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 "constants.h"
52 #include "cpu-set-util.h"
53 #include "creds-util.h"
54 #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
= -EBADF
;
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
= -EBADF
;
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
= -EBADF
, saved_stdin
= -EBADF
, saved_stdout
= -EBADF
;
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
= -EBADF
;
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] = PIPE_EBADF
;
2103 _cleanup_close_
int unshare_ready_fd
= -EBADF
;
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
= -EBADF
;
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
= -EBADF
;
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(read_dfd
>= 0 || read_dfd
== AT_FDCWD
);
2667 assert(write_dfd
>= 0);
2670 if (read_dfd
>= 0) {
2671 /* If a directory fd is specified, then read the file directly from that dir. In this case we
2672 * won't do AF_UNIX stuff (we simply don't want to recursively iterate down a tree of AF_UNIX
2673 * IPC sockets). It's OK if a file vanishes here in the time we enumerate it and intend to
2676 if (!filename_is_valid(path
)) /* safety check */
2682 } else if (path_is_absolute(path
)) {
2683 /* If this is an absolute path, read the data directly from it, and support AF_UNIX
2686 if (!path_is_valid(path
)) /* safety check */
2689 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2691 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2692 * via the source socket address in case we read off an AF_UNIX socket. */
2693 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, id
) < 0)
2699 } else if (credential_name_valid(path
)) {
2700 /* If this is a relative path, take it as credential name relative to the credentials
2701 * directory we received ourselves. We don't support the AF_UNIX stuff in this mode, since we
2702 * are operating on a credential store, i.e. this is guaranteed to be regular files. */
2704 search_path
= credential_search_path(params
, encrypted
);
2713 flags
|= READ_FULL_FILE_UNBASE64
;
2715 maxsz
= encrypted
? CREDENTIAL_ENCRYPTED_SIZE_MAX
: CREDENTIAL_SIZE_MAX
;
2718 STRV_FOREACH(d
, search_path
) {
2719 _cleanup_free_
char *j
= NULL
;
2721 j
= path_join(*d
, path
);
2725 r
= read_full_file_full(
2726 AT_FDCWD
, j
, /* path is absolute, hence pass AT_FDCWD as nop dir fd here */
2736 r
= read_full_file_full(
2746 if (r
== -ENOENT
&& (missing_ok
|| hashmap_contains(context
->set_credentials
, id
))) {
2747 /* Make a missing inherited credential non-fatal, let's just continue. After all apps
2748 * will get clear errors if we don't pass such a missing credential on as they
2749 * themselves will get ENOENT when trying to read them, which should not be much
2750 * worse than when we handle the error here and make it fatal.
2752 * Also, if the source file doesn't exist, but a fallback is set via SetCredentials=
2753 * we are fine, too. */
2754 log_debug_errno(r
, "Couldn't read inherited credential '%s', skipping: %m", path
);
2758 return log_debug_errno(r
, "Failed to read credential '%s': %m", path
);
2761 _cleanup_free_
void *plaintext
= NULL
;
2762 size_t plaintext_size
= 0;
2764 r
= decrypt_credential_and_warn(id
, now(CLOCK_REALTIME
), NULL
, NULL
, data
, size
, &plaintext
, &plaintext_size
);
2768 free_and_replace(data
, plaintext
);
2769 size
= plaintext_size
;
2772 add
= strlen(id
) + size
;
2776 r
= write_credential(write_dfd
, id
, data
, size
, uid
, ownership_ok
);
2778 return log_debug_errno(r
, "Failed to write credential '%s': %m", id
);
2784 struct load_cred_args
{
2785 const ExecContext
*context
;
2786 const ExecParameters
*params
;
2795 static int load_cred_recurse_dir_cb(
2796 RecurseDirEvent event
,
2800 const struct dirent
*de
,
2801 const struct statx
*sx
,
2804 struct load_cred_args
*args
= ASSERT_PTR(userdata
);
2805 _cleanup_free_
char *sub_id
= NULL
;
2808 if (event
!= RECURSE_DIR_ENTRY
)
2809 return RECURSE_DIR_CONTINUE
;
2811 if (!IN_SET(de
->d_type
, DT_REG
, DT_SOCK
))
2812 return RECURSE_DIR_CONTINUE
;
2814 sub_id
= strreplace(path
, "/", "_");
2818 if (!credential_name_valid(sub_id
))
2819 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential would get ID %s, which is not valid, refusing", sub_id
);
2821 if (faccessat(args
->dfd
, sub_id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0) {
2822 log_debug("Skipping credential with duplicated ID %s at %s", sub_id
, path
);
2823 return RECURSE_DIR_CONTINUE
;
2825 if (errno
!= ENOENT
)
2826 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sub_id
);
2828 r
= load_credential(
2843 return RECURSE_DIR_CONTINUE
;
2846 static int acquire_credentials(
2847 const ExecContext
*context
,
2848 const ExecParameters
*params
,
2852 bool ownership_ok
) {
2854 uint64_t left
= CREDENTIALS_TOTAL_SIZE_MAX
;
2855 _cleanup_close_
int dfd
= -EBADF
;
2856 ExecLoadCredential
*lc
;
2857 ExecSetCredential
*sc
;
2863 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
2867 /* First, load credentials off disk (or acquire via AF_UNIX socket) */
2868 HASHMAP_FOREACH(lc
, context
->load_credentials
) {
2869 _cleanup_close_
int sub_fd
= -EBADF
;
2871 /* If this is an absolute path, then try to open it as a directory. If that works, then we'll
2872 * recurse into it. If it is an absolute path but it isn't a directory, then we'll open it as
2873 * a regular file. Finally, if it's a relative path we will use it as a credential name to
2874 * propagate a credential passed to us from further up. */
2876 if (path_is_absolute(lc
->path
)) {
2877 sub_fd
= open(lc
->path
, O_DIRECTORY
|O_CLOEXEC
|O_RDONLY
);
2878 if (sub_fd
< 0 && !IN_SET(errno
,
2879 ENOTDIR
, /* Not a directory */
2880 ENOENT
)) /* Doesn't exist? */
2881 return log_debug_errno(errno
, "Failed to open '%s': %m", lc
->path
);
2885 /* Regular file (incl. a credential passed in from higher up) */
2886 r
= load_credential(
2902 /* path= */ lc
->id
, /* recurse_dir() will suffix the subdir paths from here to the top-level id */
2903 /* statx_mask= */ 0,
2904 /* n_depth_max= */ UINT_MAX
,
2905 RECURSE_DIR_SORT
|RECURSE_DIR_IGNORE_DOT
|RECURSE_DIR_ENSURE_TYPE
,
2906 load_cred_recurse_dir_cb
,
2907 &(struct load_cred_args
) {
2910 .encrypted
= lc
->encrypted
,
2914 .ownership_ok
= ownership_ok
,
2921 /* Second, we add in literally specified credentials. If the credentials already exist, we'll not add
2922 * them, so that they can act as a "default" if the same credential is specified multiple times. */
2923 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
2924 _cleanup_(erase_and_freep
) void *plaintext
= NULL
;
2928 /* Note that we check ahead of time here instead of relying on O_EXCL|O_CREAT later to return
2929 * EEXIST if the credential already exists. That's because the TPM2-based decryption is kinda
2930 * slow and involved, hence it's nice to be able to skip that if the credential already
2932 if (faccessat(dfd
, sc
->id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0)
2934 if (errno
!= ENOENT
)
2935 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sc
->id
);
2937 if (sc
->encrypted
) {
2938 r
= decrypt_credential_and_warn(sc
->id
, now(CLOCK_REALTIME
), NULL
, NULL
, sc
->data
, sc
->size
, &plaintext
, &size
);
2948 add
= strlen(sc
->id
) + size
;
2952 r
= write_credential(dfd
, sc
->id
, data
, size
, uid
, ownership_ok
);
2959 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
2962 /* After we created all keys with the right perms, also make sure the credential store as a whole is
2965 if (uid_is_valid(uid
) && uid
!= getuid()) {
2966 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
2968 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2974 if (fchown(dfd
, uid
, GID_INVALID
) < 0)
2982 static int setup_credentials_internal(
2983 const ExecContext
*context
,
2984 const ExecParameters
*params
,
2986 const char *final
, /* This is where the credential store shall eventually end up at */
2987 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
2988 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
2989 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
2992 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
2993 * if we mounted something; false if we definitely can't mount anything */
3001 if (reuse_workspace
) {
3002 r
= path_is_mount_point(workspace
, NULL
, 0);
3006 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
3008 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
3010 workspace_mounted
= -1; /* ditto */
3012 r
= path_is_mount_point(final
, NULL
, 0);
3016 /* If the final place already has something mounted, we use that. If the workspace also has
3017 * something mounted we assume it's actually the same mount (but with MS_RDONLY
3019 final_mounted
= true;
3021 if (workspace_mounted
< 0) {
3022 /* If the final place is mounted, but the workspace we isn't, then let's bind mount
3023 * the final version to the workspace, and make it writable, so that we can make
3026 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
3030 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3034 workspace_mounted
= true;
3037 final_mounted
= false;
3039 if (workspace_mounted
< 0) {
3040 /* Nothing is mounted on the workspace yet, let's try to mount something now */
3041 for (int try = 0;; try++) {
3044 /* Try "ramfs" first, since it's not swap backed */
3045 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
3047 workspace_mounted
= true;
3051 } else if (try == 1) {
3052 _cleanup_free_
char *opts
= NULL
;
3054 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%zu", (size_t) CREDENTIALS_TOTAL_SIZE_MAX
) < 0)
3057 /* Fall back to "tmpfs" otherwise */
3058 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
3060 workspace_mounted
= true;
3065 /* 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. */
3066 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
3068 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
3071 if (must_mount
) /* If we it's not OK to use the plain directory
3072 * fallback, propagate all errors too */
3075 /* If we lack privileges to bind mount stuff, then let's gracefully
3076 * proceed for compat with container envs, and just use the final dir
3079 workspace_mounted
= false;
3083 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
3084 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3088 workspace_mounted
= true;
3094 assert(!must_mount
|| workspace_mounted
> 0);
3095 where
= workspace_mounted
? workspace
: final
;
3097 (void) label_fix_full(AT_FDCWD
, where
, final
, 0);
3099 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
3103 if (workspace_mounted
) {
3104 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
3105 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
3109 /* And mount it to the final place, read-only */
3111 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
3113 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
3117 _cleanup_free_
char *parent
= NULL
;
3119 /* If we do not have our own mount put used the plain directory fallback, then we need to
3120 * open access to the top-level credential directory and the per-service directory now */
3122 r
= path_extract_directory(final
, &parent
);
3125 if (chmod(parent
, 0755) < 0)
3132 static int setup_credentials(
3133 const ExecContext
*context
,
3134 const ExecParameters
*params
,
3138 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
3144 if (!exec_context_has_credentials(context
))
3147 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
3150 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
3151 * and the subdir we mount over with a read-only file system readable by the service's user */
3152 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
3156 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
3157 if (r
< 0 && r
!= -EEXIST
)
3160 p
= path_join(q
, unit
);
3164 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
3165 if (r
< 0 && r
!= -EEXIST
)
3168 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
3170 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
3172 /* If this is not a privilege or support issue then propagate the error */
3173 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
3176 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
3177 * it into place, so that users can't access half-initialized credential stores. */
3178 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
3182 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
3183 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
3184 * after it is fully set up */
3185 u
= path_join(t
, unit
);
3189 FOREACH_STRING(i
, t
, u
) {
3190 r
= mkdir_label(i
, 0700);
3191 if (r
< 0 && r
!= -EEXIST
)
3195 r
= setup_credentials_internal(
3199 p
, /* final mount point */
3200 u
, /* temporary workspace to overmount */
3201 true, /* reuse the workspace if it is already a mount */
3202 false, /* it's OK to fall back to a plain directory if we can't mount anything */
3205 (void) rmdir(u
); /* remove the workspace again if we can. */
3210 } else if (r
== 0) {
3212 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
3213 * we can use the same directory for all cases, after turning off propagation. Question
3214 * though is: where do we turn off propagation exactly, and where do we place the workspace
3215 * directory? We need some place that is guaranteed to be a mount point in the host, and
3216 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
3217 * since we ultimately want to move the resulting file system there, i.e. we need propagation
3218 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
3219 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
3220 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
3221 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
3222 * propagation on the former, and then overmount the latter.
