert(named_iofds
[STDIN_FILENO
] >= 0);
522 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
523 return RET_NERRNO(dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
));
525 case EXEC_INPUT_DATA
: {
528 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
532 return move_fd(fd
, STDIN_FILENO
, false);
535 case EXEC_INPUT_FILE
: {
539 assert(context
->stdio_file
[STDIN_FILENO
]);
541 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
542 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
544 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
548 return move_fd(fd
, STDIN_FILENO
, false);
552 assert_not_reached();
556 static bool can_inherit_stderr_from_stdout(
557 const ExecContext
*context
,
563 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
566 if (e
== EXEC_OUTPUT_INHERIT
)
571 if (e
== EXEC_OUTPUT_NAMED_FD
)
572 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
574 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
, EXEC_OUTPUT_FILE_TRUNCATE
))
575 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
580 static int setup_output(
582 const ExecContext
*context
,
583 const ExecParameters
*params
,
586 const int named_iofds
[static 3],
590 dev_t
*journal_stream_dev
,
591 ino_t
*journal_stream_ino
) {
601 assert(journal_stream_dev
);
602 assert(journal_stream_ino
);
604 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
606 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
609 return STDOUT_FILENO
;
612 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
613 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
616 return STDERR_FILENO
;
619 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
620 o
= fixup_output(context
->std_output
, socket_fd
);
622 if (fileno
== STDERR_FILENO
) {
624 e
= fixup_output(context
->std_error
, socket_fd
);
626 /* This expects the input and output are already set up */
628 /* Don't change the stderr file descriptor if we inherit all
629 * the way and are not on a tty */
630 if (e
== EXEC_OUTPUT_INHERIT
&&
631 o
== EXEC_OUTPUT_INHERIT
&&
632 i
== EXEC_INPUT_NULL
&&
633 !is_terminal_input(context
->std_input
) &&
637 /* Duplicate from stdout if possible */
638 if (can_inherit_stderr_from_stdout(context
, o
, e
))
639 return RET_NERRNO(dup2(STDOUT_FILENO
, fileno
));
643 } else if (o
== EXEC_OUTPUT_INHERIT
) {
644 /* If input got downgraded, inherit the original value */
645 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
646 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
648 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
649 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
650 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
652 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
656 /* We need to open /dev/null here anew, to get the right access mode. */
657 return open_null_as(O_WRONLY
, fileno
);
662 case EXEC_OUTPUT_NULL
:
663 return open_null_as(O_WRONLY
, fileno
);
665 case EXEC_OUTPUT_TTY
:
666 if (is_terminal_input(i
))
667 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
669 /* We don't reset the terminal if this is just about output */
670 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
672 case EXEC_OUTPUT_KMSG
:
673 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
674 case EXEC_OUTPUT_JOURNAL
:
675 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
676 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
678 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m",
679 fileno
== STDOUT_FILENO
? "stdout" : "stderr");
680 r
= open_null_as(O_WRONLY
, fileno
);
684 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
685 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
686 * services to detect whether they are connected to the journal or not.
688 * If both stdout and stderr are connected to a stream then let's make sure to store the data
689 * about STDERR as that's usually the best way to do logging. */
691 if (fstat(fileno
, &st
) >= 0 &&
692 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
693 *journal_stream_dev
= st
.st_dev
;
694 *journal_stream_ino
= st
.st_ino
;
699 case EXEC_OUTPUT_SOCKET
:
700 assert(socket_fd
>= 0);
702 return RET_NERRNO(dup2(socket_fd
, fileno
));
704 case EXEC_OUTPUT_NAMED_FD
:
705 assert(named_iofds
[fileno
] >= 0);
707 (void) fd_nonblock(named_iofds
[fileno
], false);
708 return RET_NERRNO(dup2(named_iofds
[fileno
], fileno
));
710 case EXEC_OUTPUT_FILE
:
711 case EXEC_OUTPUT_FILE_APPEND
:
712 case EXEC_OUTPUT_FILE_TRUNCATE
: {
716 assert(context
->stdio_file
[fileno
]);
718 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
719 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
722 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
725 if (o
== EXEC_OUTPUT_FILE_APPEND
)
727 else if (o
== EXEC_OUTPUT_FILE_TRUNCATE
)
730 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
734 return move_fd(fd
, fileno
, 0);
738 assert_not_reached();
742 static int chown_terminal(int fd
, uid_t uid
) {
747 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
748 if (isatty(fd
) < 1) {
749 if (IN_SET(errno
, EINVAL
, ENOTTY
))
750 return 0; /* not a tty */
755 /* This might fail. What matters are the results. */
756 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, GID_INVALID
);
763 static int setup_confirm_stdio(
764 const ExecContext
*context
,
766 int *ret_saved_stdin
,
767 int *ret_saved_stdout
) {
769 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
772 assert(ret_saved_stdin
);
773 assert(ret_saved_stdout
);
775 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
779 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
780 if (saved_stdout
< 0)
783 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
787 r
= chown_terminal(fd
, getuid());
791 r
= reset_terminal_fd(fd
, true);
795 r
= terminal_set_size_fd(fd
, vc
, context
->tty_rows
, context
->tty_cols
);
799 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
); /* Invalidates 'fd' also on failure */
804 *ret_saved_stdin
= TAKE_FD(saved_stdin
);
805 *ret_saved_stdout
= TAKE_FD(saved_stdout
);
809 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
812 if (err
== -ETIMEDOUT
)
813 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
816 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
820 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
821 _cleanup_close_
int fd
= -1;
825 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
829 write_confirm_error_fd(err
, fd
, u
);
832 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
836 assert(saved_stdout
);
840 if (*saved_stdin
>= 0)
841 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
844 if (*saved_stdout
>= 0)
845 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
848 *saved_stdin
= safe_close(*saved_stdin
);
849 *saved_stdout
= safe_close(*saved_stdout
);
855 CONFIRM_PRETEND_FAILURE
= -1,
856 CONFIRM_PRETEND_SUCCESS
= 0,
860 static int ask_for_confirmation(const ExecContext
*context
, const char *vc
, Unit
*u
, const char *cmdline
) {
861 int saved_stdout
= -1, saved_stdin
= -1, r
;
862 _cleanup_free_
char *e
= NULL
;
865 /* For any internal errors, assume a positive response. */
866 r
= setup_confirm_stdio(context
, vc
, &saved_stdin
, &saved_stdout
);
868 write_confirm_error(r
, vc
, u
);
869 return CONFIRM_EXECUTE
;
872 /* confirm_spawn might have been disabled while we were sleeping. */
873 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
878 e
= ellipsize(cmdline
, 60, 100);
886 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
888 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
895 printf("Resuming normal execution.\n");
896 manager_disable_confirm_spawn();
900 unit_dump(u
, stdout
, " ");
901 continue; /* ask again */
903 printf("Failing execution.\n");
904 r
= CONFIRM_PRETEND_FAILURE
;
907 printf(" c - continue, proceed without asking anymore\n"
908 " D - dump, show the state of the unit\n"
909 " f - fail, don't execute the command and pretend it failed\n"
911 " i - info, show a short summary of the unit\n"
912 " j - jobs, show jobs that are in progress\n"
913 " s - skip, don't execute the command and pretend it succeeded\n"
914 " y - yes, execute the command\n");
915 continue; /* ask again */
917 printf(" Description: %s\n"
920 u
->id
, u
->description
, cmdline
);
921 continue; /* ask again */
923 manager_dump_jobs(u
->manager
, stdout
, /* patterns= */ NULL
, " ");
924 continue; /* ask again */
926 /* 'n' was removed in favor of 'f'. */
927 printf("Didn't understand 'n', did you mean 'f'?\n");
928 continue; /* ask again */
930 printf("Skipping execution.\n");
931 r
= CONFIRM_PRETEND_SUCCESS
;
937 assert_not_reached();
943 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
947 static int get_fixed_user(const ExecContext
*c
, const char **user
,
948 uid_t
*uid
, gid_t
*gid
,
949 const char **home
, const char **shell
) {
958 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
959 * (i.e. are "/" or "/bin/nologin"). */
962 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
970 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
980 r
= get_group_creds(&name
, gid
, 0);
988 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
989 const char *group
, gid_t gid
,
990 gid_t
**supplementary_gids
, int *ngids
) {
993 bool keep_groups
= false;
994 gid_t
*groups
= NULL
;
995 _cleanup_free_ gid_t
*l_gids
= NULL
;
1000 * If user is given, then lookup GID and supplementary groups list.
1001 * We avoid NSS lookups for gid=0. Also we have to initialize groups
1002 * here and as early as possible so we keep the list of supplementary
1003 * groups of the caller.
1005 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
1006 /* First step, initialize groups from /etc/groups */
1007 if (initgroups(user
, gid
) < 0)
1013 if (strv_isempty(c
->supplementary_groups
))
1017 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1018 * be positive, otherwise fail.
1021 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1022 if (ngroups_max
<= 0)
1023 return errno_or_else(EOPNOTSUPP
);
1025 l_gids
= new(gid_t
, ngroups_max
);
1031 * Lookup the list of groups that the user belongs to, we
1032 * avoid NSS lookups here too for gid=0.
1035 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1040 STRV_FOREACH(i
, c
->supplementary_groups
) {
1043 if (k
>= ngroups_max
)
1047 r
= get_group_creds(&g
, l_gids
+k
, 0);
1055 * Sets ngids to zero to drop all supplementary groups, happens
1056 * when we are under root and SupplementaryGroups= is empty.
1063 /* Otherwise get the final list of supplementary groups */
1064 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1068 *supplementary_gids
= groups
;
1076 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1079 /* Handle SupplementaryGroups= if it is not empty */
1081 r
= maybe_setgroups(ngids
, supplementary_gids
);
1086 if (gid_is_valid(gid
)) {
1087 /* Then set our gids */
1088 if (setresgid(gid
, gid
, gid
) < 0)
1095 static int set_securebits(int bits
, int mask
) {
1096 int current
, applied
;
1097 current
= prctl(PR_GET_SECUREBITS
);
1100 /* Clear all securebits defined in mask and set bits */
1101 applied
= (current
& ~mask
) | bits
;
1102 if (current
== applied
)
1104 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1109 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1113 if (!uid_is_valid(uid
))
1116 /* Sets (but doesn't look up) the uid and make sure we keep the
1117 * capabilities while doing so. For setting secure bits the capability CAP_SETPCAP is
1118 * required, so we also need keep-caps in this case.
1121 if (context
->capability_ambient_set
!= 0 || context
->secure_bits
!= 0) {
1123 /* First step: If we need to keep capabilities but
1124 * drop privileges we need to make sure we keep our
1125 * caps, while we drop privileges. */
1127 /* Add KEEP_CAPS to the securebits */
1128 r
= set_securebits(1<<SECURE_KEEP_CAPS
, 0);
1134 /* Second step: actually set the uids */
1135 if (setresuid(uid
, uid
, uid
) < 0)
1138 /* At this point we should have all necessary capabilities but
1139 are otherwise a normal user. However, the caps might got
1140 corrupted due to the setresuid() so we need clean them up
1141 later. This is done outside of this call. */
1148 static int null_conv(
1150 const struct pam_message
**msg
,
1151 struct pam_response
**resp
,
1152 void *appdata_ptr
) {
1154 /* We don't support conversations */
1156 return PAM_CONV_ERR
;
1161 static int setup_pam(
1167 char ***env
, /* updated on success */
1168 const int fds
[], size_t n_fds
) {
1172 static const struct pam_conv conv
= {
1177 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1178 _cleanup_strv_free_
char **e
= NULL
;
1179 pam_handle_t
*handle
= NULL
;
1181 int pam_code
= PAM_SUCCESS
, r
;
1182 bool close_session
= false;
1183 pid_t pam_pid
= 0, parent_pid
;
1190 /* We set up PAM in the parent process, then fork. The child
1191 * will then stay around until killed via PR_GET_PDEATHSIG or
1192 * systemd via the cgroup logic. It will then remove the PAM
1193 * session again. The parent process will exec() the actual
1194 * daemon. We do things this way to ensure that the main PID
1195 * of the daemon is the one we initially fork()ed. */
1197 r
= barrier_create(&barrier
);
1201 if (log_get_max_level() < LOG_DEBUG
)
1202 flags
|= PAM_SILENT
;
1204 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1205 if (pam_code
!= PAM_SUCCESS
) {
1211 _cleanup_free_
char *q
= NULL
;
1213 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1214 * out if that's the case, and read the TTY off it. */
1216 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1217 tty
= strjoina("/dev/", q
);
1221 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1222 if (pam_code
!= PAM_SUCCESS
)
1226 STRV_FOREACH(nv
, *env
) {
1227 pam_code
= pam_putenv(handle
, *nv
);
1228 if (pam_code
!= PAM_SUCCESS
)
1232 pam_code
= pam_acct_mgmt(handle
, flags
);
1233 if (pam_code
!= PAM_SUCCESS
)
1236 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1237 if (pam_code
!= PAM_SUCCESS
)
1238 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1240 pam_code
= pam_open_session(handle
, flags
);
1241 if (pam_code
!= PAM_SUCCESS
)
1244 close_session
= true;
1246 e
= pam_getenvlist(handle
);
1248 pam_code
= PAM_BUF_ERR
;
1252 /* Block SIGTERM, so that we know that it won't get lost in the child */
1254 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1256 parent_pid
= getpid_cached();
1258 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1262 int sig
, ret
= EXIT_PAM
;
1264 /* The child's job is to reset the PAM session on termination */
1265 barrier_set_role(&barrier
, BARRIER_CHILD
);
1267 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1268 * those fds are open here that have been opened by PAM. */
1269 (void) close_many(fds
, n_fds
);
1271 /* Drop privileges - we don't need any to pam_close_session and this will make
1272 * PR_SET_PDEATHSIG work in most cases. If this fails, ignore the error - but expect sd-pam
1273 * threads to fail to exit normally */
1275 r
= maybe_setgroups(0, NULL
);
1277 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1278 if (setresgid(gid
, gid
, gid
) < 0)
1279 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1280 if (setresuid(uid
, uid
, uid
) < 0)
1281 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1283 (void) ignore_signals(SIGPIPE
);
1285 /* Wait until our parent died. This will only work if the above setresuid() succeeds,
1286 * otherwise the kernel will not allow unprivileged parents kill their privileged children
1287 * this way. We rely on the control groups kill logic to do the rest for us. */
1288 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1291 /* Tell the parent that our setup is done. This is especially important regarding dropping
1292 * privileges. Otherwise, unit setup might race against our setresuid(2) call.