3224 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
3225 * for this purpose, but there are few other candidates that work equally well for us, and
3226 * given that the we do this in a privately namespaced short-lived single-threaded process
3227 * that no one else sees this should be OK to do. */
3229 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
3233 r
= setup_credentials_internal(
3237 p
, /* final mount point */
3238 "/dev/shm", /* temporary workspace to overmount */
3239 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
3240 true, /* insist that something is mounted, do not allow fallback to plain directory */
3245 _exit(EXIT_SUCCESS
);
3248 _exit(EXIT_FAILURE
);
3255 static int setup_smack(
3256 const Manager
*manager
,
3257 const ExecContext
*context
,
3258 int executable_fd
) {
3262 assert(executable_fd
>= 0);
3264 if (context
->smack_process_label
) {
3265 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
3268 } else if (manager
->default_smack_process_label
) {
3269 _cleanup_free_
char *exec_label
= NULL
;
3271 r
= mac_smack_read_fd(executable_fd
, SMACK_ATTR_EXEC
, &exec_label
);
3272 if (r
< 0 && !ERRNO_IS_XATTR_ABSENT(r
))
3275 r
= mac_smack_apply_pid(0, exec_label
? : manager
->default_smack_process_label
);
3284 static int compile_bind_mounts(
3285 const ExecContext
*context
,
3286 const ExecParameters
*params
,
3287 BindMount
**ret_bind_mounts
,
3288 size_t *ret_n_bind_mounts
,
3289 char ***ret_empty_directories
) {
3291 _cleanup_strv_free_
char **empty_directories
= NULL
;
3292 BindMount
*bind_mounts
;
3298 assert(ret_bind_mounts
);
3299 assert(ret_n_bind_mounts
);
3300 assert(ret_empty_directories
);
3302 n
= context
->n_bind_mounts
;
3303 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3304 if (!params
->prefix
[t
])
3307 for (size_t i
= 0; i
< context
->directories
[t
].n_items
; i
++)
3308 n
+= !context
->directories
[t
].items
[i
].only_create
;
3312 *ret_bind_mounts
= NULL
;
3313 *ret_n_bind_mounts
= 0;
3314 *ret_empty_directories
= NULL
;
3318 bind_mounts
= new(BindMount
, n
);
3322 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
3323 BindMount
*item
= context
->bind_mounts
+ i
;
3326 s
= strdup(item
->source
);
3332 d
= strdup(item
->destination
);
3339 bind_mounts
[h
++] = (BindMount
) {
3342 .read_only
= item
->read_only
,
3343 .recursive
= item
->recursive
,
3344 .ignore_enoent
= item
->ignore_enoent
,
3348 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3349 if (!params
->prefix
[t
])
3352 if (context
->directories
[t
].n_items
== 0)
3355 if (exec_directory_is_private(context
, t
) &&
3356 !exec_context_with_rootfs(context
)) {
3359 /* So this is for a dynamic user, and we need to make sure the process can access its own
3360 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3361 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3363 private_root
= path_join(params
->prefix
[t
], "private");
3364 if (!private_root
) {
3369 r
= strv_consume(&empty_directories
, private_root
);
3374 for (size_t i
= 0; i
< context
->directories
[t
].n_items
; i
++) {
3377 /* When one of the parent directories is in the list, we cannot create the symlink
3378 * for the child directory. See also the comments in setup_exec_directory(). */
3379 if (context
->directories
[t
].items
[i
].only_create
)
3382 if (exec_directory_is_private(context
, t
))
3383 s
= path_join(params
->prefix
[t
], "private", context
->directories
[t
].items
[i
].path
);
3385 s
= path_join(params
->prefix
[t
], context
->directories
[t
].items
[i
].path
);
3391 if (exec_directory_is_private(context
, t
) &&
3392 exec_context_with_rootfs(context
))
3393 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3394 * directory is not created on the root directory. So, let's bind-mount the directory
3395 * on the 'non-private' place. */
3396 d
= path_join(params
->prefix
[t
], context
->directories
[t
].items
[i
].path
);
3405 bind_mounts
[h
++] = (BindMount
) {
3409 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3411 .ignore_enoent
= false,
3418 *ret_bind_mounts
= bind_mounts
;
3419 *ret_n_bind_mounts
= n
;
3420 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3425 bind_mount_free_many(bind_mounts
, h
);
3429 /* ret_symlinks will contain a list of pairs src:dest that describes
3430 * the symlinks to create later on. For example, the symlinks needed
3431 * to safely give private directories to DynamicUser=1 users. */
3432 static int compile_symlinks(
3433 const ExecContext
*context
,
3434 const ExecParameters
*params
,
3435 char ***ret_symlinks
) {
3437 _cleanup_strv_free_
char **symlinks
= NULL
;
3442 assert(ret_symlinks
);
3444 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3445 for (size_t i
= 0; i
< context
->directories
[dt
].n_items
; i
++) {
3446 _cleanup_free_
char *private_path
= NULL
, *path
= NULL
;
3448 STRV_FOREACH(symlink
, context
->directories
[dt
].items
[i
].symlinks
) {
3449 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
3451 src_abs
= path_join(params
->prefix
[dt
], context
->directories
[dt
].items
[i
].path
);
3452 dst_abs
= path_join(params
->prefix
[dt
], *symlink
);
3453 if (!src_abs
|| !dst_abs
)
3456 r
= strv_consume_pair(&symlinks
, TAKE_PTR(src_abs
), TAKE_PTR(dst_abs
));
3461 if (!exec_directory_is_private(context
, dt
) ||
3462 exec_context_with_rootfs(context
) ||
3463 context
->directories
[dt
].items
[i
].only_create
)
3466 private_path
= path_join(params
->prefix
[dt
], "private", context
->directories
[dt
].items
[i
].path
);
3470 path
= path_join(params
->prefix
[dt
], context
->directories
[dt
].items
[i
].path
);
3474 r
= strv_consume_pair(&symlinks
, TAKE_PTR(private_path
), TAKE_PTR(path
));
3480 *ret_symlinks
= TAKE_PTR(symlinks
);
3485 static bool insist_on_sandboxing(
3486 const ExecContext
*context
,
3487 const char *root_dir
,
3488 const char *root_image
,
3489 const BindMount
*bind_mounts
,
3490 size_t n_bind_mounts
) {
3493 assert(n_bind_mounts
== 0 || bind_mounts
);
3495 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3496 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3497 * rearrange stuff in a way we cannot ignore gracefully. */
3499 if (context
->n_temporary_filesystems
> 0)
3502 if (root_dir
|| root_image
)
3505 if (context
->n_mount_images
> 0)
3508 if (context
->dynamic_user
)
3511 if (context
->n_extension_images
> 0 || !strv_isempty(context
->extension_directories
))
3514 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3516 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3517 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3520 if (context
->log_namespace
)
3526 static int apply_mount_namespace(
3528 ExecCommandFlags command_flags
,
3529 const ExecContext
*context
,
3530 const ExecParameters
*params
,
3531 const ExecRuntime
*runtime
,
3532 char **error_path
) {
3534 _cleanup_strv_free_
char **empty_directories
= NULL
, **symlinks
= NULL
;
3535 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3536 const char *root_dir
= NULL
, *root_image
= NULL
;
3537 _cleanup_free_
char *creds_path
= NULL
, *incoming_dir
= NULL
, *propagate_dir
= NULL
,
3538 *extension_dir
= NULL
;
3539 NamespaceInfo ns_info
;
3540 bool needs_sandboxing
;
3541 BindMount
*bind_mounts
= NULL
;
3542 size_t n_bind_mounts
= 0;
3547 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3548 root_image
= context
->root_image
;
3551 root_dir
= context
->root_directory
;
3554 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3558 /* Symlinks for exec dirs are set up after other mounts, before they are made read-only. */
3559 r
= compile_symlinks(context
, params
, &symlinks
);
3563 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3564 if (needs_sandboxing
) {
3565 /* The runtime struct only contains the parent of the private /tmp,
3566 * which is non-accessible to world users. Inside of it there's a /tmp
3567 * that is sticky, and that's the one we want to use here.
3568 * This does not apply when we are using /run/systemd/empty as fallback. */
3570 if (context
->private_tmp
&& runtime
) {
3571 if (streq_ptr(runtime
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3572 tmp_dir
= runtime
->tmp_dir
;
3573 else if (runtime
->tmp_dir
)
3574 tmp_dir
= strjoina(runtime
->tmp_dir
, "/tmp");
3576 if (streq_ptr(runtime
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3577 var_tmp_dir
= runtime
->var_tmp_dir
;
3578 else if (runtime
->var_tmp_dir
)
3579 var_tmp_dir
= strjoina(runtime
->var_tmp_dir
, "/tmp");
3582 ns_info
= (NamespaceInfo
) {
3583 .ignore_protect_paths
= false,
3584 .private_dev
= context
->private_devices
,
3585 .protect_control_groups
= context
->protect_control_groups
,
3586 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3587 .protect_kernel_modules
= context
->protect_kernel_modules
,
3588 .protect_kernel_logs
= context
->protect_kernel_logs
,
3589 .protect_hostname
= context
->protect_hostname
,
3590 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3591 .private_mounts
= context
->private_mounts
,
3592 .protect_home
= context
->protect_home
,
3593 .protect_system
= context
->protect_system
,
3594 .protect_proc
= context
->protect_proc
,
3595 .proc_subset
= context
->proc_subset
,
3596 .private_ipc
= context
->private_ipc
|| context
->ipc_namespace_path
,
3597 /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
3598 .mount_nosuid
= context
->no_new_privileges
&& !mac_selinux_use(),
3600 } else if (!context
->dynamic_user
&& root_dir
)
3602 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3603 * sandbox info, otherwise enforce it, don't ignore protected paths and
3604 * fail if we are enable to apply the sandbox inside the mount namespace.
3606 ns_info
= (NamespaceInfo
) {
3607 .ignore_protect_paths
= true,
3610 ns_info
= (NamespaceInfo
) {};
3612 if (context
->mount_flags
== MS_SHARED
)
3613 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3615 if (exec_context_has_credentials(context
) &&
3616 params
->prefix
[EXEC_DIRECTORY_RUNTIME
] &&
3617 FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
3618 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3625 if (MANAGER_IS_SYSTEM(u
->manager
)) {
3626 propagate_dir
= path_join("/run/systemd/propagate/", u
->id
);
3627 if (!propagate_dir
) {
3632 incoming_dir
= strdup("/run/systemd/incoming");
3633 if (!incoming_dir
) {
3638 extension_dir
= strdup("/run/systemd/unit-extensions");
3639 if (!extension_dir
) {
3644 if (asprintf(&extension_dir
, "/run/user/" UID_FMT
"/systemd/unit-extensions", geteuid()) < 0) {
3649 r
= setup_namespace(root_dir
, root_image
, context
->root_image_options
,
3650 &ns_info
, context
->read_write_paths
,
3651 needs_sandboxing
? context
->read_only_paths
: NULL
,
3652 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3653 needs_sandboxing
? context
->exec_paths
: NULL
,
3654 needs_sandboxing
? context
->no_exec_paths
: NULL
,
3659 context
->temporary_filesystems
,
3660 context
->n_temporary_filesystems
,
3661 context
->mount_images
,
3662 context
->n_mount_images
,
3666 context
->log_namespace
,
3667 context
->mount_flags
,
3668 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3669 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3670 context
->root_verity
,
3671 context
->extension_images
,
3672 context
->n_extension_images
,
3673 context
->extension_directories
,
3677 root_dir
|| root_image
? params
->notify_socket
: NULL
,
3680 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3681 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3682 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3683 * completely different execution environment. */
3685 if (insist_on_sandboxing(
3687 root_dir
, root_image
,
3690 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3691 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3692 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
3696 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3702 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
3706 static int apply_working_directory(
3707 const ExecContext
*context
,
3708 const ExecParameters
*params
,
3715 assert(exit_status
);
3717 if (context
->working_directory_home
) {
3720 *exit_status
= EXIT_CHDIR
;
3727 wd
= empty_to_root(context
->working_directory
);
3729 if (params
->flags
& EXEC_APPLY_CHROOT
)
3732 d
= prefix_roota(context
->root_directory
, wd
);
3734 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3735 *exit_status
= EXIT_CHDIR
;
3742 static int apply_root_directory(
3743 const ExecContext
*context
,
3744 const ExecParameters
*params
,
3745 const bool needs_mount_ns
,
3749 assert(exit_status
);
3751 if (params
->flags
& EXEC_APPLY_CHROOT
)
3752 if (!needs_mount_ns
&& context
->root_directory
)
3753 if (chroot(context
->root_directory
) < 0) {
3754 *exit_status
= EXIT_CHROOT
;
3761 static int setup_keyring(
3763 const ExecContext
*context
,
3764 const ExecParameters
*p
,
3765 uid_t uid
, gid_t gid
) {
3767 key_serial_t keyring
;
3776 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3777 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3778 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3779 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3780 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3781 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3783 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3786 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3787 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3788 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3789 * & group is just as nasty as acquiring a reference to the user keyring. */
3791 saved_uid
= getuid();
3792 saved_gid
= getgid();
3794 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3795 if (setregid(gid
, -1) < 0)
3796 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3799 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3800 if (setreuid(uid
, -1) < 0) {
3801 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3806 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3807 if (keyring
== -1) {
3808 if (errno
== ENOSYS
)
3809 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3810 else if (ERRNO_IS_PRIVILEGE(errno
))
3811 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3812 else if (errno
== EDQUOT
)
3813 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3815 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3820 /* When requested link the user keyring into the session keyring. */
3821 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3823 if (keyctl(KEYCTL_LINK
,
3824 KEY_SPEC_USER_KEYRING
,
3825 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3826 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3831 /* Restore uid/gid back */
3832 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3833 if (setreuid(saved_uid
, -1) < 0) {
3834 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3839 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3840 if (setregid(saved_gid
, -1) < 0)
3841 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3844 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3845 if (!sd_id128_is_null(u
->invocation_id
)) {
3848 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3850 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3852 if (keyctl(KEYCTL_SETPERM
, key
,
3853 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3854 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3855 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3860 /* Revert back uid & gid for the last time, and exit */
3861 /* no extra logging, as only the first already reported error matters */
3862 if (getuid() != saved_uid
)
3863 (void) setreuid(saved_uid
, -1);
3865 if (getgid() != saved_gid
)
3866 (void) setregid(saved_gid
, -1);
3871 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3877 array
[(*n
)++] = pair
[0];
3879 array
[(*n
)++] = pair
[1];
3882 static int close_remaining_fds(
3883 const ExecParameters
*params
,
3884 const ExecRuntime
*runtime
,
3885 const DynamicCreds
*dcreds
,
3888 const int *fds
, size_t n_fds
) {
3890 size_t n_dont_close
= 0;
3891 int dont_close
[n_fds
+ 12];
3895 if (params
->stdin_fd
>= 0)
3896 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3897 if (params
->stdout_fd
>= 0)
3898 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3899 if (params
->stderr_fd
>= 0)
3900 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3903 dont_close
[n_dont_close
++] = socket_fd
;
3905 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3906 n_dont_close
+= n_fds
;
3910 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3911 append_socket_pair(dont_close
, &n_dont_close
, runtime
->ipcns_storage_socket
);
3916 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3918 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3921 if (user_lookup_fd
>= 0)
3922 dont_close
[n_dont_close
++] = user_lookup_fd
;
3924 return close_all_fds(dont_close
, n_dont_close
);
3927 static int send_user_lookup(
3935 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3936 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3939 if (user_lookup_fd
< 0)
3942 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3945 if (writev(user_lookup_fd
,
3947 IOVEC_INIT(&uid
, sizeof(uid
)),
3948 IOVEC_INIT(&gid
, sizeof(gid
)),
3949 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3955 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3962 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3967 if (!c
->working_directory_home
)
3970 r
= get_home_dir(buf
);
3978 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3979 _cleanup_strv_free_
char ** list
= NULL
;
3986 assert(c
->dynamic_user
);
3988 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3989 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3992 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3993 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3999 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
4002 if (exec_directory_is_private(c
, t
))
4003 e
= path_join(p
->prefix
[t
], "private", c
->directories
[t
].items
[i
].path
);
4005 e
= path_join(p
->prefix
[t
], c
->directories
[t
].items
[i
].path
);
4009 r
= strv_consume(&list
, e
);
4015 *ret
= TAKE_PTR(list
);
4020 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
4021 bool using_subcgroup
;
4027 if (!params
->cgroup_path
)
4030 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
4031 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
4032 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
4033 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
4034 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
4035 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
4036 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
4037 * flag, which is only passed for the former statements, not for the latter. */
4039 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
4040 if (using_subcgroup
)
4041 p
= path_join(params
->cgroup_path
, ".control");
4043 p
= strdup(params
->cgroup_path
);
4048 return using_subcgroup
;
4051 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
4052 _cleanup_(cpu_set_reset
) CPUSet s
= {};
4058 if (!c
->numa_policy
.nodes
.set
) {
4059 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
4063 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
4069 return cpu_set_add_all(ret
, &s
);
4072 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
4075 return c
->cpu_affinity_from_numa
;
4078 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
4083 assert(*n_fds
< fds_size
);
4091 if (fd
< 3 + (int) *n_fds
) {