1294 * If the parent aborted, we'll detect this below, hence ignore return failure here. */
1295 (void) barrier_place(&barrier
);
1297 /* Check if our parent process might already have died? */
1298 if (getppid() == parent_pid
) {
1301 assert_se(sigemptyset(&ss
) >= 0);
1302 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1305 if (sigwait(&ss
, &sig
) < 0) {
1312 assert(sig
== SIGTERM
);
1317 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1318 if (pam_code
!= PAM_SUCCESS
)
1321 /* If our parent died we'll end the session */
1322 if (getppid() != parent_pid
) {
1323 pam_code
= pam_close_session(handle
, flags
);
1324 if (pam_code
!= PAM_SUCCESS
)
1331 /* NB: pam_end() when called in child processes should set PAM_DATA_SILENT to let the module
1332 * know about this. See pam_end(3) */
1333 (void) pam_end(handle
, pam_code
| flags
| PAM_DATA_SILENT
);
1337 barrier_set_role(&barrier
, BARRIER_PARENT
);
1339 /* If the child was forked off successfully it will do all the cleanups, so forget about the handle
1343 /* Unblock SIGTERM again in the parent */
1344 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1346 /* We close the log explicitly here, since the PAM modules might have opened it, but we don't want
1347 * this fd around. */
1350 /* Synchronously wait for the child to initialize. We don't care for errors as we cannot
1351 * recover. However, warn loudly if it happens. */
1352 if (!barrier_place_and_sync(&barrier
))
1353 log_error("PAM initialization failed");
1355 return strv_free_and_replace(*env
, e
);
1358 if (pam_code
!= PAM_SUCCESS
) {
1359 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1360 r
= -EPERM
; /* PAM errors do not map to errno */
1362 log_error_errno(r
, "PAM failed: %m");
1366 pam_code
= pam_close_session(handle
, flags
);
1368 (void) pam_end(handle
, pam_code
| flags
);
1378 static void rename_process_from_path(const char *path
) {
1379 char process_name
[11];
1383 /* This resulting string must fit in 10 chars (i.e. the length
1384 * of "/sbin/init") to look pretty in /bin/ps */
1388 rename_process("(...)");
1394 /* The end of the process name is usually more
1395 * interesting, since the first bit might just be
1401 process_name
[0] = '(';
1402 memcpy(process_name
+1, p
, l
);
1403 process_name
[1+l
] = ')';
1404 process_name
[1+l
+1] = 0;
1406 rename_process(process_name
);
1409 static bool context_has_address_families(const ExecContext
*c
) {
1412 return c
->address_families_allow_list
||
1413 !set_isempty(c
->address_families
);
1416 static bool context_has_syscall_filters(const ExecContext
*c
) {
1419 return c
->syscall_allow_list
||
1420 !hashmap_isempty(c
->syscall_filter
);
1423 static bool context_has_syscall_logs(const ExecContext
*c
) {
1426 return c
->syscall_log_allow_list
||
1427 !hashmap_isempty(c
->syscall_log
);
1430 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1433 if (c
->no_new_privileges
)
1436 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1439 /* We need NNP if we have any form of seccomp and are unprivileged */
1440 return c
->lock_personality
||
1441 c
->memory_deny_write_execute
||
1442 c
->private_devices
||
1444 c
->protect_hostname
||
1445 c
->protect_kernel_tunables
||
1446 c
->protect_kernel_modules
||
1447 c
->protect_kernel_logs
||
1448 context_has_address_families(c
) ||
1449 exec_context_restrict_namespaces_set(c
) ||
1450 c
->restrict_realtime
||
1451 c
->restrict_suid_sgid
||
1452 !set_isempty(c
->syscall_archs
) ||
1453 context_has_syscall_filters(c
) ||
1454 context_has_syscall_logs(c
);
1457 static bool exec_context_has_credentials(const ExecContext
*context
) {
1461 return !hashmap_isempty(context
->set_credentials
) ||
1462 !hashmap_isempty(context
->load_credentials
);
1467 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1469 if (is_seccomp_available())
1472 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1476 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1477 uint32_t negative_action
, default_action
, action
;
1483 if (!context_has_syscall_filters(c
))
1486 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1489 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1491 if (c
->syscall_allow_list
) {
1492 default_action
= negative_action
;
1493 action
= SCMP_ACT_ALLOW
;
1495 default_action
= SCMP_ACT_ALLOW
;
1496 action
= negative_action
;
1499 if (needs_ambient_hack
) {
1500 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1505 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1508 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1510 uint32_t default_action
, action
;
1516 if (!context_has_syscall_logs(c
))
1520 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1523 if (c
->syscall_log_allow_list
) {
1524 /* Log nothing but the ones listed */
1525 default_action
= SCMP_ACT_ALLOW
;
1526 action
= SCMP_ACT_LOG
;
1528 /* Log everything but the ones listed */
1529 default_action
= SCMP_ACT_LOG
;
1530 action
= SCMP_ACT_ALLOW
;
1533 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1535 /* old libseccomp */
1536 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1541 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1545 if (set_isempty(c
->syscall_archs
))
1548 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1551 return seccomp_restrict_archs(c
->syscall_archs
);
1554 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1558 if (!context_has_address_families(c
))
1561 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1564 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1567 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1571 if (!c
->memory_deny_write_execute
)
1574 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1577 return seccomp_memory_deny_write_execute();
1580 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1584 if (!c
->restrict_realtime
)
1587 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1590 return seccomp_restrict_realtime();
1593 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1597 if (!c
->restrict_suid_sgid
)
1600 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1603 return seccomp_restrict_suid_sgid();
1606 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1610 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1611 * let's protect even those systems where this is left on in the kernel. */
1613 if (!c
->protect_kernel_tunables
)
1616 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1619 return seccomp_protect_sysctl();
1622 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1626 /* Turn off module syscalls on ProtectKernelModules=yes */
1628 if (!c
->protect_kernel_modules
)
1631 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1634 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1637 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1641 if (!c
->protect_kernel_logs
)
1644 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1647 return seccomp_protect_syslog();
1650 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1654 if (!c
->protect_clock
)
1657 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1660 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1663 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1667 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1669 if (!c
->private_devices
)
1672 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1675 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1678 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1682 if (!exec_context_restrict_namespaces_set(c
))
1685 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1688 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1691 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1692 unsigned long personality
;
1698 if (!c
->lock_personality
)
1701 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1704 personality
= c
->personality
;
1706 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1707 if (personality
== PERSONALITY_INVALID
) {
1709 r
= opinionated_personality(&personality
);
1714 return seccomp_lock_personality(personality
);
1720 static int apply_restrict_filesystems(Unit
*u
, const ExecContext
*c
) {
1724 if (!exec_context_restrict_filesystems_set(c
))
1727 if (!u
->manager
->restrict_fs
) {
1728 /* LSM BPF is unsupported or lsm_bpf_setup failed */
1729 log_unit_debug(u
, "LSM BPF not supported, skipping RestrictFileSystems=");
1733 return lsm_bpf_unit_restrict_filesystems(u
, c
->restrict_filesystems
, c
->restrict_filesystems_allow_list
);
1737 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1741 if (!c
->protect_hostname
)
1744 if (ns_type_supported(NAMESPACE_UTS
)) {
1745 if (unshare(CLONE_NEWUTS
) < 0) {
1746 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1747 *ret_exit_status
= EXIT_NAMESPACE
;
1748 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1751 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1754 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1759 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1762 r
= seccomp_protect_hostname();
1764 *ret_exit_status
= EXIT_SECCOMP
;
1765 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1772 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1775 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1776 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1778 if (idle_pipe
[0] >= 0) {
1781 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1783 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1786 /* Signal systemd that we are bored and want to continue. */
1787 n
= write(idle_pipe
[3], "x", 1);
1789 /* Wait for systemd to react to the signal above. */
1790 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1793 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1797 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1800 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1802 static int build_environment(
1804 const ExecContext
*c
,
1805 const ExecParameters
*p
,
1808 const char *username
,
1810 dev_t journal_stream_dev
,
1811 ino_t journal_stream_ino
,
1814 _cleanup_strv_free_
char **our_env
= NULL
;
1823 #define N_ENV_VARS 17
1824 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1829 _cleanup_free_
char *joined
= NULL
;
1831 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1833 our_env
[n_env
++] = x
;
1835 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1837 our_env
[n_env
++] = x
;
1839 joined
= strv_join(p
->fd_names
, ":");
1843 x
= strjoin("LISTEN_FDNAMES=", joined
);
1846 our_env
[n_env
++] = x
;
1849 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1850 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1852 our_env
[n_env
++] = x
;
1854 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1856 our_env
[n_env
++] = x
;
1859 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use blocking
1860 * Varlink calls back to us for look up dynamic users in PID 1. Break the deadlock between D-Bus and
1861 * PID 1 by disabling use of PID1' NSS interface for looking up dynamic users. */
1862 if (p
->flags
& EXEC_NSS_DYNAMIC_BYPASS
) {
1863 x
= strdup("SYSTEMD_NSS_DYNAMIC_BYPASS=1");
1866 our_env
[n_env
++] = x
;
1870 x
= strjoin("HOME=", home
);
1874 path_simplify(x
+ 5);
1875 our_env
[n_env
++] = x
;
1879 x
= strjoin("LOGNAME=", username
);
1882 our_env
[n_env
++] = x
;
1884 x
= strjoin("USER=", username
);
1887 our_env
[n_env
++] = x
;
1891 x
= strjoin("SHELL=", shell
);
1895 path_simplify(x
+ 6);
1896 our_env
[n_env
++] = x
;
1899 if (!sd_id128_is_null(u
->invocation_id
)) {
1900 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1903 our_env
[n_env
++] = x
;
1906 if (exec_context_needs_term(c
)) {
1907 const char *tty_path
, *term
= NULL
;
1909 tty_path
= exec_context_tty_path(c
);
1911 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1912 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1913 * container manager passes to PID 1 ends up all the way in the console login shown. */
1915 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1916 term
= getenv("TERM");
1919 term
= default_term_for_tty(tty_path
);
1921 x
= strjoin("TERM=", term
);
1924 our_env
[n_env
++] = x
;
1927 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1928 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1931 our_env
[n_env
++] = x
;
1934 if (c
->log_namespace
) {
1935 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1939 our_env
[n_env
++] = x
;
1942 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1943 _cleanup_free_
char *joined
= NULL
;
1949 if (c
->directories
[t
].n_items
== 0)
1952 n
= exec_directory_env_name_to_string(t
);
1956 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
1957 _cleanup_free_
char *prefixed
= NULL
;
1959 prefixed
= path_join(p
->prefix
[t
], c
->directories
[t
].items
[i
].path
);
1963 if (!strextend_with_separator(&joined
, ":", prefixed
))
1967 x
= strjoin(n
, "=", joined
);
1971 our_env
[n_env
++] = x
;
1974 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
1975 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
1979 our_env
[n_env
++] = x
;
1982 if (asprintf(&x
, "SYSTEMD_EXEC_PID=" PID_FMT
, getpid_cached()) < 0)
1985 our_env
[n_env
++] = x
;
1987 our_env
[n_env
++] = NULL
;
1988 assert(n_env
<= N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1991 *ret
= TAKE_PTR(our_env
);
1996 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1997 _cleanup_strv_free_
char **pass_env
= NULL
;
2000 STRV_FOREACH(i
, c
->pass_environment
) {
2001 _cleanup_free_
char *x
= NULL
;
2007 x
= strjoin(*i
, "=", v
);
2011 if (!GREEDY_REALLOC(pass_env
, n_env
+ 2))
2014 pass_env
[n_env
++] = TAKE_PTR(x
);
2015 pass_env
[n_env
] = NULL
;
2018 *ret
= TAKE_PTR(pass_env
);
2023 bool exec_needs_mount_namespace(
2024 const ExecContext
*context
,
2025 const ExecParameters
*params
,
2026 const ExecRuntime
*runtime
) {
2030 if (context
->root_image
)
2033 if (!strv_isempty(context
->read_write_paths
) ||
2034 !strv_isempty(context
->read_only_paths
) ||
2035 !strv_isempty(context
->inaccessible_paths
) ||
2036 !strv_isempty(context
->exec_paths
) ||
2037 !strv_isempty(context
->no_exec_paths
))
2040 if (context
->n_bind_mounts
> 0)
2043 if (context
->n_temporary_filesystems
> 0)
2046 if (context
->n_mount_images
> 0)
2049 if (context
->n_extension_images
> 0)
2052 if (!strv_isempty(context
->extension_directories
))
2055 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
2058 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
2061 if (context
->private_devices
||
2062 context
->private_mounts
||
2063 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2064 context
->protect_home
!= PROTECT_HOME_NO
||
2065 context
->protect_kernel_tunables
||
2066 context
->protect_kernel_modules
||
2067 context
->protect_kernel_logs
||
2068 context
->protect_control_groups
||
2069 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2070 context
->proc_subset
!= PROC_SUBSET_ALL
||
2071 context
->private_ipc
||
2072 context
->ipc_namespace_path
)
2075 if (context
->root_directory
) {
2076 if (exec_context_get_effective_mount_apivfs(context
))
2079 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2080 if (params
&& !params
->prefix
[t
])
2083 if (context
->directories
[t
].n_items
> 0)
2088 if (context
->dynamic_user
&&
2089 (context
->directories
[EXEC_DIRECTORY_STATE
].n_items
> 0 ||
2090 context
->directories
[EXEC_DIRECTORY_CACHE
].n_items
> 0 ||
2091 context
->directories
[EXEC_DIRECTORY_LOGS
].n_items
> 0))
2094 if (context
->log_namespace
)
2100 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2101 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2102 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
2103 _cleanup_close_
int unshare_ready_fd
= -1;
2104 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2109 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2110 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2111 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2112 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2113 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2114 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2115 * continues execution normally.
2116 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2117 * does not need CAP_SETUID to write the single line mapping to itself. */
2119 /* Can only set up multiple mappings with CAP_SETUID. */
2120 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
2121 r
= asprintf(&uid_map
,
2122 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2123 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2124 ouid
, ouid
, uid
, uid
);
2126 r
= asprintf(&uid_map
,
2127 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2133 /* Can only set up multiple mappings with CAP_SETGID. */
2134 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2135 r
= asprintf(&gid_map
,
2136 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2137 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2138 ogid
, ogid
, gid
, gid
);
2140 r
= asprintf(&gid_map
,
2141 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2147 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2149 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2150 if (unshare_ready_fd
< 0)
2153 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2155 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2158 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2162 _cleanup_close_
int fd
= -1;
2166 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2167 * here, after the parent opened its own user namespace. */
2170 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2172 /* Wait until the parent unshared the user namespace */
2173 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2178 /* Disable the setgroups() system call in the child user namespace, for good. */
2179 a
= procfs_file_alloca(ppid
, "setgroups");
2180 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2182 if (errno
!= ENOENT
) {
2187 /* If the file is missing the kernel is too old, let's continue anyway. */
2189 if (write(fd
, "deny\n", 5) < 0) {
2194 fd
= safe_close(fd
);
2197 /* First write the GID map */
2198 a
= procfs_file_alloca(ppid
, "gid_map");
2199 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2204 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2208 fd
= safe_close(fd
);
2210 /* The write the UID map */
2211 a
= procfs_file_alloca(ppid
, "uid_map");
2212 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2217 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2222 _exit(EXIT_SUCCESS
);
2225 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2226 _exit(EXIT_FAILURE
);
2229 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2231 if (unshare(CLONE_NEWUSER
) < 0)
2234 /* Let the child know that the namespace is ready now */
2235 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2238 /* Try to read an error code from the child */
2239 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2242 if (n
== sizeof(r
)) { /* an error code was sent to us */
2247 if (n
!= 0) /* on success we should have read 0 bytes */
2250 r
= wait_for_terminate_and_check("(sd-userns)", TAKE_PID(pid
), 0);
2253 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2259 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2260 if (!context
->dynamic_user
)
2263 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2266 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2272 static int create_many_symlinks(const char *root
, const char *source
, char **symlinks
) {
2273 _cleanup_free_
char *src_abs
= NULL
;
2278 src_abs
= path_join(root
, source
);
2282 STRV_FOREACH(dst
, symlinks
) {
2283 _cleanup_free_
char *dst_abs
= NULL
;
2285 dst_abs
= path_join(root
, *dst
);
2289 r
= mkdir_parents_label(dst_abs
, 0755);
2293 r
= symlink_idempotent(src_abs
, dst_abs
, true);
2301 static int setup_exec_directory(
2302 const ExecContext
*context
,
2303 const ExecParameters
*params
,
2306 ExecDirectoryType type
,
2307 bool needs_mount_namespace
,
2310 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2311 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2312 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2313 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2314 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2315 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2321 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2322 assert(exit_status
);
2324 if (!params
->prefix
[type
])
2327 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2328 if (!uid_is_valid(uid
))
2330 if (!gid_is_valid(gid
))
2334 for (size_t i
= 0; i
< context
->directories
[type
].n_items
; i
++) {
2335 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2337 p
= path_join(params
->prefix
[type
], context
->directories
[type
].items
[i
].path
);
2343 r
= mkdir_parents_label(p
, 0755);
2347 if (exec_directory_is_private(context
, type
)) {
2348 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2349 * case we want to avoid leaving a directory around fully accessible that is owned by
2350 * a dynamic user whose UID is later on reused. To lock this down we use the same
2351 * trick used by container managers to prohibit host users to get access to files of
2352 * the same UID in containers: we place everything inside a directory that has an
2353 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2354 * for unprivileged host code. We then use fs namespacing to make this directory
2355 * permeable for the service itself.
2357 * Specifically: for a service which wants a special directory "foo/" we first create
2358 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2359 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2360 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2361 * unprivileged host users can't look into it. Inside of the namespace of the unit
2362 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2363 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2364 * for the service and making sure it only gets access to the dirs it needs but no
2365 * others. Tricky? Yes, absolutely, but it works!
2367 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2368 * to be owned by the service itself.
2370 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2371 * for sharing files or sockets with other services. */
2373 pp
= path_join(params
->prefix
[type
], "private");
2379 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2380 r
= mkdir_safe_label(pp
, 0700, 0, 0, MKDIR_WARN_MODE
);
2384 if (!path_extend(&pp
, context
->directories
[type
].items
[i
].path
)) {
2389 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2390 r
= mkdir_parents_label(pp
, 0755);
2394 if (is_dir(p
, false) > 0 &&
2395 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2397 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2398 * it over. Most likely the service has been upgraded from one that didn't use
2399 * DynamicUser=1, to one that does. */
2401 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2402 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2403 exec_directory_type_to_string(type
), p
, pp
);
2405 if (rename(p
, pp
) < 0) {
2410 /* Otherwise, create the actual directory for the service */
2412 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2413 if (r
< 0 && r
!= -EEXIST
)
2417 if (!context
->directories
[type
].items
[i
].only_create
) {
2418 /* And link it up from the original place.
2420 * 1) If a mount namespace is going to be used, then this symlink remains on
2421 * the host, and a new one for the child namespace will be created later.
2422 * 2) It is not necessary to create this symlink when one of its parent
2423 * directories is specified and already created. E.g.
2424 * StateDirectory=foo foo/bar
2425 * In that case, the inode points to pp and p for "foo/bar" are the same:
2426 * pp = "/var/lib/private/foo/bar"
2427 * p = "/var/lib/foo/bar"
2428 * and, /var/lib/foo is a symlink to /var/lib/private/foo. So, not only
2429 * we do not need to create the symlink, but we cannot create the symlink.
2430 * See issue #24783. */
2431 r
= symlink_idempotent(pp
, p
, true);
2437 _cleanup_free_
char *target
= NULL
;
2439 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2440 readlink_and_make_absolute(p
, &target
) >= 0) {
2441 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2443 /* This already exists and is a symlink? Interesting. Maybe it's one created
2444 * by DynamicUser=1 (see above)?
2446 * We do this for all directory types except for ConfigurationDirectory=,
2447 * since they all support the private/ symlink logic at least in some
2448 * configurations, see above. */
2450 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2454 q
= path_join(params
->prefix
[type
], "private", context
->directories
[type
].items
[i
].path
);
2460 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2461 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2465 if (path_equal(q_resolved
, target_resolved
)) {
2467 /* Hmm, apparently DynamicUser= was once turned on for this service,
2468 * but is no longer. Let's move the directory back up. */
2470 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2471 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2472 exec_directory_type_to_string(type
), q
, p
);
2474 if (unlink(p
) < 0) {
2479 if (rename(q
, p
) < 0) {
2486 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2491 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2494 /* Don't change the owner/access mode of the configuration directory,
2495 * as in the common case it is not written to by a service, and shall
2496 * not be writable. */
2498 if (stat(p
, &st
) < 0) {
2503 /* Still complain if the access mode doesn't match */
2504 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2505 log_warning("%s \'%s\' already exists but the mode is different. "
2506 "(File system: %o %sMode: %o)",
2507 exec_directory_type_to_string(type
), context
->directories
[type
].items
[i
].path
,
2508 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2515 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2516 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2517 * current UID/GID ownership.) */
2518 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2522 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2523 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2524 * assignments to exist. */
2525 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2530 /* If we are not going to run in a namespace, set up the symlinks - otherwise
2531 * they are set up later, to allow configuring empty var/run/etc. */
2532 if (!needs_mount_namespace
)
2533 for (size_t i
= 0; i
< context
->directories
[type
].n_items
; i
++) {
2534 r
= create_many_symlinks(params
->prefix
[type
],
2535 context
->directories
[type
].items
[i
].path
,
2536 context
->directories
[type
].items
[i
].symlinks
);
2544 *exit_status
= exit_status_table
[type
];
2548 static int write_credential(
2554 bool ownership_ok
) {
2556 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2557 _cleanup_close_
int fd
= -1;
2560 r
= tempfn_random_child("", "cred", &tmp
);
2564 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2570 r
= loop_write(fd
, data
, size
, /* do_poll = */ false);
2574 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2577 if (uid_is_valid(uid
) && uid
!= getuid()) {
2578 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2580 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2583 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2584 * to express: that the user gets read access and nothing
2585 * else. But if the backing fs can't support that (e.g. ramfs)
2586 * then we can use file ownership instead. But that's only safe if
2587 * we can then re-mount the whole thing read-only, so that the
2588 * user can no longer chmod() the file to gain write access. */
2591 if (fchown(fd
, uid
, GID_INVALID
) < 0)
2596 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2603 static char **credential_search_path(
2604 const ExecParameters
*params
,
2607 _cleanup_strv_free_
char **l
= NULL
;
2611 /* Assemble a search path to find credentials in. We'll look in /etc/credstore/ (and similar
2612 * directories in /usr/lib/ + /run/) for all types of credentials. If we are looking for encrypted
2613 * credentials, also look in /etc/credstore.encrypted/ (and similar dirs). */
2616 if (strv_extend(&l
, params
->received_encrypted_credentials_directory
) < 0)
2619 if (strv_extend_strv(&l
, CONF_PATHS_STRV("credstore.encrypted"), /* filter_duplicates= */ true) < 0)
2623 if (params
->received_credentials_directory
)
2624 if (strv_extend(&l
, params
->received_credentials_directory
) < 0)
2627 if (strv_extend_strv(&l
, CONF_PATHS_STRV("credstore"), /* filter_duplicates= */ true) < 0)
2630 if (DEBUG_LOGGING
) {
2631 _cleanup_free_
char *t
= strv_join(l
, ":");
2633 log_debug("Credential search path is: %s", t
);
2639 static int load_credential(
2640 const ExecContext
*context
,
2641 const ExecParameters
*params
,
2652 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
|READ_FULL_FILE_FAIL_WHEN_LARGER
;
2653 _cleanup_strv_free_
char **search_path
= NULL
;
2654 _cleanup_(erase_and_freep
) char *data
= NULL
;
2655 _cleanup_free_
char *bindname
= NULL
;
2656 const char *source
= NULL
;
2657 bool missing_ok
= true;
2658 size_t size
, add
, maxsz
;
2666 assert(write_dfd
>= 0);
2669 if (read_dfd
>= 0) {
2670 /* If a directory fd is specified, then read the file directly from that dir. In this case we
2671 * won't do AF_UNIX stuff (we simply don't want to recursively iterate down a tree of AF_UNIX
2672 * IPC sockets). It's OK if a file vanishes here in the time we enumerate it and intend to
2675 if (!filename_is_valid(path
)) /* safety check */
2681 } else if (path_is_absolute(path
)) {
2682 /* If this is an absolute path, read the data directly from it, and support AF_UNIX
2685 if (!path_is_valid(path
)) /* safety check */
2688 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2690 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2691 * via the source socket address in case we read off an AF_UNIX socket. */
2692 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, id
) < 0)
2698 } else if (credential_name_valid(path
)) {
2699 /* If this is a relative path, take it as credential name relative to the credentials
2700 * directory we received ourselves. We don't support the AF_UNIX stuff in this mode, since we
2701 * are operating on a credential store, i.e. this is guaranteed to be regular files. */
2703 search_path
= credential_search_path(params
, encrypted
);
2712 flags
|= READ_FULL_FILE_UNBASE64
;
2714 maxsz
= encrypted
? CREDENTIAL_ENCRYPTED_SIZE_MAX
: CREDENTIAL_SIZE_MAX
;
2717 STRV_FOREACH(d
, search_path
) {
2718 _cleanup_free_
char *j
= NULL
;
2720 j
= path_join(*d
, path
);
2724 r
= read_full_file_full(
2725 AT_FDCWD
, j
, /* path is absolute, hence pass AT_FDCWD as nop dir fd here */
2735 r
= read_full_file_full(
2745 if (r
== -ENOENT
&& (missing_ok
|| hashmap_contains(context
->set_credentials
, id
))) {
2746 /* Make a missing inherited credential non-fatal, let's just continue. After all apps
2747 * will get clear errors if we don't pass such a missing credential on as they
2748 * themselves will get ENOENT when trying to read them, which should not be much
2749 * worse than when we handle the error here and make it fatal.
2751 * Also, if the source file doesn't exist, but a fallback is set via SetCredentials=
2752 * we are fine, too. */
2753 log_debug_errno(r
, "Couldn't read inherited credential '%s', skipping: %m", path
);
2757 return log_debug_errno(r
, "Failed to read credential '%s': %m", path
);
2760 _cleanup_free_
void *plaintext
= NULL
;
2761 size_t plaintext_size
= 0;
2763 r
= decrypt_credential_and_warn(id
, now(CLOCK_REALTIME
), NULL
, NULL
, data
, size
, &plaintext
, &plaintext_size
);
2767 free_and_replace(data
, plaintext
);
2768 size
= plaintext_size
;
2771 add
= strlen(id
) + size
;
2775 r
= write_credential(write_dfd
, id
, data
, size
, uid
, ownership_ok
);
2777 return log_debug_errno(r
, "Failed to write credential '%s': %m", id
);
2783 struct load_cred_args
{
2784 const ExecContext
*context
;
2785 const ExecParameters
*params
;
2794 static int load_cred_recurse_dir_cb(
2795 RecurseDirEvent event
,
2799 const struct dirent
*de
,
2800 const struct statx
*sx
,
2803 struct load_cred_args
*args
= ASSERT_PTR(userdata
);
2804 _cleanup_free_
char *sub_id
= NULL
;
2807 if (event
!= RECURSE_DIR_ENTRY
)
2808 return RECURSE_DIR_CONTINUE
;
2810 if (!IN_SET(de
->d_type
, DT_REG
, DT_SOCK
))
2811 return RECURSE_DIR_CONTINUE
;
2813 sub_id
= strreplace(path
, "/", "_");
2817 if (!credential_name_valid(sub_id
))
2818 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential would get ID %s, which is not valid, refusing", sub_id
);
2820 if (faccessat(args
->dfd
, sub_id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0) {
2821 log_debug("Skipping credential with duplicated ID %s at %s", sub_id
, path
);
2822 return RECURSE_DIR_CONTINUE
;
2824 if (errno
!= ENOENT
)
2825 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sub_id
);
2827 r
= load_credential(
2842 return RECURSE_DIR_CONTINUE
;
2845 static int acquire_credentials(
2846 const ExecContext
*context
,
2847 const ExecParameters
*params
,
2851 bool ownership_ok
) {
2853 uint64_t left
= CREDENTIALS_TOTAL_SIZE_MAX
;
2854 _cleanup_close_
int dfd
= -1;
2855 ExecLoadCredential
*lc
;
2856 ExecSetCredential
*sc
;
2862 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
2866 /* First, load credentials off disk (or acquire via AF_UNIX socket) */
2867 HASHMAP_FOREACH(lc
, context
->load_credentials
) {
2868 _cleanup_close_
int sub_fd
= -1;
2870 /* If this is an absolute path, then try to open it as a directory. If that works, then we'll
2871 * recurse into it. If it is an absolute path but it isn't a directory, then we'll open it as
2872 * a regular file. Finally, if it's a relative path we will use it as a credential name to
2873 * propagate a credential passed to us from further up. */
2875 if (path_is_absolute(lc
->path
)) {
2876 sub_fd
= open(lc
->path
, O_DIRECTORY
|O_CLOEXEC
|O_RDONLY
);
2877 if (sub_fd
< 0 && !IN_SET(errno
,
2878 ENOTDIR
, /* Not a directory */
2879 ENOENT
)) /* Doesn't exist? */
2880 return log_debug_errno(errno
, "Failed to open '%s': %m", lc
->path
);
2884 /* Regular file (incl. a credential passed in from higher up) */
2885 r
= load_credential(
2901 /* path= */ lc
->id
, /* recurse_dir() will suffix the subdir paths from here to the top-level id */
2902 /* statx_mask= */ 0,
2903 /* n_depth_max= */ UINT_MAX
,
2904 RECURSE_DIR_SORT
|RECURSE_DIR_IGNORE_DOT
|RECURSE_DIR_ENSURE_TYPE
,
2905 load_cred_recurse_dir_cb
,
2906 &(struct load_cred_args
) {
2909 .encrypted
= lc
->encrypted
,
2913 .ownership_ok
= ownership_ok
,
2920 /* Second, we add in literally specified credentials. If the credentials already exist, we'll not add
2921 * them, so that they can act as a "default" if the same credential is specified multiple times. */
2922 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
2923 _cleanup_(erase_and_freep
) void *plaintext
= NULL
;
2927 /* Note that we check ahead of time here instead of relying on O_EXCL|O_CREAT later to return
2928 * EEXIST if the credential already exists. That's because the TPM2-based decryption is kinda
2929 * slow and involved, hence it's nice to be able to skip that if the credential already
2931 if (faccessat(dfd
, sc
->id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0)
2933 if (errno
!= ENOENT
)
2934 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sc
->id
);
2936 if (sc
->encrypted
) {
2937 r
= decrypt_credential_and_warn(sc
->id
, now(CLOCK_REALTIME
), NULL
, NULL
, sc
->data
, sc
->size
, &plaintext
, &size
);
2947 add
= strlen(sc
->id
) + size
;
2951 r
= write_credential(dfd
, sc
->id
, data
, size
, uid
, ownership_ok
);
2958 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
2961 /* After we created all keys with the right perms, also make sure the credential store as a whole is
2964 if (uid_is_valid(uid
) && uid
!= getuid()) {
2965 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
2967 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2973 if (fchown(dfd
, uid
, GID_INVALID
) < 0)
2981 static int setup_credentials_internal(
2982 const ExecContext
*context
,
2983 const ExecParameters
*params
,
2985 const char *final
, /* This is where the credential store shall eventually end up at */
2986 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
2987 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
2988 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
2991 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
2992 * if we mounted something; false if we definitely can't mount anything */
3000 if (reuse_workspace
) {
3001 r
= path_is_mount_point(workspace
, NULL
, 0);
3005 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
3007 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
3009 workspace_mounted
= -1; /* ditto */
3011 r
= path_is_mount_point(final
, NULL
, 0);
3015 /* If the final place already has something mounted, we use that. If the workspace also has
3016 * something mounted we assume it's actually the same mount (but with MS_RDONLY
3018 final_mounted
= true;
3020 if (workspace_mounted
< 0) {
3021 /* If the final place is mounted, but the workspace we isn't, then let's bind mount
3022 * the final version to the workspace, and make it writable, so that we can make
3025 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
3029 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3033 workspace_mounted
= true;
3036 final_mounted
= false;
3038 if (workspace_mounted
< 0) {
3039 /* Nothing is mounted on the workspace yet, let's try to mount something now */
3040 for (int try = 0;; try++) {
3043 /* Try "ramfs" first, since it's not swap backed */
3044 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
3046 workspace_mounted
= true;
3050 } else if (try == 1) {
3051 _cleanup_free_
char *opts
= NULL
;
3053 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%zu", (size_t) CREDENTIALS_TOTAL_SIZE_MAX
) < 0)
3056 /* Fall back to "tmpfs" otherwise */
3057 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
3059 workspace_mounted
= true;
3064 /* 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. */
3065 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
3067 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
3070 if (must_mount
) /* If we it's not OK to use the plain directory
3071 * fallback, propagate all errors too */
3074 /* If we lack privileges to bind mount stuff, then let's gracefully
3075 * proceed for compat with container envs, and just use the final dir
3078 workspace_mounted
= false;
3082 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
3083 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3087 workspace_mounted
= true;
3093 assert(!must_mount
|| workspace_mounted
> 0);
3094 where
= workspace_mounted
? workspace
: final
;
3096 (void) label_fix_full(AT_FDCWD
, where
, final
, 0);
3098 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
3102 if (workspace_mounted
) {
3103 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
3104 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3108 /* And mount it to the final place, read-only */
3110 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
3112 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
3116 _cleanup_free_
char *parent
= NULL
;
3118 /* If we do not have our own mount put used the plain directory fallback, then we need to
3119 * open access to the top-level credential directory and the per-service directory now */
3121 r
= path_extract_directory(final
, &parent
);
3124 if (chmod(parent
, 0755) < 0)
3131 static int setup_credentials(
3132 const ExecContext
*context
,
3133 const ExecParameters
*params
,
3137 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
3143 if (!exec_context_has_credentials(context
))
3146 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
3149 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
3150 * and the subdir we mount over with a read-only file system readable by the service's user */
3151 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
3155 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
3156 if (r
< 0 && r
!= -EEXIST
)
3159 p
= path_join(q
, unit
);
3163 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
3164 if (r
< 0 && r
!= -EEXIST
)
3167 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
3169 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
3171 /* If this is not a privilege or support issue then propagate the error */
3172 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
3175 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
3176 * it into place, so that users can't access half-initialized credential stores. */
3177 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
3181 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
3182 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
3183 * after it is fully set up */
3184 u
= path_join(t
, unit
);
3188 FOREACH_STRING(i
, t
, u
) {
3189 r
= mkdir_label(i
, 0700);
3190 if (r
< 0 && r
!= -EEXIST
)
3194 r
= setup_credentials_internal(
3198 p
, /* final mount point */
3199 u
, /* temporary workspace to overmount */
3200 true, /* reuse the workspace if it is already a mount */
3201 false, /* it's OK to fall back to a plain directory if we can't mount anything */
3204 (void) rmdir(u
); /* remove the workspace again if we can. */
3209 } else if (r
== 0) {
3211 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
3212 * we can use the same directory for all cases, after turning off propagation. Question
3213 * though is: where do we turn off propagation exactly, and where do we place the workspace
3214 * directory? We need some place that is guaranteed to be a mount point in the host, and
3215 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
3216 * since we ultimately want to move the resulting file system there, i.e. we need propagation
3217 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
3218 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
3219 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
3220 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
3221 * propagation on the former, and then overmount the latter.