4092 /* Let's move the fd up, so that it's outside of the fd range we will use to store
4093 * the fds we pass to the process (or which are closed only during execve). */
4095 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
4099 close_and_replace(fd
, r
);
4102 *ret_fd
= fds
[*n_fds
] = fd
;
4107 static int exec_child(
4109 const ExecCommand
*command
,
4110 const ExecContext
*context
,
4111 const ExecParameters
*params
,
4112 ExecRuntime
*runtime
,
4113 DynamicCreds
*dcreds
,
4115 const int named_iofds
[static 3],
4117 size_t n_socket_fds
,
4118 size_t n_storage_fds
,
4123 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **joined_exec_search_path
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
4124 int r
, ngids
= 0, exec_fd
;
4125 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
4126 const char *username
= NULL
, *groupname
= NULL
;
4127 _cleanup_free_
char *home_buffer
= NULL
;
4128 const char *home
= NULL
, *shell
= NULL
;
4129 char **final_argv
= NULL
;
4130 dev_t journal_stream_dev
= 0;
4131 ino_t journal_stream_ino
= 0;
4132 bool userns_set_up
= false;
4133 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
4134 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
4135 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
4136 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
4138 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
4139 bool use_selinux
= false;
4142 bool use_smack
= false;
4145 bool use_apparmor
= false;
4147 uid_t saved_uid
= getuid();
4148 gid_t saved_gid
= getgid();
4149 uid_t uid
= UID_INVALID
;
4150 gid_t gid
= GID_INVALID
;
4151 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
4152 n_keep_fds
; /* total number of fds not to close */
4154 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
4155 int ngids_after_pam
= 0;
4161 assert(exit_status
);
4163 /* Explicitly test for CVE-2021-4034 inspired invocations */
4164 assert(command
->path
);
4165 assert(!strv_isempty(command
->argv
));
4167 rename_process_from_path(command
->path
);
4169 /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main
4170 * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially,
4171 * both of which will be demoted to SIG_DFL. */
4172 (void) default_signals(SIGNALS_CRASH_HANDLER
,
4175 if (context
->ignore_sigpipe
)
4176 (void) ignore_signals(SIGPIPE
);
4178 r
= reset_signal_mask();
4180 *exit_status
= EXIT_SIGNAL_MASK
;
4181 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
4184 if (params
->idle_pipe
)
4185 do_idle_pipe_dance(params
->idle_pipe
);
4187 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
4188 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
4189 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
4190 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
4193 log_set_open_when_needed(true);
4195 /* In case anything used libc syslog(), close this here, too */
4198 int keep_fds
[n_fds
+ 3];
4199 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
4202 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
4204 *exit_status
= EXIT_FDS
;
4205 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4209 if (unit
->manager
->restrict_fs
) {
4210 int bpf_map_fd
= lsm_bpf_map_restrict_fs_fd(unit
);
4211 if (bpf_map_fd
< 0) {
4212 *exit_status
= EXIT_FDS
;
4213 return log_unit_error_errno(unit
, bpf_map_fd
, "Failed to get restrict filesystems BPF map fd: %m");
4216 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, bpf_map_fd
, &bpf_map_fd
);
4218 *exit_status
= EXIT_FDS
;
4219 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4224 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
4226 *exit_status
= EXIT_FDS
;
4227 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
4230 if (!context
->same_pgrp
&&
4232 *exit_status
= EXIT_SETSID
;
4233 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
4236 exec_context_tty_reset(context
, params
);
4238 if (unit_shall_confirm_spawn(unit
)) {
4239 _cleanup_free_
char *cmdline
= NULL
;
4241 cmdline
= quote_command_line(command
->argv
, SHELL_ESCAPE_EMPTY
);
4243 *exit_status
= EXIT_MEMORY
;
4247 r
= ask_for_confirmation(context
, params
->confirm_spawn
, unit
, cmdline
);
4248 if (r
!= CONFIRM_EXECUTE
) {
4249 if (r
== CONFIRM_PRETEND_SUCCESS
) {
4250 *exit_status
= EXIT_SUCCESS
;
4253 *exit_status
= EXIT_CONFIRM
;
4254 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
4255 "Execution cancelled by the user");
4259 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
4260 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
4261 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
4262 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
4263 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
4264 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
4265 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
4266 *exit_status
= EXIT_MEMORY
;
4267 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
4270 if (context
->dynamic_user
&& dcreds
) {
4271 _cleanup_strv_free_
char **suggested_paths
= NULL
;
4273 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
4274 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
4275 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
4276 *exit_status
= EXIT_USER
;
4277 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
4280 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
4282 *exit_status
= EXIT_MEMORY
;
4286 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
4288 *exit_status
= EXIT_USER
;
4290 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4291 "Failed to update dynamic user credentials: User or group with specified name already exists.");
4292 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
4295 if (!uid_is_valid(uid
)) {
4296 *exit_status
= EXIT_USER
;
4297 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "UID validation failed for \""UID_FMT
"\"", uid
);
4300 if (!gid_is_valid(gid
)) {
4301 *exit_status
= EXIT_USER
;
4302 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "GID validation failed for \""GID_FMT
"\"", gid
);
4306 username
= dcreds
->user
->name
;
4309 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
4311 *exit_status
= EXIT_USER
;
4312 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
4315 r
= get_fixed_group(context
, &groupname
, &gid
);
4317 *exit_status
= EXIT_GROUP
;
4318 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
4322 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
4323 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
4324 &supplementary_gids
, &ngids
);
4326 *exit_status
= EXIT_GROUP
;
4327 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
4330 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
4332 *exit_status
= EXIT_USER
;
4333 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
4336 user_lookup_fd
= safe_close(user_lookup_fd
);
4338 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
4340 *exit_status
= EXIT_CHDIR
;
4341 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
4344 /* If a socket is connected to STDIN/STDOUT/STDERR, we
4345 * must sure to drop O_NONBLOCK */
4347 (void) fd_nonblock(socket_fd
, false);
4349 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
4350 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
4351 if (params
->cgroup_path
) {
4352 _cleanup_free_
char *p
= NULL
;
4354 r
= exec_parameters_get_cgroup_path(params
, &p
);
4356 *exit_status
= EXIT_CGROUP
;
4357 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
4360 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
4361 if (r
== -EUCLEAN
) {
4362 *exit_status
= EXIT_CGROUP
;
4363 return log_unit_error_errno(unit
, r
, "Failed to attach process to cgroup %s "
4364 "because the cgroup or one of its parents or "
4365 "siblings is in the threaded mode: %m", p
);
4368 *exit_status
= EXIT_CGROUP
;
4369 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
4373 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4374 r
= open_shareable_ns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
, CLONE_NEWNET
);
4376 *exit_status
= EXIT_NETWORK
;
4377 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
4381 if (context
->ipc_namespace_path
&& runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
4382 r
= open_shareable_ns_path(runtime
->ipcns_storage_socket
, context
->ipc_namespace_path
, CLONE_NEWIPC
);
4384 *exit_status
= EXIT_NAMESPACE
;
4385 return log_unit_error_errno(unit
, r
, "Failed to open IPC namespace path %s: %m", context
->ipc_namespace_path
);
4389 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
4391 *exit_status
= EXIT_STDIN
;
4392 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
4395 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
4397 *exit_status
= EXIT_STDOUT
;
4398 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
4401 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
4403 *exit_status
= EXIT_STDERR
;
4404 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
4407 if (context
->oom_score_adjust_set
) {
4408 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
4409 * prohibit write access to this file, and we shouldn't trip up over that. */
4410 r
= set_oom_score_adjust(context
->oom_score_adjust
);
4411 if (ERRNO_IS_PRIVILEGE(r
))
4412 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
4414 *exit_status
= EXIT_OOM_ADJUST
;
4415 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
4419 if (context
->coredump_filter_set
) {
4420 r
= set_coredump_filter(context
->coredump_filter
);
4421 if (ERRNO_IS_PRIVILEGE(r
))
4422 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
4424 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
4427 if (context
->nice_set
) {
4428 r
= setpriority_closest(context
->nice
);
4430 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
4433 if (context
->cpu_sched_set
) {
4434 struct sched_param param
= {
4435 .sched_priority
= context
->cpu_sched_priority
,
4438 r
= sched_setscheduler(0,
4439 context
->cpu_sched_policy
|
4440 (context
->cpu_sched_reset_on_fork
?
4441 SCHED_RESET_ON_FORK
: 0),
4444 *exit_status
= EXIT_SETSCHEDULER
;
4445 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
4449 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
4450 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
4451 const CPUSet
*cpu_set
;
4453 if (context
->cpu_affinity_from_numa
) {
4454 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
4456 *exit_status
= EXIT_CPUAFFINITY
;
4457 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
4460 cpu_set
= &converted_cpu_set
;
4462 cpu_set
= &context
->cpu_set
;
4464 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
4465 *exit_status
= EXIT_CPUAFFINITY
;
4466 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
4470 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
4471 r
= apply_numa_policy(&context
->numa_policy
);
4472 if (r
== -EOPNOTSUPP
)
4473 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
4475 *exit_status
= EXIT_NUMA_POLICY
;
4476 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
4480 if (context
->ioprio_set
)
4481 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
4482 *exit_status
= EXIT_IOPRIO
;
4483 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
4486 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
4487 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
4488 *exit_status
= EXIT_TIMERSLACK
;
4489 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
4492 if (context
->personality
!= PERSONALITY_INVALID
) {
4493 r
= safe_personality(context
->personality
);
4495 *exit_status
= EXIT_PERSONALITY
;
4496 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
4500 if (context
->utmp_id
) {
4501 const char *line
= context
->tty_path
?