3223 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
3224 * for this purpose, but there are few other candidates that work equally well for us, and
3225 * given that the we do this in a privately namespaced short-lived single-threaded process
3226 * that no one else sees this should be OK to do. */
3228 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
3232 r
= setup_credentials_internal(
3236 p
, /* final mount point */
3237 "/dev/shm", /* temporary workspace to overmount */
3238 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
3239 true, /* insist that something is mounted, do not allow fallback to plain directory */
3244 _exit(EXIT_SUCCESS
);
3247 _exit(EXIT_FAILURE
);
3254 static int setup_smack(
3255 const Manager
*manager
,
3256 const ExecContext
*context
,
3257 int executable_fd
) {
3261 assert(executable_fd
>= 0);
3263 if (context
->smack_process_label
) {
3264 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
3267 } else if (manager
->default_smack_process_label
) {
3268 _cleanup_free_
char *exec_label
= NULL
;
3270 r
= mac_smack_read_fd(executable_fd
, SMACK_ATTR_EXEC
, &exec_label
);
3271 if (r
< 0 && !ERRNO_IS_XATTR_ABSENT(r
))
3274 r
= mac_smack_apply_pid(0, exec_label
? : manager
->default_smack_process_label
);
3283 static int compile_bind_mounts(
3284 const ExecContext
*context
,
3285 const ExecParameters
*params
,
3286 BindMount
**ret_bind_mounts
,
3287 size_t *ret_n_bind_mounts
,
3288 char ***ret_empty_directories
) {
3290 _cleanup_strv_free_
char **empty_directories
= NULL
;
3291 BindMount
*bind_mounts
;
3297 assert(ret_bind_mounts
);
3298 assert(ret_n_bind_mounts
);
3299 assert(ret_empty_directories
);
3301 n
= context
->n_bind_mounts
;
3302 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3303 if (!params
->prefix
[t
])
3306 for (size_t i
= 0; i
< context
->directories
[t
].n_items
; i
++)
3307 n
+= !context
->directories
[t
].items
[i
].only_create
;
3311 *ret_bind_mounts
= NULL
;
3312 *ret_n_bind_mounts
= 0;
3313 *ret_empty_directories
= NULL
;
3317 bind_mounts
= new(BindMount
, n
);
3321 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
3322 BindMount
*item
= context
->bind_mounts
+ i
;
3325 s
= strdup(item
->source
);
3331 d
= strdup(item
->destination
);
3338 bind_mounts
[h
++] = (BindMount
) {
3341 .read_only
= item
->read_only
,
3342 .recursive
= item
->recursive
,
3343 .ignore_enoent
= item
->ignore_enoent
,
3347 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3348 if (!params
->prefix
[t
])
3351 if (context
->directories
[t
].n_items
== 0)
3354 if (exec_directory_is_private(context
, t
) &&
3355 !exec_context_with_rootfs(context
)) {
3358 /* So this is for a dynamic user, and we need to make sure the process can access its own
3359 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3360 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3362 private_root
= path_join(params
->prefix
[t
], "private");
3363 if (!private_root
) {
3368 r
= strv_consume(&empty_directories
, private_root
);
3373 for (size_t i
= 0; i
< context
->directories
[t
].n_items
; i
++) {
3376 /* When one of the parent directories is in the list, we cannot create the symlink
3377 * for the child directory. See also the comments in setup_exec_directory(). */
3378 if (context
->directories
[t
].items
[i
].only_create
)
3381 if (exec_directory_is_private(context
, t
))
3382 s
= path_join(params
->prefix
[t
], "private", context
->directories
[t
].items
[i
].path
);
3384 s
= path_join(params
->prefix
[t
], context
->directories
[t
].items
[i
].path
);
3390 if (exec_directory_is_private(context
, t
) &&
3391 exec_context_with_rootfs(context
))
3392 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3393 * directory is not created on the root directory. So, let's bind-mount the directory
3394 * on the 'non-private' place. */
3395 d
= path_join(params
->prefix
[t
], context
->directories
[t
].items
[i
].path
);
3404 bind_mounts
[h
++] = (BindMount
) {
3408 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3410 .ignore_enoent
= false,
3417 *ret_bind_mounts
= bind_mounts
;
3418 *ret_n_bind_mounts
= n
;
3419 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3424 bind_mount_free_many(bind_mounts
, h
);
3428 /* ret_symlinks will contain a list of pairs src:dest that describes
3429 * the symlinks to create later on. For example, the symlinks needed
3430 * to safely give private directories to DynamicUser=1 users. */
3431 static int compile_symlinks(
3432 const ExecContext
*context
,
3433 const ExecParameters
*params
,
3434 char ***ret_symlinks
) {
3436 _cleanup_strv_free_
char **symlinks
= NULL
;
3441 assert(ret_symlinks
);
3443 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3444 for (size_t i
= 0; i
< context
->directories
[dt
].n_items
; i
++) {
3445 _cleanup_free_
char *private_path
= NULL
, *path
= NULL
;
3447 STRV_FOREACH(symlink
, context
->directories
[dt
].items
[i
].symlinks
) {
3448 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
3450 src_abs
= path_join(params
->prefix
[dt
], context
->directories
[dt
].items
[i
].path
);
3451 dst_abs
= path_join(params
->prefix
[dt
], *symlink
);
3452 if (!src_abs
|| !dst_abs
)
3455 r
= strv_consume_pair(&symlinks
, TAKE_PTR(src_abs
), TAKE_PTR(dst_abs
));
3460 if (!exec_directory_is_private(context
, dt
) ||
3461 exec_context_with_rootfs(context
) ||
3462 context
->directories
[dt
].items
[i
].only_create
)
3465 private_path
= path_join(params
->prefix
[dt
], "private", context
->directories
[dt
].items
[i
].path
);
3469 path
= path_join(params
->prefix
[dt
], context
->directories
[dt
].items
[i
].path
);
3473 r
= strv_consume_pair(&symlinks
, TAKE_PTR(private_path
), TAKE_PTR(path
));
3479 *ret_symlinks
= TAKE_PTR(symlinks
);
3484 static bool insist_on_sandboxing(
3485 const ExecContext
*context
,
3486 const char *root_dir
,
3487 const char *root_image
,
3488 const BindMount
*bind_mounts
,
3489 size_t n_bind_mounts
) {
3492 assert(n_bind_mounts
== 0 || bind_mounts
);
3494 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3495 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3496 * rearrange stuff in a way we cannot ignore gracefully. */
3498 if (context
->n_temporary_filesystems
> 0)
3501 if (root_dir
|| root_image
)
3504 if (context
->n_mount_images
> 0)
3507 if (context
->dynamic_user
)
3510 if (context
->n_extension_images
> 0 || !strv_isempty(context
->extension_directories
))
3513 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3515 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3516 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3519 if (context
->log_namespace
)
3525 static int apply_mount_namespace(
3527 ExecCommandFlags command_flags
,
3528 const ExecContext
*context
,
3529 const ExecParameters
*params
,
3530 const ExecRuntime
*runtime
,
3531 char **error_path
) {
3533 _cleanup_strv_free_
char **empty_directories
= NULL
, **symlinks
= NULL
;
3534 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3535 const char *root_dir
= NULL
, *root_image
= NULL
;
3536 _cleanup_free_
char *creds_path
= NULL
, *incoming_dir
= NULL
, *propagate_dir
= NULL
,
3537 *extension_dir
= NULL
;
3538 NamespaceInfo ns_info
;
3539 bool needs_sandboxing
;
3540 BindMount
*bind_mounts
= NULL
;
3541 size_t n_bind_mounts
= 0;
3546 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3547 root_image
= context
->root_image
;
3550 root_dir
= context
->root_directory
;
3553 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3557 /* Symlinks for exec dirs are set up after other mounts, before they are made read-only. */
3558 r
= compile_symlinks(context
, params
, &symlinks
);
3562 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3563 if (needs_sandboxing
) {
3564 /* The runtime struct only contains the parent of the private /tmp,
3565 * which is non-accessible to world users. Inside of it there's a /tmp
3566 * that is sticky, and that's the one we want to use here.
3567 * This does not apply when we are using /run/systemd/empty as fallback. */
3569 if (context
->private_tmp
&& runtime
) {
3570 if (streq_ptr(runtime
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3571 tmp_dir
= runtime
->tmp_dir
;
3572 else if (runtime
->tmp_dir
)
3573 tmp_dir
= strjoina(runtime
->tmp_dir
, "/tmp");
3575 if (streq_ptr(runtime
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3576 var_tmp_dir
= runtime
->var_tmp_dir
;
3577 else if (runtime
->var_tmp_dir
)
3578 var_tmp_dir
= strjoina(runtime
->var_tmp_dir
, "/tmp");
3581 ns_info
= (NamespaceInfo
) {
3582 .ignore_protect_paths
= false,
3583 .private_dev
= context
->private_devices
,
3584 .protect_control_groups
= context
->protect_control_groups
,
3585 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3586 .protect_kernel_modules
= context
->protect_kernel_modules
,
3587 .protect_kernel_logs
= context
->protect_kernel_logs
,
3588 .protect_hostname
= context
->protect_hostname
,
3589 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3590 .private_mounts
= context
->private_mounts
,
3591 .protect_home
= context
->protect_home
,
3592 .protect_system
= context
->protect_system
,
3593 .protect_proc
= context
->protect_proc
,
3594 .proc_subset
= context
->proc_subset
,
3595 .private_ipc
= context
->private_ipc
|| context
->ipc_namespace_path
,
3596 /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
3597 .mount_nosuid
= context
->no_new_privileges
&& !mac_selinux_use(),
3599 } else if (!context
->dynamic_user
&& root_dir
)
3601 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3602 * sandbox info, otherwise enforce it, don't ignore protected paths and
3603 * fail if we are enable to apply the sandbox inside the mount namespace.
3605 ns_info
= (NamespaceInfo
) {
3606 .ignore_protect_paths
= true,
3609 ns_info
= (NamespaceInfo
) {};
3611 if (context
->mount_flags
== MS_SHARED
)
3612 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3614 if (exec_context_has_credentials(context
) &&
3615 params
->prefix
[EXEC_DIRECTORY_RUNTIME
] &&
3616 FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
3617 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3624 if (MANAGER_IS_SYSTEM(u
->manager
)) {
3625 propagate_dir
= path_join("/run/systemd/propagate/", u
->id
);
3626 if (!propagate_dir
) {
3631 incoming_dir
= strdup("/run/systemd/incoming");
3632 if (!incoming_dir
) {
3637 extension_dir
= strdup("/run/systemd/unit-extensions");
3638 if (!extension_dir
) {
3643 if (asprintf(&extension_dir
, "/run/user/" UID_FMT
"/systemd/unit-extensions", geteuid()) < 0) {
3648 r
= setup_namespace(root_dir
, root_image
, context
->root_image_options
,
3649 &ns_info
, context
->read_write_paths
,
3650 needs_sandboxing
? context
->read_only_paths
: NULL
,
3651 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3652 needs_sandboxing
? context
->exec_paths
: NULL
,
3653 needs_sandboxing
? context
->no_exec_paths
: NULL
,
3658 context
->temporary_filesystems
,
3659 context
->n_temporary_filesystems
,
3660 context
->mount_images
,
3661 context
->n_mount_images
,
3665 context
->log_namespace
,
3666 context
->mount_flags
,
3667 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3668 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3669 context
->root_verity
,
3670 context
->extension_images
,
3671 context
->n_extension_images
,
3672 context
->extension_directories
,
3676 root_dir
|| root_image
? params
->notify_socket
: NULL
,
3679 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3680 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3681 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3682 * completely different execution environment. */
3684 if (insist_on_sandboxing(
3686 root_dir
, root_image
,
3689 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3690 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3691 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
3695 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3701 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
3705 static int apply_working_directory(
3706 const ExecContext
*context
,
3707 const ExecParameters
*params
,
3714 assert(exit_status
);
3716 if (context
->working_directory_home
) {
3719 *exit_status
= EXIT_CHDIR
;
3726 wd
= empty_to_root(context
->working_directory
);
3728 if (params
->flags
& EXEC_APPLY_CHROOT
)
3731 d
= prefix_roota(context
->root_directory
, wd
);
3733 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3734 *exit_status
= EXIT_CHDIR
;
3741 static int apply_root_directory(
3742 const ExecContext
*context
,
3743 const ExecParameters
*params
,
3744 const bool needs_mount_ns
,
3748 assert(exit_status
);
3750 if (params
->flags
& EXEC_APPLY_CHROOT
)
3751 if (!needs_mount_ns
&& context
->root_directory
)
3752 if (chroot(context
->root_directory
) < 0) {
3753 *exit_status
= EXIT_CHROOT
;
3760 static int setup_keyring(
3762 const ExecContext
*context
,
3763 const ExecParameters
*p
,
3764 uid_t uid
, gid_t gid
) {
3766 key_serial_t keyring
;
3775 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3776 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3777 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3778 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3779 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3780 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3782 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3785 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3786 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3787 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3788 * & group is just as nasty as acquiring a reference to the user keyring. */
3790 saved_uid
= getuid();
3791 saved_gid
= getgid();
3793 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3794 if (setregid(gid
, -1) < 0)
3795 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3798 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3799 if (setreuid(uid
, -1) < 0) {
3800 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3805 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3806 if (keyring
== -1) {
3807 if (errno
== ENOSYS
)
3808 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3809 else if (ERRNO_IS_PRIVILEGE(errno
))
3810 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3811 else if (errno
== EDQUOT
)
3812 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3814 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3819 /* When requested link the user keyring into the session keyring. */
3820 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3822 if (keyctl(KEYCTL_LINK
,
3823 KEY_SPEC_USER_KEYRING
,
3824 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3825 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3830 /* Restore uid/gid back */
3831 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3832 if (setreuid(saved_uid
, -1) < 0) {
3833 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3838 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3839 if (setregid(saved_gid
, -1) < 0)
3840 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3843 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3844 if (!sd_id128_is_null(u
->invocation_id
)) {
3847 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3849 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3851 if (keyctl(KEYCTL_SETPERM
, key
,
3852 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3853 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3854 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3859 /* Revert back uid & gid for the last time, and exit */
3860 /* no extra logging, as only the first already reported error matters */
3861 if (getuid() != saved_uid
)
3862 (void) setreuid(saved_uid
, -1);
3864 if (getgid() != saved_gid
)
3865 (void) setregid(saved_gid
, -1);
3870 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3876 array
[(*n
)++] = pair
[0];
3878 array
[(*n
)++] = pair
[1];
3881 static int close_remaining_fds(
3882 const ExecParameters
*params
,
3883 const ExecRuntime
*runtime
,
3884 const DynamicCreds
*dcreds
,
3887 const int *fds
, size_t n_fds
) {
3889 size_t n_dont_close
= 0;
3890 int dont_close
[n_fds
+ 12];
3894 if (params
->stdin_fd
>= 0)
3895 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3896 if (params
->stdout_fd
>= 0)
3897 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3898 if (params
->stderr_fd
>= 0)
3899 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3902 dont_close
[n_dont_close
++] = socket_fd
;
3904 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3905 n_dont_close
+= n_fds
;
3909 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3910 append_socket_pair(dont_close
, &n_dont_close
, runtime
->ipcns_storage_socket
);
3915 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3917 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3920 if (user_lookup_fd
>= 0)
3921 dont_close
[n_dont_close
++] = user_lookup_fd
;
3923 return close_all_fds(dont_close
, n_dont_close
);
3926 static int send_user_lookup(
3934 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3935 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3938 if (user_lookup_fd
< 0)
3941 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3944 if (writev(user_lookup_fd
,
3946 IOVEC_INIT(&uid
, sizeof(uid
)),
3947 IOVEC_INIT(&gid
, sizeof(gid
)),
3948 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3954 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3961 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3966 if (!c
->working_directory_home
)
3969 r
= get_home_dir(buf
);
3977 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3978 _cleanup_strv_free_
char ** list
= NULL
;
3985 assert(c
->dynamic_user
);
3987 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3988 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3991 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3992 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3998 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
4001 if (exec_directory_is_private(c
, t
))
4002 e
= path_join(p
->prefix
[t
], "private", c
->directories
[t
].items
[i
].path
);
4004 e
= path_join(p
->prefix
[t
], c
->directories
[t
].items
[i
].path
);
4008 r
= strv_consume(&list
, e
);
4014 *ret
= TAKE_PTR(list
);
4019 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
4020 bool using_subcgroup
;
4026 if (!params
->cgroup_path
)
4029 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
4030 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
4031 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
4032 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
4033 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
4034 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
4035 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
4036 * flag, which is only passed for the former statements, not for the latter. */
4038 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
4039 if (using_subcgroup
)
4040 p
= path_join(params
->cgroup_path
, ".control");
4042 p
= strdup(params
->cgroup_path
);
4047 return using_subcgroup
;
4050 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
4051 _cleanup_(cpu_set_reset
) CPUSet s
= {};
4057 if (!c
->numa_policy
.nodes
.set
) {
4058 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
4062 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
4068 return cpu_set_add_all(ret
, &s
);
4071 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
4074 return c
->cpu_affinity_from_numa
;
4077 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
4082 assert(*n_fds
< fds_size
);
4090 if (fd
< 3 + (int) *n_fds
) {
4091 /* Let's move the fd up, so that it's outside of the fd range we will use to store
4092 * the fds we pass to the process (or which are closed only during execve). */
4094 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
4098 close_and_replace(fd
, r
);
4101 *ret_fd
= fds
[*n_fds
] = fd
;
4106 static int exec_child(
4108 const ExecCommand
*command
,
4109 const ExecContext
*context
,
4110 const ExecParameters
*params
,
4111 ExecRuntime
*runtime
,
4112 DynamicCreds
*dcreds
,
4114 const int named_iofds
[static 3],
4116 size_t n_socket_fds
,
4117 size_t n_storage_fds
,
4122 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **joined_exec_search_path
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
4123 int r
, ngids
= 0, exec_fd
;
4124 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
4125 const char *username
= NULL
, *groupname
= NULL
;
4126 _cleanup_free_
char *home_buffer
= NULL
;
4127 const char *home
= NULL
, *shell
= NULL
;
4128 char **final_argv
= NULL
;
4129 dev_t journal_stream_dev
= 0;
4130 ino_t journal_stream_ino
= 0;
4131 bool userns_set_up
= false;
4132 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
4133 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
4134 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
4135 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
4137 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
4138 bool use_selinux
= false;
4141 bool use_smack
= false;
4144 bool use_apparmor
= false;
4146 uid_t saved_uid
= getuid();
4147 gid_t saved_gid
= getgid();
4148 uid_t uid
= UID_INVALID
;
4149 gid_t gid
= GID_INVALID
;
4150 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
4151 n_keep_fds
; /* total number of fds not to close */
4153 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
4154 int ngids_after_pam
= 0;
4160 assert(exit_status
);
4162 /* Explicitly test for CVE-2021-4034 inspired invocations */
4163 assert(command
->path
);
4164 assert(!strv_isempty(command
->argv
));
4166 rename_process_from_path(command
->path
);
4168 /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main
4169 * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially,
4170 * both of which will be demoted to SIG_DFL. */
4171 (void) default_signals(SIGNALS_CRASH_HANDLER
,
4174 if (context
->ignore_sigpipe
)
4175 (void) ignore_signals(SIGPIPE
);
4177 r
= reset_signal_mask();
4179 *exit_status
= EXIT_SIGNAL_MASK
;
4180 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
4183 if (params
->idle_pipe
)
4184 do_idle_pipe_dance(params
->idle_pipe
);
4186 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
4187 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
4188 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
4189 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
4192 log_set_open_when_needed(true);
4194 /* In case anything used libc syslog(), close this here, too */
4197 int keep_fds
[n_fds
+ 3];
4198 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
4201 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
4203 *exit_status
= EXIT_FDS
;
4204 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4208 if (unit
->manager
->restrict_fs
) {
4209 int bpf_map_fd
= lsm_bpf_map_restrict_fs_fd(unit
);
4210 if (bpf_map_fd
< 0) {
4211 *exit_status
= EXIT_FDS
;
4212 return log_unit_error_errno(unit
, bpf_map_fd
, "Failed to get restrict filesystems BPF map fd: %m");
4215 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, bpf_map_fd
, &bpf_map_fd
);
4217 *exit_status
= EXIT_FDS
;
4218 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4223 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
4225 *exit_status
= EXIT_FDS
;
4226 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
4229 if (!context
->same_pgrp
&&
4231 *exit_status
= EXIT_SETSID
;
4232 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
4235 exec_context_tty_reset(context
, params
);
4237 if (unit_shall_confirm_spawn(unit
)) {
4238 _cleanup_free_
char *cmdline
= NULL
;
4240 cmdline
= quote_command_line(command
->argv
, SHELL_ESCAPE_EMPTY
);
4242 *exit_status
= EXIT_MEMORY
;
4246 r
= ask_for_confirmation(context
, params
->confirm_spawn
, unit
, cmdline
);
4247 if (r
!= CONFIRM_EXECUTE
) {
4248 if (r
== CONFIRM_PRETEND_SUCCESS
) {
4249 *exit_status
= EXIT_SUCCESS
;
4252 *exit_status
= EXIT_CONFIRM
;
4253 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
4254 "Execution cancelled by the user");
4258 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
4259 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
4260 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
4261 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
4262 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
4263 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
4264 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
4265 *exit_status
= EXIT_MEMORY
;
4266 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
4269 if (context
->dynamic_user
&& dcreds
) {
4270 _cleanup_strv_free_
char **suggested_paths
= NULL
;
4272 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
4273 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
4274 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
4275 *exit_status
= EXIT_USER
;
4276 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
4279 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
4281 *exit_status
= EXIT_MEMORY
;
4285 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
4287 *exit_status
= EXIT_USER
;
4289 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4290 "Failed to update dynamic user credentials: User or group with specified name already exists.");
4291 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
4294 if (!uid_is_valid(uid
)) {
4295 *exit_status
= EXIT_USER
;
4296 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "UID validation failed for \""UID_FMT
"\"", uid
);
4299 if (!gid_is_valid(gid
)) {
4300 *exit_status
= EXIT_USER
;
4301 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "GID validation failed for \""GID_FMT
"\"", gid
);
4305 username
= dcreds
->user
->name
;
4308 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
4310 *exit_status
= EXIT_USER
;
4311 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
4314 r
= get_fixed_group(context
, &groupname
, &gid
);
4316 *exit_status
= EXIT_GROUP
;
4317 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
4321 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
4322 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
4323 &supplementary_gids
, &ngids
);
4325 *exit_status
= EXIT_GROUP
;
4326 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
4329 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
4331 *exit_status
= EXIT_USER
;
4332 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
4335 user_lookup_fd
= safe_close(user_lookup_fd
);
4337 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
4339 *exit_status
= EXIT_CHDIR
;
4340 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
4343 /* If a socket is connected to STDIN/STDOUT/STDERR, we
4344 * must sure to drop O_NONBLOCK */
4346 (void) fd_nonblock(socket_fd
, false);
4348 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
4349 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
4350 if (params
->cgroup_path
) {
4351 _cleanup_free_
char *p
= NULL
;
4353 r
= exec_parameters_get_cgroup_path(params
, &p
);
4355 *exit_status
= EXIT_CGROUP
;
4356 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
4359 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
4360 if (r
== -EUCLEAN
) {
4361 *exit_status
= EXIT_CGROUP
;
4362 return log_unit_error_errno(unit
, r
, "Failed to attach process to cgroup %s "
4363 "because the cgroup or one of its parents or "
4364 "siblings is in the threaded mode: %m", p
);
4367 *exit_status
= EXIT_CGROUP
;
4368 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
4372 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4373 r
= open_shareable_ns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
, CLONE_NEWNET
);
4375 *exit_status
= EXIT_NETWORK
;
4376 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
4380 if (context
->ipc_namespace_path
&& runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
4381 r
= open_shareable_ns_path(runtime
->ipcns_storage_socket
, context
->ipc_namespace_path
, CLONE_NEWIPC
);
4383 *exit_status
= EXIT_NAMESPACE
;
4384 return log_unit_error_errno(unit
, r
, "Failed to open IPC namespace path %s: %m", context
->ipc_namespace_path
);
4388 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
4390 *exit_status
= EXIT_STDIN
;
4391 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
4394 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
4396 *exit_status
= EXIT_STDOUT
;
4397 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
4400 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
4402 *exit_status
= EXIT_STDERR
;
4403 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
4406 if (context
->oom_score_adjust_set
) {
4407 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
4408 * prohibit write access to this file, and we shouldn't trip up over that. */
4409 r
= set_oom_score_adjust(context
->oom_score_adjust
);
4410 if (ERRNO_IS_PRIVILEGE(r
))
4411 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
4413 *exit_status
= EXIT_OOM_ADJUST
;
4414 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
4418 if (context
->coredump_filter_set
) {
4419 r
= set_coredump_filter(context
->coredump_filter
);
4420 if (ERRNO_IS_PRIVILEGE(r
))
4421 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
4423 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
4426 if (context
->nice_set
) {
4427 r
= setpriority_closest(context
->nice
);
4429 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
4432 if (context
->cpu_sched_set
) {
4433 struct sched_param param
= {
4434 .sched_priority
= context
->cpu_sched_priority
,
4437 r
= sched_setscheduler(0,
4438 context
->cpu_sched_policy
|
4439 (context
->cpu_sched_reset_on_fork
?
4440 SCHED_RESET_ON_FORK
: 0),
4443 *exit_status
= EXIT_SETSCHEDULER
;
4444 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
4448 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
4449 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
4450 const CPUSet
*cpu_set
;
4452 if (context
->cpu_affinity_from_numa
) {
4453 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
4455 *exit_status
= EXIT_CPUAFFINITY
;
4456 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
4459 cpu_set
= &converted_cpu_set
;
4461 cpu_set
= &context
->cpu_set
;
4463 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
4464 *exit_status
= EXIT_CPUAFFINITY
;
4465 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
4469 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
4470 r
= apply_numa_policy(&context
->numa_policy
);
4471 if (r
== -EOPNOTSUPP
)
4472 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
4474 *exit_status
= EXIT_NUMA_POLICY
;
4475 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
4479 if (context
->ioprio_set
)
4480 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
4481 *exit_status
= EXIT_IOPRIO
;
4482 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
4485 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
4486 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
4487 *exit_status
= EXIT_TIMERSLACK
;
4488 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
4491 if (context
->personality
!= PERSONALITY_INVALID
) {
4492 r
= safe_personality(context
->personality
);
4494 *exit_status
= EXIT_PERSONALITY
;
4495 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
4499 if (context
->utmp_id
) {
4500 const char *line
= context
->tty_path
?
4501 (path_startswith(context
->tty_path
, "/dev/") ?: context
->tty_path
) :
4503 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
4505 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
4506 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
4511 if (uid_is_valid(uid
)) {
4512 r
= chown_terminal(STDIN_FILENO
, uid
);
4514 *exit_status
= EXIT_STDIN
;
4515 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
4519 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4520 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4521 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4522 * touch a single hierarchy too. */
4523 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
4524 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
4526 *exit_status
= EXIT_CGROUP
;
4527 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4531 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4533 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4534 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, needs_mount_namespace
, exit_status
);
4536 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4539 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4540 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4542 *exit_status
= EXIT_CREDENTIALS
;
4543 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4547 r
= build_environment(
4559 *exit_status
= EXIT_MEMORY
;
4563 r
= build_pass_environment(context
, &pass_env
);
4565 *exit_status
= EXIT_MEMORY
;
4569 /* The $PATH variable is set to the default path in params->environment. However, this is overridden
4570 * if user-specified fields have $PATH set. The intention is to also override $PATH if the unit does
4571 * not specify PATH but the unit has ExecSearchPath. */
4572 if (!strv_isempty(context
->exec_search_path
)) {
4573 _cleanup_free_
char *joined
= NULL
;
4575 joined
= strv_join(context
->exec_search_path
, ":");
4577 *exit_status
= EXIT_MEMORY
;
4581 r
= strv_env_assign(&joined_exec_search_path
, "PATH", joined
);
4583 *exit_status
= EXIT_MEMORY
;
4588 accum_env
= strv_env_merge(params
->environment
,
4590 joined_exec_search_path
,
4592 context
->environment
,
4595 *exit_status
= EXIT_MEMORY
;
4598 accum_env
= strv_env_clean(accum_env
);
4600 (void) umask(context
->umask
);
4602 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4604 *exit_status
= EXIT_KEYRING
;
4605 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4608 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted
4610 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4612 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked
4613 * for it, and the kernel doesn't actually support ambient caps. */
4614 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4616 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly
4617 * excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not
4619 if (needs_ambient_hack
)
4620 needs_setuid
= false;
4622 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4624 if (needs_sandboxing
) {
4625 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on
4626 * /sys being present. The actual MAC context application will happen later, as late as
4627 * possible, to avoid impacting our own code paths. */
4630 use_selinux
= mac_selinux_use();
4633 use_smack
= mac_smack_use();
4636 use_apparmor
= mac_apparmor_use();
4640 if (needs_sandboxing
) {
4643 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4644 * is set here. (See below.) */
4646 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4648 *exit_status
= EXIT_LIMITS
;
4649 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4653 if (needs_setuid
&& context
->pam_name
&& username
) {
4654 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4655 * wins here. (See above.) */
4657 /* All fds passed in the fds array will be closed in the pam child process. */
4658 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4660 *exit_status
= EXIT_PAM
;
4661 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4664 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4665 if (ngids_after_pam
< 0) {
4666 *exit_status
= EXIT_MEMORY
;
4667 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4671 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4672 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4673 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4674 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4676 userns_set_up
= true;
4677 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4679 *exit_status
= EXIT_USER
;
4680 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4684 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4686 if (ns_type_supported(NAMESPACE_NET
)) {
4687 r
= setup_shareable_ns(runtime
->netns_storage_socket
, CLONE_NEWNET
);
4689 log_unit_warning_errno(unit
, r
,
4690 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4692 *exit_status
= EXIT_NETWORK
;
4693 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4695 } else if (context
->network_namespace_path
) {
4696 *exit_status
= EXIT_NETWORK
;
4697 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4698 "NetworkNamespacePath= is not supported, refusing.");
4700 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4703 if ((context
->private_ipc
|| context
->ipc_namespace_path
) && runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
4705 if (ns_type_supported(NAMESPACE_IPC
)) {
4706 r
= setup_shareable_ns(runtime
->ipcns_storage_socket
, CLONE_NEWIPC
);
4708 log_unit_warning_errno(unit
, r
,
4709 "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
4711 *exit_status
= EXIT_NAMESPACE
;
4712 return log_unit_error_errno(unit
, r
, "Failed to set up IPC namespacing: %m");
4714 } else if (context
->ipc_namespace_path
) {
4715 *exit_status
= EXIT_NAMESPACE
;
4716 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4717 "IPCNamespacePath= is not supported, refusing.");
4719 log_unit_warning(unit
, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
4722 if (needs_mount_namespace
) {
4723 _cleanup_free_
char *error_path
= NULL
;
4725 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4727 *exit_status
= EXIT_NAMESPACE
;
4728 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4729 error_path
? ": " : "", strempty(error_path
));
4733 if (needs_sandboxing
) {
4734 r
= apply_protect_hostname(unit
, context
, exit_status
);
4739 /* Drop groups as early as possible.
4740 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4741 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4743 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4744 int ngids_to_enforce
= 0;
4746 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4751 if (ngids_to_enforce
< 0) {
4752 *exit_status
= EXIT_MEMORY
;
4753 return log_unit_error_errno(unit
,
4755 "Failed to merge group lists. Group membership might be incorrect: %m");
4758 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4760 *exit_status
= EXIT_GROUP
;
4761 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4765 /* If the user namespace was not set up above, try to do it now.