4502 (path_startswith(context
->tty_path
, "/dev/") ?: context
->tty_path
) :
4504 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
4506 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
4507 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
4512 if (uid_is_valid(uid
)) {
4513 r
= chown_terminal(STDIN_FILENO
, uid
);
4515 *exit_status
= EXIT_STDIN
;
4516 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
4520 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4521 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4522 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4523 * touch a single hierarchy too. */
4524 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
4525 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
4527 *exit_status
= EXIT_CGROUP
;
4528 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4532 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4534 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4535 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, needs_mount_namespace
, exit_status
);
4537 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4540 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4541 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4543 *exit_status
= EXIT_CREDENTIALS
;
4544 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4548 r
= build_environment(
4560 *exit_status
= EXIT_MEMORY
;
4564 r
= build_pass_environment(context
, &pass_env
);
4566 *exit_status
= EXIT_MEMORY
;
4570 /* The $PATH variable is set to the default path in params->environment. However, this is overridden
4571 * if user-specified fields have $PATH set. The intention is to also override $PATH if the unit does
4572 * not specify PATH but the unit has ExecSearchPath. */
4573 if (!strv_isempty(context
->exec_search_path
)) {
4574 _cleanup_free_
char *joined
= NULL
;
4576 joined
= strv_join(context
->exec_search_path
, ":");
4578 *exit_status
= EXIT_MEMORY
;
4582 r
= strv_env_assign(&joined_exec_search_path
, "PATH", joined
);
4584 *exit_status
= EXIT_MEMORY
;
4589 accum_env
= strv_env_merge(params
->environment
,
4591 joined_exec_search_path
,
4593 context
->environment
,
4596 *exit_status
= EXIT_MEMORY
;
4599 accum_env
= strv_env_clean(accum_env
);
4601 (void) umask(context
->umask
);
4603 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4605 *exit_status
= EXIT_KEYRING
;
4606 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4609 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted
4611 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4613 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked
4614 * for it, and the kernel doesn't actually support ambient caps. */
4615 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4617 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly
4618 * excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not
4620 if (needs_ambient_hack
)
4621 needs_setuid
= false;
4623 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4625 if (needs_sandboxing
) {
4626 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on
4627 * /sys being present. The actual MAC context application will happen later, as late as
4628 * possible, to avoid impacting our own code paths. */
4631 use_selinux
= mac_selinux_use();
4634 use_smack
= mac_smack_use();
4637 use_apparmor
= mac_apparmor_use();
4641 if (needs_sandboxing
) {
4644 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4645 * is set here. (See below.) */
4647 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4649 *exit_status
= EXIT_LIMITS
;
4650 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4654 if (needs_setuid
&& context
->pam_name
&& username
) {
4655 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4656 * wins here. (See above.) */
4658 /* All fds passed in the fds array will be closed in the pam child process. */
4659 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4661 *exit_status
= EXIT_PAM
;
4662 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4665 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4666 if (ngids_after_pam
< 0) {
4667 *exit_status
= EXIT_MEMORY
;
4668 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4672 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4673 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4674 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4675 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4677 userns_set_up
= true;
4678 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4680 *exit_status
= EXIT_USER
;
4681 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4685 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4687 if (ns_type_supported(NAMESPACE_NET
)) {
4688 r
= setup_shareable_ns(runtime
->netns_storage_socket
, CLONE_NEWNET
);
4690 log_unit_warning_errno(unit
, r
,
4691 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4693 *exit_status
= EXIT_NETWORK
;
4694 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4696 } else if (context
->network_namespace_path
) {
4697 *exit_status
= EXIT_NETWORK
;
4698 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4699 "NetworkNamespacePath= is not supported, refusing.");
4701 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4704 if ((context
->private_ipc
|| context
->ipc_namespace_path
) && runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
4706 if (ns_type_supported(NAMESPACE_IPC
)) {
4707 r
= setup_shareable_ns(runtime
->ipcns_storage_socket
, CLONE_NEWIPC
);
4709 log_unit_warning_errno(unit
, r
,
4710 "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
4712 *exit_status
= EXIT_NAMESPACE
;
4713 return log_unit_error_errno(unit
, r
, "Failed to set up IPC namespacing: %m");
4715 } else if (context
->ipc_namespace_path
) {
4716 *exit_status
= EXIT_NAMESPACE
;
4717 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4718 "IPCNamespacePath= is not supported, refusing.");
4720 log_unit_warning(unit
, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
4723 if (needs_mount_namespace
) {
4724 _cleanup_free_
char *error_path
= NULL
;
4726 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4728 *exit_status
= EXIT_NAMESPACE
;
4729 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4730 error_path
? ": " : "", strempty(error_path
));
4734 if (needs_sandboxing
) {
4735 r
= apply_protect_hostname(unit
, context
, exit_status
);
4740 /* Drop groups as early as possible.
4741 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4742 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4744 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4745 int ngids_to_enforce
= 0;
4747 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4752 if (ngids_to_enforce
< 0) {
4753 *exit_status
= EXIT_MEMORY
;
4754 return log_unit_error_errno(unit
,
4756 "Failed to merge group lists. Group membership might be incorrect: %m");
4759 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4761 *exit_status
= EXIT_GROUP
;
4762 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4766 /* If the user namespace was not set up above, try to do it now.
4767 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4768 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4769 * case of mount namespaces being less privileged when the mount point list is copied from a
4770 * different user namespace). */
4772 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4773 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4775 *exit_status
= EXIT_USER
;
4776 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4780 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4783 _cleanup_free_
char *executable
= NULL
;
4784 _cleanup_close_
int executable_fd
= -EBADF
;
4785 r
= find_executable_full(command
->path
, /* root= */ NULL
, context
->exec_search_path
, false, &executable
, &executable_fd
);
4787 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4788 log_unit_struct_errno(unit
, LOG_INFO
, r
,
4789 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4790 LOG_UNIT_INVOCATION_ID(unit
),
4791 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4793 "EXECUTABLE=%s", command
->path
);
4797 *exit_status
= EXIT_EXEC
;
4799 return log_unit_struct_errno(unit
, LOG_INFO
, r
,
4800 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4801 LOG_UNIT_INVOCATION_ID(unit
),
4802 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4804 "EXECUTABLE=%s", command
->path
);
4807 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, executable_fd
, &executable_fd
);
4809 *exit_status
= EXIT_FDS
;
4810 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4814 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
) {
4819 else if (params
->n_socket_fds
== 1)
4820 /* If stdin is not connected to a socket but we are triggered by exactly one socket unit then we
4821 * use context from that fd to compute the label. */
4822 fd
= params
->fds
[0];
4825 r
= mac_selinux_get_child_mls_label(fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4827 if (!context
->selinux_context_ignore
) {
4828 *exit_status
= EXIT_SELINUX_CONTEXT
;
4829 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4831 log_unit_debug_errno(unit
, r
, "Failed to determine SELinux context, ignoring: %m");
4837 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4838 * more aggressive this time since socket_fd and the netns and ipcns fds we don't need anymore. We do keep the exec_fd
4839 * however if we have it as we want to keep it open until the final execve(). */
4841 r
= close_all_fds(keep_fds
, n_keep_fds
);
4843 r
= shift_fds(fds
, n_fds
);
4845 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4847 *exit_status
= EXIT_FDS
;
4848 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4851 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4852 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4853 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4856 secure_bits
= context
->secure_bits
;
4858 if (needs_sandboxing
) {
4861 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4862 * requested. (Note this is placed after the general resource limit initialization, see
4863 * above, in order to take precedence.) */
4864 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4865 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4866 *exit_status
= EXIT_LIMITS
;
4867 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4872 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4873 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4875 r
= setup_smack(unit
->manager
, context
, executable_fd
);
4876 if (r
< 0 && !context
->smack_process_label_ignore
) {
4877 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4878 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4883 bset
= context
->capability_bounding_set
;
4884 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4885 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4886 * instead of us doing that */
4887 if (needs_ambient_hack
)
4888 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4889 (UINT64_C(1) << CAP_SETUID
) |
4890 (UINT64_C(1) << CAP_SETGID
);
4892 if (!cap_test_all(bset
)) {
4893 r
= capability_bounding_set_drop(bset
, false);
4895 *exit_status
= EXIT_CAPABILITIES
;
4896 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4900 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4902 * To be able to raise the ambient capabilities after setresuid() they have to be
4903 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4904 * After setresuid() the ambient capabilities can be raised as they are present in
4905 * the permitted and inhertiable set. However it is possible that someone wants to
4906 * set ambient capabilities without changing the user, so we also set the ambient
4907 * capabilities here.
4908 * The requested ambient capabilities are raised in the inheritable set if the
4909 * second argument is true. */
4910 if (!needs_ambient_hack
) {
4911 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4913 *exit_status
= EXIT_CAPABILITIES
;
4914 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4919 /* chroot to root directory first, before we lose the ability to chroot */
4920 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4922 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4925 if (uid_is_valid(uid
)) {
4926 r
= enforce_user(context
, uid
);
4928 *exit_status
= EXIT_USER
;
4929 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4932 if (!needs_ambient_hack
&&
4933 context
->capability_ambient_set
!= 0) {
4935 /* Raise the ambient capabilities after user change. */
4936 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4938 *exit_status
= EXIT_CAPABILITIES
;
4939 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4945 /* Apply working directory here, because the working directory might be on NFS and only the user running
4946 * this service might have the correct privilege to change to the working directory */
4947 r
= apply_working_directory(context
, params
, home
, exit_status
);
4949 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4951 if (needs_sandboxing
) {
4952 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4953 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4954 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4955 * are restricted. */
4959 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4962 r
= setexeccon(exec_context
);
4964 if (!context
->selinux_context_ignore
) {
4965 *exit_status
= EXIT_SELINUX_CONTEXT
;
4966 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4968 log_unit_debug_errno(unit
, r
, "Failed to change SELinux context to %s, ignoring: %m", exec_context
);
4975 if (use_apparmor
&& context
->apparmor_profile
) {
4976 r
= aa_change_onexec(context
->apparmor_profile
);
4977 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4978 *exit_status
= EXIT_APPARMOR_PROFILE
;
4979 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4984 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4985 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4987 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4988 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4989 * effective set here.
4990 * The effective set is overwritten during execve with the following values:
4991 * - ambient set (for non-root processes)
4992 * - (inheritable | bounding) set for root processes)
4994 * Hence there is no security impact to raise it in the effective set before execve
4996 r
= capability_gain_cap_setpcap(NULL
);
4998 *exit_status
= EXIT_CAPABILITIES
;
4999 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
5001 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
5002 *exit_status
= EXIT_SECUREBITS
;
5003 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
5007 if (context_has_no_new_privileges(context
))
5008 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
5009 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
5010 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
5014 r
= apply_address_families(unit
, context
);
5016 *exit_status
= EXIT_ADDRESS_FAMILIES
;
5017 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
5020 r
= apply_memory_deny_write_execute(unit
, context
);
5022 *exit_status
= EXIT_SECCOMP
;
5023 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
5026 r
= apply_restrict_realtime(unit
, context
);
5028 *exit_status
= EXIT_SECCOMP
;
5029 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
5032 r
= apply_restrict_suid_sgid(unit
, context
);
5034 *exit_status
= EXIT_SECCOMP
;
5035 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
5038 r
= apply_restrict_namespaces(unit
, context
);
5040 *exit_status
= EXIT_SECCOMP
;
5041 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
5044 r
= apply_protect_sysctl(unit
, context
);
5046 *exit_status
= EXIT_SECCOMP
;
5047 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
5050 r
= apply_protect_kernel_modules(unit
, context
);
5052 *exit_status
= EXIT_SECCOMP
;
5053 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
5056 r
= apply_protect_kernel_logs(unit
, context
);
5058 *exit_status
= EXIT_SECCOMP
;
5059 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
5062 r
= apply_protect_clock(unit
, context
);
5064 *exit_status
= EXIT_SECCOMP
;
5065 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
5068 r
= apply_private_devices(unit
, context
);
5070 *exit_status
= EXIT_SECCOMP
;
5071 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
5074 r
= apply_syscall_archs(unit
, context
);
5076 *exit_status
= EXIT_SECCOMP
;
5077 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
5080 r
= apply_lock_personality(unit
, context
);
5082 *exit_status
= EXIT_SECCOMP
;
5083 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
5086 r
= apply_syscall_log(unit
, context
);
5088 *exit_status
= EXIT_SECCOMP
;
5089 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
5092 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
5093 * by the filter as little as possible. */
5094 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
5096 *exit_status
= EXIT_SECCOMP
;
5097 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
5102 r
= apply_restrict_filesystems(unit
, context
);
5104 *exit_status
= EXIT_BPF
;
5105 return log_unit_error_errno(unit
, r
, "Failed to restrict filesystems: %m");
5111 if (!strv_isempty(context
->unset_environment
)) {
5114 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
5116 *exit_status
= EXIT_MEMORY
;
5120 strv_free_and_replace(accum_env
, ee
);
5123 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
5124 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
5125 if (!replaced_argv
) {
5126 *exit_status
= EXIT_MEMORY
;
5129 final_argv
= replaced_argv
;
5131 final_argv
= command
->argv
;
5133 if (DEBUG_LOGGING
) {
5134 _cleanup_free_
char *line
= NULL
;
5136 line
= quote_command_line(final_argv
, SHELL_ESCAPE_EMPTY
);
5138 *exit_status
= EXIT_MEMORY
;
5142 log_unit_struct(unit
, LOG_DEBUG
,
5143 "EXECUTABLE=%s", executable
,
5144 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
));
5150 /* We have finished with all our initializations. Let's now let the manager know that. From this point
5151 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
5153 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
5154 *exit_status
= EXIT_EXEC
;
5155 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
5159 r
= fexecve_or_execve(executable_fd
, executable
, final_argv
, accum_env
);
5164 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
5165 * that POLLHUP on it no longer means execve() succeeded. */
5167 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
5168 *exit_status
= EXIT_EXEC
;
5169 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
5173 *exit_status
= EXIT_EXEC
;
5174 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
5177 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
5178 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
5180 int exec_spawn(Unit
*unit
,
5181 ExecCommand
*command
,
5182 const ExecContext
*context
,
5183 const ExecParameters
*params
,
5184 ExecRuntime
*runtime
,
5185 DynamicCreds
*dcreds
,
5188 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
5189 _cleanup_free_
char *subcgroup_path
= NULL
;
5190 _cleanup_strv_free_
char **files_env
= NULL
;
5191 size_t n_storage_fds
= 0, n_socket_fds
= 0;
5192 _cleanup_free_
char *line
= NULL
;
5200 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
5202 if (context
->std_input
== EXEC_INPUT_SOCKET
||
5203 context
->std_output
== EXEC_OUTPUT_SOCKET
||
5204 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
5206 if (params
->n_socket_fds
> 1)
5207 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got more than one socket.");
5209 if (params
->n_socket_fds
== 0)
5210 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got no socket.");
5212 socket_fd
= params
->fds
[0];
5216 n_socket_fds
= params
->n_socket_fds
;
5217 n_storage_fds
= params
->n_storage_fds
;
5220 r
= exec_context_named_iofds(context
, params
, named_iofds
);
5222 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
5224 r
= exec_context_load_environment(unit
, context
, &files_env
);
5226 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
5228 line
= quote_command_line(command
->argv
, SHELL_ESCAPE_EMPTY
);
5232 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
5233 and, until the next SELinux policy changes, we save further reloads in future children. */
5234 mac_selinux_maybe_reload();
5236 log_unit_struct(unit
, LOG_DEBUG
,
5237 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
5238 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
5239 the mount namespace in the child, but we want to log
5240 from the parent, so we need to use the (possibly
5241 inaccurate) path here. */
5242 LOG_UNIT_INVOCATION_ID(unit
));
5244 if (params
->cgroup_path
) {
5245 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
5247 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
5248 if (r
> 0) { /* We are using a child cgroup */
5249 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
5251 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
5253 /* Normally we would not propagate the xattrs to children but since we created this
5254 * sub-cgroup internally we should do it. */
5255 cgroup_oomd_xattr_apply(unit
, subcgroup_path
);
5256 cgroup_log_xattr_apply(unit
, subcgroup_path
);
5262 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
5265 int exit_status
= EXIT_SUCCESS
;
5267 r
= exec_child(unit
,
5279 unit
->manager
->user_lookup_fds
[1],
5283 const char *status
=
5284 exit_status_to_string(exit_status
,
5285 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
5287 log_unit_struct_errno(unit
, LOG_ERR
, r
,
5288 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
5289 LOG_UNIT_INVOCATION_ID(unit
),
5290 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
5291 status
, command
->path
),
5292 "EXECUTABLE=%s", command
->path
);
5298 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
5300 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
5301 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
5302 * process will be killed too). */
5304 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
5306 exec_status_start(&command
->exec_status
, pid
);
5312 void exec_context_init(ExecContext
*c
) {
5316 c
->ioprio
= IOPRIO_DEFAULT_CLASS_AND_PRIO
;
5317 c
->cpu_sched_policy
= SCHED_OTHER
;
5318 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
5319 c
->syslog_level_prefix
= true;
5320 c
->ignore_sigpipe
= true;
5321 c
->timer_slack_nsec
= NSEC_INFINITY
;
5322 c
->personality
= PERSONALITY_INVALID
;
5323 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5324 c
->directories
[t
].mode
= 0755;
5325 c
->timeout_clean_usec
= USEC_INFINITY
;
5326 c
->capability_bounding_set
= CAP_ALL
;
5327 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
5328 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
5329 c
->log_level_max
= -1;
5331 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
5333 c
->tty_rows
= UINT_MAX
;
5334 c
->tty_cols
= UINT_MAX
;
5335 numa_policy_reset(&c
->numa_policy
);
5338 void exec_context_done(ExecContext
*c
) {
5341 c
->environment
= strv_free(c
->environment
);
5342 c
->environment_files
= strv_free(c
->environment_files
);
5343 c
->pass_environment
= strv_free(c
->pass_environment
);
5344 c
->unset_environment
= strv_free(c
->unset_environment
);
5346 rlimit_free_all(c
->rlimit
);
5348 for (size_t l
= 0; l
< 3; l
++) {
5349 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
5350 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
5353 c
->working_directory
= mfree(c
->working_directory
);
5354 c
->root_directory
= mfree(c
->root_directory
);
5355 c
->root_image
= mfree(c
->root_image
);
5356 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
5357 c
->root_hash
= mfree(c
->root_hash
);
5358 c
->root_hash_size
= 0;
5359 c
->root_hash_path
= mfree(c
->root_hash_path
);
5360 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
5361 c
->root_hash_sig_size
= 0;
5362 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
5363 c
->root_verity
= mfree(c
->root_verity
);
5364 c
->extension_images
= mount_image_free_many(c
->extension_images
, &c
->n_extension_images
);
5365 c
->extension_directories
= strv_free(c
->extension_directories
);
5366 c
->tty_path
= mfree(c
->tty_path
);
5367 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
5368 c
->user
= mfree(c
->user
);
5369 c
->group
= mfree(c
->group
);
5371 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
5373 c
->pam_name
= mfree(c
->pam_name
);
5375 c
->read_only_paths
= strv_free(c
->read_only_paths
);
5376 c
->read_write_paths
= strv_free(c
->read_write_paths
);
5377 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
5378 c
->exec_paths
= strv_free(c
->exec_paths
);
5379 c