4766 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4767 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4768 * case of mount namespaces being less privileged when the mount point list is copied from a
4769 * different user namespace). */
4771 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4772 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4774 *exit_status
= EXIT_USER
;
4775 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4779 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4782 _cleanup_free_
char *executable
= NULL
;
4783 _cleanup_close_
int executable_fd
= -1;
4784 r
= find_executable_full(command
->path
, /* root= */ NULL
, context
->exec_search_path
, false, &executable
, &executable_fd
);
4786 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4787 log_unit_struct_errno(unit
, LOG_INFO
, r
,
4788 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4789 LOG_UNIT_INVOCATION_ID(unit
),
4790 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4792 "EXECUTABLE=%s", command
->path
);
4796 *exit_status
= EXIT_EXEC
;
4798 return log_unit_struct_errno(unit
, LOG_INFO
, r
,
4799 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4800 LOG_UNIT_INVOCATION_ID(unit
),
4801 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4803 "EXECUTABLE=%s", command
->path
);
4806 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, executable_fd
, &executable_fd
);
4808 *exit_status
= EXIT_FDS
;
4809 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4813 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
) {
4818 else if (params
->n_socket_fds
== 1)
4819 /* If stdin is not connected to a socket but we are triggered by exactly one socket unit then we
4820 * use context from that fd to compute the label. */
4821 fd
= params
->fds
[0];
4824 r
= mac_selinux_get_child_mls_label(fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4826 if (!context
->selinux_context_ignore
) {
4827 *exit_status
= EXIT_SELINUX_CONTEXT
;
4828 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4830 log_unit_debug_errno(unit
, r
, "Failed to determine SELinux context, ignoring: %m");
4836 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4837 * more aggressive this time since socket_fd and the netns and ipcns fds we don't need anymore. We do keep the exec_fd
4838 * however if we have it as we want to keep it open until the final execve(). */
4840 r
= close_all_fds(keep_fds
, n_keep_fds
);
4842 r
= shift_fds(fds
, n_fds
);
4844 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4846 *exit_status
= EXIT_FDS
;
4847 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4850 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4851 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4852 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4855 secure_bits
= context
->secure_bits
;
4857 if (needs_sandboxing
) {
4860 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4861 * requested. (Note this is placed after the general resource limit initialization, see
4862 * above, in order to take precedence.) */
4863 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4864 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4865 *exit_status
= EXIT_LIMITS
;
4866 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4871 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4872 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4874 r
= setup_smack(unit
->manager
, context
, executable_fd
);
4875 if (r
< 0 && !context
->smack_process_label_ignore
) {
4876 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4877 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4882 bset
= context
->capability_bounding_set
;
4883 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4884 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4885 * instead of us doing that */
4886 if (needs_ambient_hack
)
4887 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4888 (UINT64_C(1) << CAP_SETUID
) |
4889 (UINT64_C(1) << CAP_SETGID
);
4891 if (!cap_test_all(bset
)) {
4892 r
= capability_bounding_set_drop(bset
, false);
4894 *exit_status
= EXIT_CAPABILITIES
;
4895 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4899 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4901 * To be able to raise the ambient capabilities after setresuid() they have to be
4902 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4903 * After setresuid() the ambient capabilities can be raised as they are present in
4904 * the permitted and inhertiable set. However it is possible that someone wants to
4905 * set ambient capabilities without changing the user, so we also set the ambient
4906 * capabilities here.
4907 * The requested ambient capabilities are raised in the inheritable set if the
4908 * second argument is true. */
4909 if (!needs_ambient_hack
) {
4910 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4912 *exit_status
= EXIT_CAPABILITIES
;
4913 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4918 /* chroot to root directory first, before we lose the ability to chroot */
4919 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4921 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4924 if (uid_is_valid(uid
)) {
4925 r
= enforce_user(context
, uid
);
4927 *exit_status
= EXIT_USER
;
4928 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4931 if (!needs_ambient_hack
&&
4932 context
->capability_ambient_set
!= 0) {
4934 /* Raise the ambient capabilities after user change. */
4935 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4937 *exit_status
= EXIT_CAPABILITIES
;
4938 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4944 /* Apply working directory here, because the working directory might be on NFS and only the user running
4945 * this service might have the correct privilege to change to the working directory */
4946 r
= apply_working_directory(context
, params
, home
, exit_status
);
4948 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4950 if (needs_sandboxing
) {
4951 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4952 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4953 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4954 * are restricted. */
4958 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4961 r
= setexeccon(exec_context
);
4963 if (!context
->selinux_context_ignore
) {
4964 *exit_status
= EXIT_SELINUX_CONTEXT
;
4965 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4967 log_unit_debug_errno(unit
, r
, "Failed to change SELinux context to %s, ignoring: %m", exec_context
);
4974 if (use_apparmor
&& context
->apparmor_profile
) {
4975 r
= aa_change_onexec(context
->apparmor_profile
);
4976 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4977 *exit_status
= EXIT_APPARMOR_PROFILE
;
4978 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4983 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4984 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4986 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4987 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4988 * effective set here.
4989 * The effective set is overwritten during execve with the following values:
4990 * - ambient set (for non-root processes)
4991 * - (inheritable | bounding) set for root processes)
4993 * Hence there is no security impact to raise it in the effective set before execve
4995 r
= capability_gain_cap_setpcap(NULL
);
4997 *exit_status
= EXIT_CAPABILITIES
;
4998 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
5000 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
5001 *exit_status
= EXIT_SECUREBITS
;
5002 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
5006 if (context_has_no_new_privileges(context
))
5007 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
5008 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
5009 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
5013 r
= apply_address_families(unit
, context
);
5015 *exit_status
= EXIT_ADDRESS_FAMILIES
;
5016 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
5019 r
= apply_memory_deny_write_execute(unit
, context
);
5021 *exit_status
= EXIT_SECCOMP
;
5022 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
5025 r
= apply_restrict_realtime(unit
, context
);
5027 *exit_status
= EXIT_SECCOMP
;
5028 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
5031 r
= apply_restrict_suid_sgid(unit
, context
);
5033 *exit_status
= EXIT_SECCOMP
;
5034 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
5037 r
= apply_restrict_namespaces(unit
, context
);
5039 *exit_status
= EXIT_SECCOMP
;
5040 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
5043 r
= apply_protect_sysctl(unit
, context
);
5045 *exit_status
= EXIT_SECCOMP
;
5046 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
5049 r
= apply_protect_kernel_modules(unit
, context
);
5051 *exit_status
= EXIT_SECCOMP
;
5052 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
5055 r
= apply_protect_kernel_logs(unit
, context
);
5057 *exit_status
= EXIT_SECCOMP
;
5058 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
5061 r
= apply_protect_clock(unit
, context
);
5063 *exit_status
= EXIT_SECCOMP
;
5064 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
5067 r
= apply_private_devices(unit
, context
);
5069 *exit_status
= EXIT_SECCOMP
;
5070 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
5073 r
= apply_syscall_archs(unit
, context
);
5075 *exit_status
= EXIT_SECCOMP
;
5076 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
5079 r
= apply_lock_personality(unit
, context
);
5081 *exit_status
= EXIT_SECCOMP
;
5082 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
5085 r
= apply_syscall_log(unit
, context
);
5087 *exit_status
= EXIT_SECCOMP
;
5088 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
5091 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
5092 * by the filter as little as possible. */
5093 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
5095 *exit_status
= EXIT_SECCOMP
;
5096 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
5101 r
= apply_restrict_filesystems(unit
, context
);
5103 *exit_status
= EXIT_BPF
;
5104 return log_unit_error_errno(unit
, r
, "Failed to restrict filesystems: %m");
5110 if (!strv_isempty(context
->unset_environment
)) {
5113 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
5115 *exit_status
= EXIT_MEMORY
;
5119 strv_free_and_replace(accum_env
, ee
);
5122 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
5123 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
5124 if (!replaced_argv
) {
5125 *exit_status
= EXIT_MEMORY
;
5128 final_argv
= replaced_argv
;
5130 final_argv
= command
->argv
;
5132 if (DEBUG_LOGGING
) {
5133 _cleanup_free_
char *line
= NULL
;
5135 line
= quote_command_line(final_argv
, SHELL_ESCAPE_EMPTY
);
5137 *exit_status
= EXIT_MEMORY
;
5141 log_unit_struct(unit
, LOG_DEBUG
,
5142 "EXECUTABLE=%s", executable
,
5143 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
));
5149 /* We have finished with all our initializations. Let's now let the manager know that. From this point
5150 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
5152 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
5153 *exit_status
= EXIT_EXEC
;
5154 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
5158 r
= fexecve_or_execve(executable_fd
, executable
, final_argv
, accum_env
);
5163 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
5164 * that POLLHUP on it no longer means execve() succeeded. */
5166 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
5167 *exit_status
= EXIT_EXEC
;
5168 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
5172 *exit_status
= EXIT_EXEC
;
5173 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
5176 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
5177 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
5179 int exec_spawn(Unit
*unit
,
5180 ExecCommand
*command
,
5181 const ExecContext
*context
,
5182 const ExecParameters
*params
,
5183 ExecRuntime
*runtime
,
5184 DynamicCreds
*dcreds
,
5187 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
5188 _cleanup_free_
char *subcgroup_path
= NULL
;
5189 _cleanup_strv_free_
char **files_env
= NULL
;
5190 size_t n_storage_fds
= 0, n_socket_fds
= 0;
5191 _cleanup_free_
char *line
= NULL
;
5199 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
5201 if (context
->std_input
== EXEC_INPUT_SOCKET
||
5202 context
->std_output
== EXEC_OUTPUT_SOCKET
||
5203 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
5205 if (params
->n_socket_fds
> 1)
5206 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got more than one socket.");
5208 if (params
->n_socket_fds
== 0)
5209 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got no socket.");
5211 socket_fd
= params
->fds
[0];
5215 n_socket_fds
= params
->n_socket_fds
;
5216 n_storage_fds
= params
->n_storage_fds
;
5219 r
= exec_context_named_iofds(context
, params
, named_iofds
);
5221 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
5223 r
= exec_context_load_environment(unit
, context
, &files_env
);
5225 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
5227 line
= quote_command_line(command
->argv
, SHELL_ESCAPE_EMPTY
);
5231 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
5232 and, until the next SELinux policy changes, we save further reloads in future children. */
5233 mac_selinux_maybe_reload();
5235 log_unit_struct(unit
, LOG_DEBUG
,
5236 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
5237 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
5238 the mount namespace in the child, but we want to log
5239 from the parent, so we need to use the (possibly
5240 inaccurate) path here. */
5241 LOG_UNIT_INVOCATION_ID(unit
));
5243 if (params
->cgroup_path
) {
5244 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
5246 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
5247 if (r
> 0) { /* We are using a child cgroup */
5248 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
5250 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
5252 /* Normally we would not propagate the oomd xattrs to children but since we created this
5253 * sub-cgroup internally we should do it. */
5254 cgroup_oomd_xattr_apply(unit
, subcgroup_path
);
5260 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
5263 int exit_status
= EXIT_SUCCESS
;
5265 r
= exec_child(unit
,
5277 unit
->manager
->user_lookup_fds
[1],
5281 const char *status
=
5282 exit_status_to_string(exit_status
,
5283 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
5285 log_unit_struct_errno(unit
, LOG_ERR
, r
,
5286 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
5287 LOG_UNIT_INVOCATION_ID(unit
),
5288 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
5289 status
, command
->path
),
5290 "EXECUTABLE=%s", command
->path
);
5296 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
5298 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
5299 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
5300 * process will be killed too). */
5302 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
5304 exec_status_start(&command
->exec_status
, pid
);
5310 void exec_context_init(ExecContext
*c
) {
5314 c
->ioprio
= IOPRIO_DEFAULT_CLASS_AND_PRIO
;
5315 c
->cpu_sched_policy
= SCHED_OTHER
;
5316 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
5317 c
->syslog_level_prefix
= true;
5318 c
->ignore_sigpipe
= true;
5319 c
->timer_slack_nsec
= NSEC_INFINITY
;
5320 c
->personality
= PERSONALITY_INVALID
;
5321 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5322 c
->directories
[t
].mode
= 0755;
5323 c
->timeout_clean_usec
= USEC_INFINITY
;
5324 c
->capability_bounding_set
= CAP_ALL
;
5325 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
5326 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
5327 c
->log_level_max
= -1;
5329 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
5331 c
->tty_rows
= UINT_MAX
;
5332 c
->tty_cols
= UINT_MAX
;
5333 numa_policy_reset(&c
->numa_policy
);
5336 void exec_context_done(ExecContext
*c
) {
5339 c
->environment
= strv_free(c
->environment
);
5340 c
->environment_files
= strv_free(c
->environment_files
);
5341 c
->pass_environment
= strv_free(c
->pass_environment
);
5342 c
->unset_environment
= strv_free(c
->unset_environment
);
5344 rlimit_free_all(c
->rlimit
);
5346 for (size_t l
= 0; l
< 3; l
++) {
5347 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
5348 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
5351 c
->working_directory
= mfree(c
->working_directory
);
5352 c
->root_directory
= mfree(c
->root_directory
);
5353 c
->root_image
= mfree(c
->root_image
);
5354 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
5355 c
->root_hash
= mfree(c
->root_hash
);
5356 c
->root_hash_size
= 0;
5357 c
->root_hash_path
= mfree(c
->root_hash_path
);
5358 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
5359 c
->root_hash_sig_size
= 0;
5360 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
5361 c
->root_verity
= mfree(c
->root_verity
);
5362 c
->extension_images
= mount_image_free_many(c
->extension_images
, &c
->n_extension_images
);
5363 c
->extension_directories
= strv_free(c
->extension_directories
);
5364 c
->tty_path
= mfree(c
->tty_path
);
5365 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
5366 c
->user
= mfree(c
->user
);
5367 c
->group
= mfree(c
->group
);
5369 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
5371 c
->pam_name
= mfree(c
->pam_name
);
5373 c
->read_only_paths
= strv_free(c
->read_only_paths
);
5374 c
->read_write_paths
= strv_free(c
->read_write_paths
);
5375 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
5376 c
->exec_paths
= strv_free(c
->exec_paths
);
5377 c
->no_exec_paths
= strv_free(c
->no_exec_paths
);
5378 c
->exec_search_path
= strv_free(c
->exec_search_path
);
5380 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
5381 c
->bind_mounts
= NULL
;
5382 c
->n_bind_mounts
= 0;
5383 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
5384 c
->temporary_filesystems
= NULL
;
5385 c
->n_temporary_filesystems
= 0;
5386 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
5388 cpu_set_reset(&c
->cpu_set
);
5389 numa_policy_reset(&c
->numa_policy
);
5391 c
->utmp_id
= mfree(c
->utmp_id
);
5392 c
->selinux_context
= mfree(c
->selinux_context
);
5393 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
5394 c
->smack_process_label
= mfree(c
->smack_process_label
);
5396 c
->restrict_filesystems
= set_free(c
->restrict_filesystems
);
5398 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
5399 c
->syscall_archs
= set_free(c
->syscall_archs
);
5400 c
->address_families
= set_free(c
->address_families
);
5402 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5403 exec_directory_done(&c
->directories
[t
]);
5405 c
->log_level_max
= -1;
5407 exec_context_free_log_extra_fields(c
);
5409 c
->log_ratelimit_interval_usec
= 0;
5410 c
->log_ratelimit_burst
= 0;
5412 c
->stdin_data
= mfree(c
->stdin_data
);
5413 c
->stdin_data_size
= 0;
5415 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
5416 c
->ipc_namespace_path
= mfree(c
->ipc_namespace_path
);
5418 c
->log_namespace
= mfree(c
->log_namespace
);
5420 c
->load_credentials
= hashmap_free(c
->load_credentials
);
5421 c
->set_credentials
= hashmap_free(c
->set_credentials
);
5424 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
5427 if (!runtime_prefix
)
5430 for (size_t i
= 0; i
< c
->directories
[EXEC_DIRECTORY_RUNTIME
].n_items
; i
++) {
5431 _cleanup_free_
char *p
= NULL
;
5433 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
5434 p
= path_join(runtime_prefix
, "private", c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].path
);
5436 p
= path_join(runtime_prefix
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].path
);
5440 /* We execute this synchronously, since we need to be sure this is gone when we start the
5442 (void) rm_rf(p
, REMOVE_ROOT
);
5444 STRV_FOREACH(symlink
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].symlinks
) {
5445 _cleanup_free_
char *symlink_abs
= NULL
;
5447 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
5448 symlink_abs
= path_join(runtime_prefix