->no_exec_paths
= strv_free(c
->no_exec_paths
);
5380 c
->exec_search_path
= strv_free(c
->exec_search_path
);
5382 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
5383 c
->bind_mounts
= NULL
;
5384 c
->n_bind_mounts
= 0;
5385 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
5386 c
->temporary_filesystems
= NULL
;
5387 c
->n_temporary_filesystems
= 0;
5388 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
5390 cpu_set_reset(&c
->cpu_set
);
5391 numa_policy_reset(&c
->numa_policy
);
5393 c
->utmp_id
= mfree(c
->utmp_id
);
5394 c
->selinux_context
= mfree(c
->selinux_context
);
5395 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
5396 c
->smack_process_label
= mfree(c
->smack_process_label
);
5398 c
->restrict_filesystems
= set_free(c
->restrict_filesystems
);
5400 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
5401 c
->syscall_archs
= set_free(c
->syscall_archs
);
5402 c
->address_families
= set_free(c
->address_families
);
5404 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5405 exec_directory_done(&c
->directories
[t
]);
5407 c
->log_level_max
= -1;
5409 exec_context_free_log_extra_fields(c
);
5410 c
->log_filter_allowed_patterns
= set_free(c
->log_filter_allowed_patterns
);
5411 c
->log_filter_denied_patterns
= set_free(c
->log_filter_denied_patterns
);
5413 c
->log_ratelimit_interval_usec
= 0;
5414 c
->log_ratelimit_burst
= 0;
5416 c
->stdin_data
= mfree(c
->stdin_data
);
5417 c
->stdin_data_size
= 0;
5419 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
5420 c
->ipc_namespace_path
= mfree(c
->ipc_namespace_path
);
5422 c
->log_namespace
= mfree(c
->log_namespace
);
5424 c
->load_credentials
= hashmap_free(c
->load_credentials
);
5425 c
->set_credentials
= hashmap_free(c
->set_credentials
);
5428 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
5431 if (!runtime_prefix
)
5434 for (size_t i
= 0; i
< c
->directories
[EXEC_DIRECTORY_RUNTIME
].n_items
; i
++) {
5435 _cleanup_free_
char *p
= NULL
;
5437 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
5438 p
= path_join(runtime_prefix
, "private", c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].path
);
5440 p
= path_join(runtime_prefix
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].path
);
5444 /* We execute this synchronously, since we need to be sure this is gone when we start the
5446 (void) rm_rf(p
, REMOVE_ROOT
);
5448 STRV_FOREACH(symlink
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].symlinks
) {
5449 _cleanup_free_
char *symlink_abs
= NULL
;
5451 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
5452 symlink_abs
= path_join(runtime_prefix
, "private", *symlink
);
5454 symlink_abs
= path_join(runtime_prefix
, *symlink
);
5458 (void) unlink(symlink_abs
);
5466 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
5467 _cleanup_free_
char *p
= NULL
;
5471 if (!runtime_prefix
|| !unit
)
5474 p
= path_join(runtime_prefix
, "credentials", unit
);
5478 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
5479 * unmount it, and afterwards remove the mount point */
5480 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
5481 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
5486 static void exec_command_done(ExecCommand
*c
) {
5489 c
->path
= mfree(c
->path
);
5490 c
->argv
= strv_free(c
->argv
);
5493 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
5494 for (size_t i
= 0; i
< n
; i
++)
5495 exec_command_done(c
+i
);
5498 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
5502 LIST_REMOVE(command
, c
, i
);
5503 exec_command_done(i
);
5510 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
5511 for (size_t i
= 0; i
< n
; i
++)
5512 c
[i
] = exec_command_free_list(c
[i
]);
5515 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
5516 for (size_t i
= 0; i
< n
; i
++)
5517 exec_status_reset(&c
[i
].exec_status
);
5520 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
5521 for (size_t i
= 0; i
< n
; i
++)
5522 LIST_FOREACH(command
, z
, c
[i
])
5523 exec_status_reset(&z
->exec_status
);
5526 typedef struct InvalidEnvInfo
{
5531 static void invalid_env(const char *p
, void *userdata
) {
5532 InvalidEnvInfo
*info
= userdata
;
5534 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
5537 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
5543 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
5546 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
5549 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
5552 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
5555 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
5558 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
5565 static int exec_context_named_iofds(
5566 const ExecContext
*c
,
5567 const ExecParameters
*p
,
5568 int named_iofds
[static 3]) {
5571 const char* stdio_fdname
[3];
5576 assert(named_iofds
);
5578 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
5579 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
5580 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
5582 for (size_t i
= 0; i
< 3; i
++)
5583 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
5585 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
5587 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
5588 if (named_iofds
[STDIN_FILENO
] < 0 &&
5589 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
5590 stdio_fdname
[STDIN_FILENO
] &&
5591 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
5593 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
5596 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
5597 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
5598 stdio_fdname
[STDOUT_FILENO
] &&
5599 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
5601 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5604 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5605 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5606 stdio_fdname
[STDERR_FILENO
] &&
5607 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5609 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5613 return targets
== 0 ? 0 : -ENOENT
;
5616 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***ret
) {
5617 _cleanup_strv_free_
char **v
= NULL
;
5623 STRV_FOREACH(i
, c
->environment_files
) {
5624 _cleanup_globfree_ glob_t pglob
= {};
5625 bool ignore
= false;
5633 if (!path_is_absolute(fn
)) {
5639 /* Filename supports globbing, take all matching files */
5640 r
= safe_glob(fn
, 0, &pglob
);
5647 /* When we don't match anything, -ENOENT should be returned */
5648 assert(pglob
.gl_pathc
> 0);
5650 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5651 _cleanup_strv_free_
char **p
= NULL
;
5653 r
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5660 /* Log invalid environment variables with filename */
5662 InvalidEnvInfo info
= {
5664 .path
= pglob
.gl_pathv
[n
]
5667 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5673 char **m
= strv_env_merge(v
, p
);
5677 strv_free_and_replace(v
, m
);
5687 static bool tty_may_match_dev_console(const char *tty
) {
5688 _cleanup_free_
char *resolved
= NULL
;
5693 tty
= skip_dev_prefix(tty
);
5695 /* trivial identity? */
5696 if (streq(tty
, "console"))
5699 if (resolve_dev_console(&resolved
) < 0)
5700 return true; /* if we could not resolve, assume it may */
5702 /* "tty0" means the active VC, so it may be the same sometimes */
5703 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5706 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5709 return ec
->tty_reset
||
5711 ec
->tty_vt_disallocate
||
5712 is_terminal_input(ec
->std_input
) ||
5713 is_terminal_output(ec
->std_output
) ||
5714 is_terminal_output(ec
->std_error
);
5717 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5719 return exec_context_may_touch_tty(ec
) &&
5720 tty_may_match_dev_console(exec_context_tty_path(ec
));
5723 static void strv_fprintf(FILE *f
, char **l
) {
5727 fprintf(f
, " %s", *g
);
5730 static void strv_dump(FILE* f
, const char *prefix
, const char *name
, char **strv
) {
5735 if (!strv_isempty(strv
)) {
5736 fprintf(f
, "%s%s:", prefix
, name
);
5737 strv_fprintf(f
, strv
);
5742 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5748 prefix
= strempty(prefix
);
5752 "%sWorkingDirectory: %s\n"
5753 "%sRootDirectory: %s\n"
5754 "%sNonBlocking: %s\n"
5755 "%sPrivateTmp: %s\n"
5756 "%sPrivateDevices: %s\n"
5757 "%sProtectKernelTunables: %s\n"
5758 "%sProtectKernelModules: %s\n"
5759 "%sProtectKernelLogs: %s\n"
5760 "%sProtectClock: %s\n"
5761 "%sProtectControlGroups: %s\n"
5762 "%sPrivateNetwork: %s\n"
5763 "%sPrivateUsers: %s\n"
5764 "%sProtectHome: %s\n"
5765 "%sProtectSystem: %s\n"
5766 "%sMountAPIVFS: %s\n"
5767 "%sIgnoreSIGPIPE: %s\n"
5768 "%sMemoryDenyWriteExecute: %s\n"
5769 "%sRestrictRealtime: %s\n"
5770 "%sRestrictSUIDSGID: %s\n"
5771 "%sKeyringMode: %s\n"
5772 "%sProtectHostname: %s\n"
5773 "%sProtectProc: %s\n"
5774 "%sProcSubset: %s\n",
5776 prefix
, empty_to_root(c
->working_directory
),
5777 prefix
, empty_to_root(c
->root_directory
),
5778 prefix
, yes_no(c
->non_blocking
),
5779 prefix
, yes_no(c
->private_tmp
),
5780 prefix
, yes_no(c
->private_devices
),
5781 prefix
, yes_no(c
->protect_kernel_tunables
),
5782 prefix
, yes_no(c
->protect_kernel_modules
),
5783 prefix
, yes_no(c
->protect_kernel_logs
),
5784 prefix
, yes_no(c
->protect_clock
),
5785 prefix
, yes_no(c
->protect_control_groups
),
5786 prefix
, yes_no(c
->private_network
),
5787 prefix
, yes_no(c
->private_users
),
5788 prefix
, protect_home_to_string(c
->protect_home
),
5789 prefix
, protect_system_to_string(c
->protect_system
),
5790 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
5791 prefix
, yes_no(c
->ignore_sigpipe
),
5792 prefix
, yes_no(c
->memory_deny_write_execute
),
5793 prefix
, yes_no(c
->restrict_realtime
),
5794 prefix
, yes_no(c
->restrict_suid_sgid
),
5795 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5796 prefix
, yes_no(c
->protect_hostname
),
5797 prefix
, protect_proc_to_string(c
->protect_proc
),
5798 prefix
, proc_subset_to_string(c
->proc_subset
));
5801 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5803 if (c
->root_image_options
) {
5804 fprintf(f
, "%sRootImageOptions:", prefix
);
5805 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5806 if (!isempty(o
->options
))
5807 fprintf(f
, " %s:%s",
5808 partition_designator_to_string(o
->partition_designator
),
5814 _cleanup_free_
char *encoded
= NULL
;
5815 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5817 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5820 if (c
->root_hash_path
)
5821 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5823 if (c
->root_hash_sig
) {
5824 _cleanup_free_
char *encoded
= NULL