, "private", *symlink
);
5450 symlink_abs
= path_join(runtime_prefix
, *symlink
);
5454 (void) unlink(symlink_abs
);
5462 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
5463 _cleanup_free_
char *p
= NULL
;
5467 if (!runtime_prefix
|| !unit
)
5470 p
= path_join(runtime_prefix
, "credentials", unit
);
5474 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
5475 * unmount it, and afterwards remove the mount point */
5476 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
5477 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
5482 static void exec_command_done(ExecCommand
*c
) {
5485 c
->path
= mfree(c
->path
);
5486 c
->argv
= strv_free(c
->argv
);
5489 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
5490 for (size_t i
= 0; i
< n
; i
++)
5491 exec_command_done(c
+i
);
5494 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
5498 LIST_REMOVE(command
, c
, i
);
5499 exec_command_done(i
);
5506 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
5507 for (size_t i
= 0; i
< n
; i
++)
5508 c
[i
] = exec_command_free_list(c
[i
]);
5511 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
5512 for (size_t i
= 0; i
< n
; i
++)
5513 exec_status_reset(&c
[i
].exec_status
);
5516 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
5517 for (size_t i
= 0; i
< n
; i
++)
5518 LIST_FOREACH(command
, z
, c
[i
])
5519 exec_status_reset(&z
->exec_status
);
5522 typedef struct InvalidEnvInfo
{
5527 static void invalid_env(const char *p
, void *userdata
) {
5528 InvalidEnvInfo
*info
= userdata
;
5530 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
5533 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
5539 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
5542 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
5545 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
5548 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
5551 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
5554 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
5561 static int exec_context_named_iofds(
5562 const ExecContext
*c
,
5563 const ExecParameters
*p
,
5564 int named_iofds
[static 3]) {
5567 const char* stdio_fdname
[3];
5572 assert(named_iofds
);
5574 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
5575 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
5576 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
5578 for (size_t i
= 0; i
< 3; i
++)
5579 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
5581 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
5583 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
5584 if (named_iofds
[STDIN_FILENO
] < 0 &&
5585 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
5586 stdio_fdname
[STDIN_FILENO
] &&
5587 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
5589 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
5592 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
5593 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
5594 stdio_fdname
[STDOUT_FILENO
] &&
5595 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
5597 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5600 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5601 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5602 stdio_fdname
[STDERR_FILENO
] &&
5603 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5605 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5609 return targets
== 0 ? 0 : -ENOENT
;
5612 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***ret
) {
5613 _cleanup_strv_free_
char **v
= NULL
;
5619 STRV_FOREACH(i
, c
->environment_files
) {
5620 _cleanup_globfree_ glob_t pglob
= {};
5621 bool ignore
= false;
5629 if (!path_is_absolute(fn
)) {
5635 /* Filename supports globbing, take all matching files */
5636 r
= safe_glob(fn
, 0, &pglob
);
5643 /* When we don't match anything, -ENOENT should be returned */
5644 assert(pglob
.gl_pathc
> 0);
5646 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5647 _cleanup_strv_free_
char **p
= NULL
;
5649 r
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5656 /* Log invalid environment variables with filename */
5658 InvalidEnvInfo info
= {
5660 .path
= pglob
.gl_pathv
[n
]
5663 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5669 char **m
= strv_env_merge(v
, p
);
5673 strv_free_and_replace(v
, m
);
5683 static bool tty_may_match_dev_console(const char *tty
) {
5684 _cleanup_free_
char *resolved
= NULL
;
5689 tty
= skip_dev_prefix(tty
);
5691 /* trivial identity? */
5692 if (streq(tty
, "console"))
5695 if (resolve_dev_console(&resolved
) < 0)
5696 return true; /* if we could not resolve, assume it may */
5698 /* "tty0" means the active VC, so it may be the same sometimes */
5699 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5702 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5705 return ec
->tty_reset
||
5707 ec
->tty_vt_disallocate
||
5708 is_terminal_input(ec
->std_input
) ||
5709 is_terminal_output(ec
->std_output
) ||
5710 is_terminal_output(ec
->std_error
);
5713 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5715 return exec_context_may_touch_tty(ec
) &&
5716 tty_may_match_dev_console(exec_context_tty_path(ec
));
5719 static void strv_fprintf(FILE *f
, char **l
) {
5723 fprintf(f
, " %s", *g
);
5726 static void strv_dump(FILE* f
, const char *prefix
, const char *name
, char **strv
) {
5731 if (!strv_isempty(strv
)) {
5732 fprintf(f
, "%s%s:", prefix
, name
);
5733 strv_fprintf(f
, strv
);
5738 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5744 prefix
= strempty(prefix
);
5748 "%sWorkingDirectory: %s\n"
5749 "%sRootDirectory: %s\n"
5750 "%sNonBlocking: %s\n"
5751 "%sPrivateTmp: %s\n"
5752 "%sPrivateDevices: %s\n"
5753 "%sProtectKernelTunables: %s\n"
5754 "%sProtectKernelModules: %s\n"
5755 "%sProtectKernelLogs: %s\n"
5756 "%sProtectClock: %s\n"
5757 "%sProtectControlGroups: %s\n"
5758 "%sPrivateNetwork: %s\n"
5759 "%sPrivateUsers: %s\n"
5760 "%sProtectHome: %s\n"
5761 "%sProtectSystem: %s\n"
5762 "%sMountAPIVFS: %s\n"
5763 "%sIgnoreSIGPIPE: %s\n"
5764 "%sMemoryDenyWriteExecute: %s\n"
5765 "%sRestrictRealtime: %s\n"
5766 "%sRestrictSUIDSGID: %s\n"
5767 "%sKeyringMode: %s\n"
5768 "%sProtectHostname: %s\n"
5769 "%sProtectProc: %s\n"
5770 "%sProcSubset: %s\n",
5772 prefix
, empty_to_root(c
->working_directory
),
5773 prefix
, empty_to_root(c
->root_directory
),
5774 prefix
, yes_no(c
->non_blocking
),
5775 prefix
, yes_no(c
->private_tmp
),
5776 prefix
, yes_no(c
->private_devices
),
5777 prefix
, yes_no(c
->protect_kernel_tunables
),
5778 prefix
, yes_no(c
->protect_kernel_modules
),
5779 prefix
, yes_no(c
->protect_kernel_logs
),
5780 prefix
, yes_no(c
->protect_clock
),
5781 prefix
, yes_no(c
->protect_control_groups
),
5782 prefix
, yes_no(c
->private_network
),
5783 prefix
, yes_no(c
->private_users
),
5784 prefix
, protect_home_to_string(c
->protect_home
),
5785 prefix
, protect_system_to_string(c
->protect_system
),
5786 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
5787 prefix
, yes_no(c
->ignore_sigpipe
),
5788 prefix
, yes_no(c
->memory_deny_write_execute
),
5789 prefix
, yes_no(c
->restrict_realtime
),
5790 prefix
, yes_no(c
->restrict_suid_sgid
),
5791 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5792 prefix
, yes_no(c
->protect_hostname
),
5793 prefix
, protect_proc_to_string(c
->protect_proc
),
5794 prefix
, proc_subset_to_string(c
->proc_subset
));
5797 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5799 if (c
->root_image_options
) {
5800 fprintf(f
, "%sRootImageOptions:", prefix
);
5801 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5802 if (!isempty(o
->options
))
5803 fprintf(f
, " %s:%s",
5804 partition_designator_to_string(o
->partition_designator
),
5810 _cleanup_free_
char *encoded
= NULL
;
5811 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5813 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5816 if (c
->root_hash_path
)
5817 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5819 if (c
->root_hash_sig
) {
5820 _cleanup_free_
char *encoded
= NULL
;
5822 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5824 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5827 if (c
->root_hash_sig_path
)
5828 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5831 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5833 STRV_FOREACH(e
, c
->environment
)
5834 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5836 STRV_FOREACH(e
, c
->environment_files
)
5837 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5839 STRV_FOREACH(e
, c
->pass_environment
)
5840 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5842 STRV_FOREACH(e
, c
->unset_environment
)
5843 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5845 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5847 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5848 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5850 for (size_t i
= 0; i
< c
->directories
[dt
].n_items
; i
++) {
5851 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].items
[i
].path
);
5853 STRV_FOREACH(d
, c
->directories
[dt
].items
[i
].symlinks
)
5854 fprintf(f
, "%s%s: %s:%s\n", prefix
, exec_directory_type_symlink_to_string(dt
), c
->directories
[dt
].items
[i
].path
, *d
);
5858 fprintf(f
, "%sTimeoutCleanSec: %s\n", prefix
, FORMAT_TIMESPAN(c
->timeout_clean_usec
, USEC_PER_SEC
));
5861 fprintf(f
, "%sNice: %i\n", prefix
, c
->nice
);
5863 if (c
->oom_score_adjust_set
)
5864 fprintf(f
, "%sOOMScoreAdjust: %i\n", prefix
, c
->oom_score_adjust
);
5866 if (c
->coredump_filter_set
)
5867 fprintf(f
, "%sCoredumpFilter: 0x%"PRIx64
"\n", prefix
, c
->coredump_filter
);
5869 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5871 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5872 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5873 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5874 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5877 if (c
->ioprio_set
) {
5878 _cleanup_free_
char *class_str
= NULL
;
5880 r
= ioprio_class_to_string_alloc(ioprio_prio_class(c
->ioprio
), &class_str
);
5882 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5884 fprintf(f
, "%sIOPriority: %d\n", prefix
, ioprio_prio_data(c
->ioprio
));
5887 if (c
->cpu_sched_set
) {
5888 _cleanup_free_
char *policy_str
= NULL
;
5890 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5892 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5895 "%sCPUSchedulingPriority: %i\n"
5896 "%sCPUSchedulingResetOnFork: %s\n",
5897 prefix
, c
->cpu_sched_priority
,
5898 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5901 if (c
->cpu_set
.set
) {
5902 _cleanup_free_
char *affinity
= NULL
;
5904 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5905 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5908 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5909 _cleanup_free_
char *nodes
= NULL
;
5911 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5912 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5913 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5916 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5917 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5920 "%sStandardInput: %s\n"
5921 "%sStandardOutput: %s\n"
5922 "%sStandardError: %s\n",
5923 prefix
, exec_input_to_string(c
->std_input
),
5924 prefix
, exec_output_to_string(c
->std_output
),
5925 prefix
, exec_output_to_string(c
->std_error
));
5927 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5928 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5929 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5930 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5931 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5932 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5934 if (c
->std_input
== EXEC_INPUT_FILE
)
5935 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5936 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5937 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5938 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5939 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5940 if (c
->std_output
== EXEC_OUTPUT_FILE_TRUNCATE
)
5941 fprintf(f
, "%sStandardOutputFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5942 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5943 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5944 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5945 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5946 if (c
->std_error
== EXEC_OUTPUT_FILE_TRUNCATE
)
5947 fprintf(f
, "%sStandardErrorFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5953 "%sTTYVHangup: %s\n"
5954 "%sTTYVTDisallocate: %s\n"
5956 "%sTTYColumns: %u\n",
5957 prefix
, c
->tty_path
,
5958 prefix
, yes_no(c
->tty_reset
),
5959 prefix
, yes_no(c
->tty_vhangup
),
5960 prefix
, yes_no(c
->tty_vt_disallocate
),
5961 prefix
, c
->tty_rows
,
5962 prefix
, c
->tty_cols
);
5964 if (IN_SET(c
->std_output
,
5966 EXEC_OUTPUT_JOURNAL
,
5967 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5968 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5969 IN_SET(c
->std_error
,
5971 EXEC_OUTPUT_JOURNAL
,
5972 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5973 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5975 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5977 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5979 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5981 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5983 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5986 if (c
->log_level_max
>= 0) {
5987 _cleanup_free_
char *t
= NULL
;
5989 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5991 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5994 if (c
->log_ratelimit_interval_usec
> 0)
5996 "%sLogRateLimitIntervalSec: %s\n",
5997 prefix
, FORMAT_TIMESPAN(c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
5999 if (c
->log_ratelimit_burst
> 0)
6000 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
6002 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
6003 fprintf(f
, "%sLogExtraFields: ", prefix
);
6004 fwrite(c
->log_extra_fields
[j
].iov_base
,
6005 1, c
->log_extra_fields
[j
].iov_len
,
6010 if (c
->log_namespace
)
6011 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
6013 if (c
->secure_bits
) {
6014 _cleanup_free_
char *str
= NULL
;
6016 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
6018 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
6021 if (c
->capability_bounding_set
!= CAP_ALL
) {
6022 _cleanup_free_
char *str
= NULL
;
6024 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
6026 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
6029 if (c
->capability_ambient_set
!= 0) {
6030 _cleanup_free_
char *str
= NULL
;
6032 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
6034 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
6038 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
6040 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
6042 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
6044 strv_dump(f
, prefix
, "SupplementaryGroups", c
->supplementary_groups
);
6047 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
6049 strv_dump(f
, prefix
, "ReadWritePaths", c
->read_write_paths
);
6050 strv_dump(f
, prefix
, "ReadOnlyPaths", c
->read_only_paths
);
6051 strv_dump(f
, prefix
, "InaccessiblePaths", c
->inaccessible_paths
);
6052 strv_dump(f
, prefix
, "ExecPaths", c
->exec_paths
);
6053 strv_dump(f
, prefix
, "NoExecPaths", c
->no_exec_paths
);
6054 strv_dump(f
, prefix
, "ExecSearchPath", c
->exec_search_path
);
6056 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
6057 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
6058 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
6059 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
6060 c
->bind_mounts
[i
].source
,
6061 c
->bind_mounts
[i
].destination
,
6062 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
6064 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
6065 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
6067 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
6069 isempty(t
->options
) ? "" : ":",
6070 strempty(t
->options
));
6075 "%sUtmpIdentifier: %s\n",
6076 prefix
, c
->utmp_id
);
6078 if (c
->selinux_context
)
6080 "%sSELinuxContext: %s%s\n",
6081 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
6083 if (c
->apparmor_profile
)
6085 "%sAppArmorProfile: %s%s\n",
6086 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
6088 if (c
->smack_process_label
)
6090 "%sSmackProcessLabel: %s%s\n",
6091 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
6093 if (c
->personality
!= PERSONALITY_INVALID
)
6095 "%sPersonality: %s\n",
6096 prefix
, strna(personality_to_string(c
->personality
)));
6099 "%sLockPersonality: %s\n",
6100 prefix
, yes_no(c
->lock_personality
));
6102 if (c
->syscall_filter
) {
6104 "%sSystemCallFilter: ",
6107 if (!c
->syscall_allow_list
)
6113 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
6114 _cleanup_free_
char *name
= NULL
;
6115 const char *errno_name
= NULL
;
6116 int num
= PTR_TO_INT(val
);
6123 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
6124 fputs(strna(name
), f
);
6127 errno_name
= seccomp_errno_or_action_to_string(num
);
6129 fprintf(f
, ":%s", errno_name
);
6131 fprintf(f
, ":%d", num
);
6139 if (c
->syscall_archs
) {
6141 "%sSystemCallArchitectures:",
6146 SET_FOREACH(id
, c
->syscall_archs
)
6147 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
6152 if (exec_context_restrict_namespaces_set(c
)) {
6153 _cleanup_free_
char *s
= NULL
;
6155 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
6157 fprintf(f
, "%sRestrictNamespaces: %s\n",
6162 if (exec_context_restrict_filesystems_set(c
)) {
6164 SET_FOREACH(fs
, c
->restrict_filesystems
)
6165 fprintf(f
, "%sRestrictFileSystems: %s\n", prefix
, fs
);
6169 if (c
->network_namespace_path
)
6171 "%sNetworkNamespacePath: %s\n",
6172 prefix
, c
->network_namespace_path
);
6174 if (c
->syscall_errno
> 0) {
6175 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
6178 const char *errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
6180 fputs(errno_name
, f
);
6182 fprintf(f
, "%d", c
->syscall_errno
);
6187 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
6188 fprintf(f
, "%sMountImages: %s%s:%s", prefix
,
6189 c
->mount_images
[i
].ignore_enoent
? "-": "",
6190 c
->mount_images
[i
].source
,
6191 c
->mount_images
[i
].destination
);
6192 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
6193 fprintf(f
, ":%s:%s",
6194 partition_designator_to_string(o
->partition_designator
),
6195 strempty(o
->options
));
6199 for (size_t i
= 0; i
< c
->n_extension_images
; i
++) {
6200 fprintf(f
, "%sExtensionImages: %s%s", prefix
,
6201 c
->extension_images
[i
].ignore_enoent
? "-": "",
6202 c
->extension_images
[i
].source
);
6203 LIST_FOREACH(mount_options
, o
, c
->extension_images
[i
].mount_options
)
6204 fprintf(f
, ":%s:%s",
6205 partition_designator_to_string(o
->partition_designator
),
6206 strempty(o
->options
));
6210 strv_dump(f
, prefix
, "ExtensionDirectories", c
->extension_directories
);
6213 bool exec_context_maintains_privileges(const ExecContext
*c
) {
6216 /* Returns true if the process forked off would run under
6217 * an unchanged UID or as root. */
6222 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
6228 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
6236 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
6238 return IOPRIO_DEFAULT_CLASS_AND_PRIO
;
6240 return ioprio_normalize(p
);
6243 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
6246 /* Explicit setting wins */
6247 if (c
->mount_apivfs_set
)
6248 return c
->mount_apivfs
;
6250 /* Default to "yes" if root directory or image are specified */
6251 if (exec_context_with_rootfs(c
))
6257 void exec_context_free_log_extra_fields(ExecContext
*c
) {
6260 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
6261 free(c
->log_extra_fields
[l
].iov_base
);
6262 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
6263 c
->n_log_extra_fields
= 0;
6266 void exec_context_revert_tty(ExecContext
*c
) {
6267 _cleanup_close_
int fd
= -1;
6274 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
6275 exec_context_tty_reset(c
, NULL
);
6277 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
6278 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
6279 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
6280 if (!exec_context_may_touch_tty(c
))
6283 path
= exec_context_tty_path(c
);
6287 fd
= open(path
, O_PATH
|O_CLOEXEC
);
6289 return (void) log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
6290 "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m",
6293 if (fstat(fd
, &st
) < 0)
6294 return (void) log_warning_errno(errno
, "Failed to stat TTY '%s', ignoring: %m", path
);
6296 /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check
6297 * if things are a character device, since a proper check either means we'd have to open the TTY and
6298 * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects
6299 * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother
6300 * with this at all? → https://github.com/systemd/systemd/issues/19213 */
6301 if (!S_ISCHR(st
.st_mode
))
6302 return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path
);
6304 r
= fchmod_and_chown(fd
, TTY_MODE
, 0, TTY_GID
);
6306 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
6309 int exec_context_get_clean_directories(
6315 _cleanup_strv_free_
char **l
= NULL
;
6322 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
6323 if (!FLAGS_SET(mask
, 1U << t
))
6329 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
6332 j
= path_join(prefix
[t
], c
->directories
[t
].items
[i
].path
);
6336 r
= strv_consume(&l
, j
);
6340 /* Also remove private directories unconditionally. */
6341 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
6342 j
= path_join(prefix
[t
], "private", c
->directories
[t
].items
[i
].path
);
6346 r
= strv_consume(&l
, j
);
6351 STRV_FOREACH(symlink
, c
->directories
[t
].items
[i
].symlinks
) {
6352 j
= path_join(prefix
[t
], *symlink
);
6356 r
= strv_consume(&l
, j
);
6367 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
6368 ExecCleanMask mask
= 0;
6373 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
6374 if (c
->directories
[t
].n_items
> 0)
6381 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
6388 dual_timestamp_get(&s
->start_timestamp
);
6391 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
6399 dual_timestamp_get(&s
->exit_timestamp
);
6404 if (context
&& context
->utmp_id
)
6405 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
6408 void exec_status_reset(ExecStatus
*s
) {
6411 *s
= (ExecStatus
) {};
6414 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
6421 prefix
= strempty(prefix
);
6424 "%sPID: "PID_FMT
"\n",
6427 if (dual_timestamp_is_set(&s
->start_timestamp
))
6429 "%sStart Timestamp: %s\n",
6430 prefix
, FORMAT_TIMESTAMP(s
->start_timestamp
.realtime
));
6432 if (dual_timestamp_is_set(&s
->exit_timestamp
))
6434 "%sExit Timestamp: %s\n"
6436 "%sExit Status: %i\n",
6437 prefix
, FORMAT_TIMESTAMP(s
->exit_timestamp
.realtime
),
6438 prefix
, sigchld_code_to_string(s
->code
),
6442 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6443 _cleanup_free_
char *cmd
= NULL
;
6444 const char *prefix2
;
6449 prefix
= strempty(prefix
);
6450 prefix2
= strjoina(prefix
, "\t");
6452 cmd
= quote_command_line(c
->argv
, SHELL_ESCAPE_EMPTY
);
6455 "%sCommand Line: %s\n",
6456 prefix
, strnull(cmd
));
6458 exec_status_dump(&c
->exec_status
, f
, prefix2
);
6461 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6464 prefix
= strempty(prefix
);
6466 LIST_FOREACH(command
, i
, c
)
6467 exec_command_dump(i
, f
, prefix
);
6470 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
6477 /* It's kind of important, that we keep the order here */
6478 LIST_FIND_TAIL(command
, *l
, end
);
6479 LIST_INSERT_AFTER(command
, *l
, end
, e
);
6484 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
6492 l
= strv_new_ap(path
, ap
);
6504 free_and_replace(c
->path
, p
);
6506 return strv_free_and_replace(c
->argv
, l
);
6509 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
6510 _cleanup_strv_free_
char **l
= NULL
;
6518 l
= strv_new_ap(path
, ap
);
6524 r
= strv_extend_strv(&c
->argv
, l
, false);
6531 static void *remove_tmpdir_thread(void *p
) {
6532 _cleanup_free_
char *path
= p
;
6534 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
6538 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
6545 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
6547 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
6549 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6550 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
6552 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
6554 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
6559 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6560 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
6562 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
6564 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
6566 rt
->var_tmp_dir
= NULL
;
6569 rt
->id
= mfree(rt
->id
);
6570 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
6571 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
6572 safe_close_pair(rt
->netns_storage_socket
);
6573 safe_close_pair(rt
->ipcns_storage_socket
);
6577 static void exec_runtime_freep(ExecRuntime
**rt
) {
6578 (void) exec_runtime_free(*rt
, false);
6581 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
6582 _cleanup_free_
char *id_copy
= NULL
;
6587 id_copy
= strdup(id
);
6591 n
= new(ExecRuntime
, 1);
6595 *n
= (ExecRuntime
) {
6596 .id
= TAKE_PTR(id_copy
),
6597 .netns_storage_socket
= { -1, -1 },
6598 .ipcns_storage_socket
= { -1, -1 },
6605 static int exec_runtime_add(
6610 int netns_storage_socket
[2],
6611 int ipcns_storage_socket
[2],
6612 ExecRuntime
**ret
) {
6614 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6620 /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */
6622 r
= exec_runtime_allocate(&rt
, id
);
6626 r
= hashmap_ensure_put(&m
->exec_runtime_by_id
, &string_hash_ops
, rt
->id
, rt
);
6630 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6631 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6632 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6634 if (netns_storage_socket
) {
6635 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6636 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6639 if (ipcns_storage_socket
) {
6640 rt
->ipcns_storage_socket
[0] = TAKE_FD(ipcns_storage_socket
[0]);
6641 rt
->ipcns_storage_socket
[1] = TAKE_FD(ipcns_storage_socket
[1]);
6648 /* do not remove created ExecRuntime object when the operation succeeds. */
6653 static int exec_runtime_make(
6655 const ExecContext
*c
,
6657 ExecRuntime
**ret
) {
6659 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6660 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 }, ipcns_storage_socket
[2] = { -1, -1 };
6667 /* It is not necessary to create ExecRuntime object. */
6668 if (!c
->private_network
&& !c
->private_ipc
&& !c
->private_tmp
&& !c
->network_namespace_path
) {
6673 if (c
->private_tmp
&&
6674 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6675 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6676 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6677 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6682 if (c
->private_network
|| c
->network_namespace_path
) {
6683 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6687 if (c
->private_ipc
|| c
->ipc_namespace_path
) {
6688 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, ipcns_storage_socket
) < 0)
6692 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ipcns_storage_socket
, ret
);
6699 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6707 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6709 /* We already have an ExecRuntime object, let's increase the ref count and reuse it */
6717 /* If not found, then create a new object. */
6718 r
= exec_runtime_make(m
, c
, id
, &rt
);
6722 /* When r == 0, it is not necessary to create ExecRuntime object. */
6728 /* increment reference counter. */
6734 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6738 assert(rt
->n_ref
> 0);
6744 return exec_runtime_free(rt
, destroy
);
6747 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6754 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6755 fprintf(f
, "exec-runtime=%s", rt
->id
);
6758 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6760 if (rt
->var_tmp_dir
)
6761 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6763 if (rt
->netns_storage_socket
[0] >= 0) {
6766 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6770 fprintf(f
, " netns-socket-0=%i", copy
);
6773 if (rt
->netns_storage_socket
[1] >= 0) {
6776 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6780 fprintf(f
, " netns-socket-1=%i", copy
);
6783 if (rt
->ipcns_storage_socket
[0] >= 0) {
6786 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[0]);
6790 fprintf(f
, " ipcns-socket-0=%i", copy
);
6793 if (rt
->ipcns_storage_socket
[1] >= 0) {
6796 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[1]);
6800 fprintf(f
, " ipcns-socket-1=%i", copy
);
6809 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6810 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6814 /* This is for the migration from old (v237 or earlier) deserialization text.
6815 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6816 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6817 * so or not from the serialized text, then we always creates a new object owned by this. */
6823 /* Manager manages ExecRuntime objects by the unit id.
6824 * So, we omit the serialized text when the unit does not have id (yet?)... */
6825 if (isempty(u
->id
)) {
6826 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6830 if (hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
) < 0)
6833 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6835 if (exec_runtime_allocate(&rt_create
, u
->id
) < 0)
6841 if (streq(key
, "tmp-dir")) {
6842 if (free_and_strdup_warn(&rt
->tmp_dir
, value
) < 0)
6845 } else if (streq(key
, "var-tmp-dir")) {
6846 if (free_and_strdup_warn(&rt
->var_tmp_dir
, value
) < 0)
6849 } else if (streq(key
, "netns-socket-0")) {
6852 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6853 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6857 safe_close(rt
->netns_storage_socket
[0]);
6858 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6860 } else if (streq(key
, "netns-socket-1")) {
6863 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6864 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6868 safe_close(rt
->netns_storage_socket
[1]);
6869 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6874 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6876 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6878 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6882 rt_create
->manager
= u
->manager
;
6885 TAKE_PTR(rt_create
);
6891 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6892 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6894 int r
, netns_fdpair
[] = {-1, -1}, ipcns_fdpair
[] = {-1, -1};
6895 const char *p
, *v
= ASSERT_PTR(value
);
6901 n
= strcspn(v
, " ");
6902 id
= strndupa_safe(v
, n
);
6907 v
= startswith(p
, "tmp-dir=");
6909 n
= strcspn(v
, " ");
6910 tmp_dir
= strndup(v
, n
);
6918 v
= startswith(p
, "var-tmp-dir=");
6920 n
= strcspn(v
, " ");
6921 var_tmp_dir
= strndup(v
, n
);
6929 v
= startswith(p
, "netns-socket-0=");
6933 n
= strcspn(v
, " ");
6934 buf
= strndupa_safe(v
, n
);
6936 r
= safe_atoi(buf
, &netns_fdpair
[0]);
6938 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6939 if (!fdset_contains(fds
, netns_fdpair
[0]))
6940 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6941 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair
[0]);
6942 netns_fdpair
[0] = fdset_remove(fds
, netns_fdpair
[0]);
6948 v
= startswith(p
, "netns-socket-1=");
6952 n
= strcspn(v
, " ");
6953 buf
= strndupa_safe(v
, n
);
6955 r
= safe_atoi(buf
, &netns_fdpair
[1]);
6957 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6958 if (!fdset_contains(fds
, netns_fdpair
[1]))
6959 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6960 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair
[1]);
6961 netns_fdpair
[1] = fdset_remove(fds
, netns_fdpair
[1]);
6967 v
= startswith(p
, "ipcns-socket-0=");
6971 n
= strcspn(v
, " ");
6972 buf
= strndupa_safe(v
, n
);
6974 r
= safe_atoi(buf
, &ipcns_fdpair
[0]);
6976 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf
);
6977 if (!fdset_contains(fds
, ipcns_fdpair
[0]))
6978 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6979 "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair
[0]);
6980 ipcns_fdpair
[0] = fdset_remove(fds
, ipcns_fdpair
[0]);
6986 v
= startswith(p
, "ipcns-socket-1=");
6990 n
= strcspn(v
, " ");
6991 buf
= strndupa_safe(v
, n
);
6993 r
= safe_atoi(buf
, &ipcns_fdpair
[1]);
6995 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf
);
6996 if (!fdset_contains(fds
, ipcns_fdpair
[1]))
6997 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6998 "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair
[1]);
6999 ipcns_fdpair
[1] = fdset_remove(fds
, ipcns_fdpair
[1]);
7003 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_fdpair
, ipcns_fdpair
, NULL
);
7005 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
7009 void exec_runtime_vacuum(Manager
*m
) {
7014 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
7016 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
7020 (void) exec_runtime_free(rt
, false);
7024 void exec_params_clear(ExecParameters
*p
) {
7028 p
->environment
= strv_free(p
->environment
);
7029 p
->fd_names
= strv_free(p
->fd_names
);
7030 p
->fds
= mfree(p
->fds
);
7031 p
->exec_fd
= safe_close(p
->exec_fd
);
7034 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
7043 ExecLoadCredential
*exec_load_credential_free(ExecLoadCredential
*lc
) {
7052 void exec_directory_done(ExecDirectory
*d
) {
7056 for (size_t i
= 0; i
< d
->n_items
; i
++) {
7057 free(d
->items
[i
].path
);
7058 strv_free(d
->items
[i
].symlinks
);
7061 d
->items
= mfree(d
->items
);
7066 static ExecDirectoryItem
*exec_directory_find(ExecDirectory
*d
, const char *path
) {
7070 for (size_t i
= 0; i
< d
->n_items
; i
++)
7071 if (path_equal(d
->items
[i
].path
, path
))
7072 return &d
->items
[i
];
7077 int exec_directory_add(ExecDirectory
*d
, const char *path
, const char *symlink
) {
7078 _cleanup_strv_free_
char **s
= NULL
;
7079 _cleanup_free_
char *p
= NULL
;
7080 ExecDirectoryItem
*existing
;
7086 existing
= exec_directory_find(d
, path
);
7088 r
= strv_extend(&existing
->symlinks
, symlink
);
7092 return 0; /* existing item is updated */
7100 s
= strv_new(symlink
);
7105 if (!GREEDY_REALLOC(d
->items
, d
->n_items
+ 1))
7108 d
->items
[d
->n_items
++] = (ExecDirectoryItem
) {
7109 .path
= TAKE_PTR(p
),
7110 .symlinks
= TAKE_PTR(s
),
7113 return 1; /* new item is added */
7116 static int exec_directory_item_compare_func(const ExecDirectoryItem
*a
, const ExecDirectoryItem
*b
) {
7120 return path_compare(a
->path
, b
->path
);
7123 void exec_directory_sort(ExecDirectory
*d
) {
7126 /* Sort the exec directories to make always parent directories processed at first in
7127 * setup_exec_directory(), e.g., even if StateDirectory=foo/bar foo, we need to create foo at first,
7128 * then foo/bar. Also, set .only_create flag if one of the parent directories is contained in the
7129 * list. See also comments in setup_exec_directory() and issue #24783. */
7131 if (d
->n_items
<= 1)
7134 typesafe_qsort(d
->items
, d
->n_items
, exec_directory_item_compare_func
);
7136 for (size_t i
= 1; i
< d
->n_items
; i
++)
7137 for (size_t j
= 0; j
< i
; j
++)
7138 if (path_startswith(d
->items
[i
].path
, d
->items
[j
].path
)) {
7139 d
->items
[i
].only_create
= true;
7144 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
7145 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_load_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecLoadCredential
, exec_load_credential_free
);
7147 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
7148 [EXEC_INPUT_NULL
] = "null",
7149 [EXEC_INPUT_TTY
] = "tty",
7150 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
7151 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
7152 [EXEC_INPUT_SOCKET
] = "socket",
7153 [EXEC_INPUT_NAMED_FD
] = "fd",
7154 [EXEC_INPUT_DATA
] = "data",
7155 [EXEC_INPUT_FILE
] = "file",
7158 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
7160 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
7161 [EXEC_OUTPUT_INHERIT
] = "inherit",
7162 [EXEC_OUTPUT_NULL
] = "null",
7163 [EXEC_OUTPUT_TTY
] = "tty",
7164 [EXEC_OUTPUT_KMSG
] = "kmsg",
7165 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
7166 [EXEC_OUTPUT_JOURNAL
] = "journal",
7167 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
7168 [EXEC_OUTPUT_SOCKET
] = "socket",
7169 [EXEC_OUTPUT_NAMED_FD
] = "fd",
7170 [EXEC_OUTPUT_FILE
] = "file",
7171 [EXEC_OUTPUT_FILE_APPEND
] = "append",
7172 [EXEC_OUTPUT_FILE_TRUNCATE
] = "truncate",
7175 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
7177 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
7178 [EXEC_UTMP_INIT
] = "init",
7179 [EXEC_UTMP_LOGIN
] = "login",
7180 [EXEC_UTMP_USER
] = "user",
7183 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
7185 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
7186 [EXEC_PRESERVE_NO
] = "no",
7187 [EXEC_PRESERVE_YES
] = "yes",
7188 [EXEC_PRESERVE_RESTART
] = "restart",
7191 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
7193 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
7194 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7195 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
7196 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
7197 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
7198 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
7199 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
7202 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
7204 /* This table maps ExecDirectoryType to the symlink setting it is configured with in the unit */
7205 static const char* const exec_directory_type_symlink_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7206 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectorySymlink",
7207 [EXEC_DIRECTORY_STATE
] = "StateDirectorySymlink",
7208 [EXEC_DIRECTORY_CACHE
] = "CacheDirectorySymlink",
7209 [EXEC_DIRECTORY_LOGS
] = "LogsDirectorySymlink",
7210 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectorySymlink",
7213 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type_symlink
, ExecDirectoryType
);
7215 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
7216 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
7217 * directories, specifically .timer units with their timestamp touch file. */
7218 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7219 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
7220 [EXEC_DIRECTORY_STATE
] = "state",
7221 [EXEC_DIRECTORY_CACHE
] = "cache",
7222 [EXEC_DIRECTORY_LOGS
] = "logs",
7223 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
7226 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
7228 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
7229 * the service payload in. */
7230 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7231 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
7232 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
7233 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
7234 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
7235 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
7238 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
7240 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
7241 [EXEC_KEYRING_INHERIT
] = "inherit",
7242 [EXEC_KEYRING_PRIVATE
] = "private",
7243 [EXEC_KEYRING_SHARED
] = "shared",
7246 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);