;
5826 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5828 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5831 if (c
->root_hash_sig_path
)
5832 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5835 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5837 STRV_FOREACH(e
, c
->environment
)
5838 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5840 STRV_FOREACH(e
, c
->environment_files
)
5841 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5843 STRV_FOREACH(e
, c
->pass_environment
)
5844 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5846 STRV_FOREACH(e
, c
->unset_environment
)
5847 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5849 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5851 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5852 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5854 for (size_t i
= 0; i
< c
->directories
[dt
].n_items
; i
++) {
5855 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].items
[i
].path
);
5857 STRV_FOREACH(d
, c
->directories
[dt
].items
[i
].symlinks
)
5858 fprintf(f
, "%s%s: %s:%s\n", prefix
, exec_directory_type_symlink_to_string(dt
), c
->directories
[dt
].items
[i
].path
, *d
);
5862 fprintf(f
, "%sTimeoutCleanSec: %s\n", prefix
, FORMAT_TIMESPAN(c
->timeout_clean_usec
, USEC_PER_SEC
));
5865 fprintf(f
, "%sNice: %i\n", prefix
, c
->nice
);
5867 if (c
->oom_score_adjust_set
)
5868 fprintf(f
, "%sOOMScoreAdjust: %i\n", prefix
, c
->oom_score_adjust
);
5870 if (c
->coredump_filter_set
)
5871 fprintf(f
, "%sCoredumpFilter: 0x%"PRIx64
"\n", prefix
, c
->coredump_filter
);
5873 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5875 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5876 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5877 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5878 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5881 if (c
->ioprio_set
) {
5882 _cleanup_free_
char *class_str
= NULL
;
5884 r
= ioprio_class_to_string_alloc(ioprio_prio_class(c
->ioprio
), &class_str
);
5886 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5888 fprintf(f
, "%sIOPriority: %d\n", prefix
, ioprio_prio_data(c
->ioprio
));
5891 if (c
->cpu_sched_set
) {
5892 _cleanup_free_
char *policy_str
= NULL
;
5894 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5896 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5899 "%sCPUSchedulingPriority: %i\n"
5900 "%sCPUSchedulingResetOnFork: %s\n",
5901 prefix
, c
->cpu_sched_priority
,
5902 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5905 if (c
->cpu_set
.set
) {
5906 _cleanup_free_
char *affinity
= NULL
;
5908 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5909 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5912 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5913 _cleanup_free_
char *nodes
= NULL
;
5915 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5916 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5917 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5920 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5921 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5924 "%sStandardInput: %s\n"
5925 "%sStandardOutput: %s\n"
5926 "%sStandardError: %s\n",
5927 prefix
, exec_input_to_string(c
->std_input
),
5928 prefix
, exec_output_to_string(c
->std_output
),
5929 prefix
, exec_output_to_string(c
->std_error
));
5931 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5932 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5933 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5934 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5935 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5936 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5938 if (c
->std_input
== EXEC_INPUT_FILE
)
5939 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5940 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5941 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5942 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5943 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5944 if (c
->std_output
== EXEC_OUTPUT_FILE_TRUNCATE
)
5945 fprintf(f
, "%sStandardOutputFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5946 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5947 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5948 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5949 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5950 if (c
->std_error
== EXEC_OUTPUT_FILE_TRUNCATE
)
5951 fprintf(f
, "%sStandardErrorFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5957 "%sTTYVHangup: %s\n"
5958 "%sTTYVTDisallocate: %s\n"
5960 "%sTTYColumns: %u\n",
5961 prefix
, c
->tty_path
,
5962 prefix
, yes_no(c
->tty_reset
),
5963 prefix
, yes_no(c
->tty_vhangup
),
5964 prefix
, yes_no(c
->tty_vt_disallocate
),
5965 prefix
, c
->tty_rows
,
5966 prefix
, c
->tty_cols
);
5968 if (IN_SET(c
->std_output
,
5970 EXEC_OUTPUT_JOURNAL
,
5971 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5972 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5973 IN_SET(c
->std_error
,
5975 EXEC_OUTPUT_JOURNAL
,
5976 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5977 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5979 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5981 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5983 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5985 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5987 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5990 if (c
->log_level_max
>= 0) {
5991 _cleanup_free_
char *t
= NULL
;
5993 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5995 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5998 if (c
->log_ratelimit_interval_usec
> 0)
6000 "%sLogRateLimitIntervalSec: %s\n",
6001 prefix
, FORMAT_TIMESPAN(c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
6003 if (c
->log_ratelimit_burst
> 0)
6004 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
6006 if (!set_isempty(c
->log_filter_allowed_patterns
) || !set_isempty(c
->log_filter_denied_patterns
)) {
6007 fprintf(f
, "%sLogFilterPatterns:", prefix
);
6010 SET_FOREACH(pattern
, c
->log_filter_allowed_patterns
)
6011 fprintf(f
, " %s", pattern
);
6012 SET_FOREACH(pattern
, c
->log_filter_denied_patterns
)
6013 fprintf(f
, " ~%s", pattern
);
6017 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
6018 fprintf(f
, "%sLogExtraFields: ", prefix
);
6019 fwrite(c
->log_extra_fields
[j
].iov_base
,
6020 1, c
->log_extra_fields
[j
].iov_len
,
6025 if (c
->log_namespace
)
6026 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
6028 if (c
->secure_bits
) {
6029 _cleanup_free_
char *str
= NULL
;
6031 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
6033 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
6036 if (c
->capability_bounding_set
!= CAP_ALL
) {
6037 _cleanup_free_
char *str
= NULL
;
6039 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
6041 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
6044 if (c
->capability_ambient_set
!= 0) {
6045 _cleanup_free_
char *str
= NULL
;
6047 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
6049 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
6053 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
6055 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
6057 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
6059 strv_dump(f
, prefix
, "SupplementaryGroups", c
->supplementary_groups
);
6062 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
6064 strv_dump(f
, prefix
, "ReadWritePaths", c
->read_write_paths
);
6065 strv_dump(f
, prefix
, "ReadOnlyPaths", c
->read_only_paths
);
6066 strv_dump(f
, prefix
, "InaccessiblePaths", c
->inaccessible_paths
);
6067 strv_dump(f
, prefix
, "ExecPaths", c
->exec_paths
);
6068 strv_dump(f
, prefix
, "NoExecPaths", c
->no_exec_paths
);
6069 strv_dump(f
, prefix
, "ExecSearchPath", c
->exec_search_path
);
6071 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
6072 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
6073 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
6074 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
6075 c
->bind_mounts
[i
].source
,
6076 c
->bind_mounts
[i
].destination
,
6077 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
6079 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
6080 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
6082 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
6084 isempty(t
->options
) ? "" : ":",
6085 strempty(t
->options
));
6090 "%sUtmpIdentifier: %s\n",
6091 prefix
, c
->utmp_id
);
6093 if (c
->selinux_context
)
6095 "%sSELinuxContext: %s%s\n",
6096 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
6098 if (c
->apparmor_profile
)
6100 "%sAppArmorProfile: %s%s\n",
6101 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
6103 if (c
->smack_process_label
)
6105 "%sSmackProcessLabel: %s%s\n",
6106 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
6108 if (c
->personality
!= PERSONALITY_INVALID
)
6110 "%sPersonality: %s\n",
6111 prefix
, strna(personality_to_string(c
->personality
)));
6114 "%sLockPersonality: %s\n",
6115 prefix
, yes_no(c
->lock_personality
));
6117 if (c
->syscall_filter
) {
6119 "%sSystemCallFilter: ",
6122 if (!c
->syscall_allow_list
)
6128 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
6129 _cleanup_free_
char *name
= NULL
;
6130 const char *errno_name
= NULL
;
6131 int num
= PTR_TO_INT(val
);
6138 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
6139 fputs(strna(name
), f
);
6142 errno_name
= seccomp_errno_or_action_to_string(num
);
6144 fprintf(f
, ":%s", errno_name
);
6146 fprintf(f
, ":%d", num
);
6154 if (c
->syscall_archs
) {
6156 "%sSystemCallArchitectures:",
6161 SET_FOREACH(id
, c
->syscall_archs
)
6162 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
6167 if (exec_context_restrict_namespaces_set(c
)) {
6168 _cleanup_free_
char *s
= NULL
;
6170 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
6172 fprintf(f
, "%sRestrictNamespaces: %s\n",
6177 if (exec_context_restrict_filesystems_set(c
)) {
6179 SET_FOREACH(fs
, c
->restrict_filesystems
)
6180 fprintf(f
, "%sRestrictFileSystems: %s\n", prefix
, fs
);
6184 if (c
->network_namespace_path
)
6186 "%sNetworkNamespacePath: %s\n",
6187 prefix
, c
->network_namespace_path
);
6189 if (c
->syscall_errno
> 0) {
6190 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
6193 const char *errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
6195 fputs(errno_name
, f
);
6197 fprintf(f
, "%d", c
->syscall_errno
);
6202 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
6203 fprintf(f
, "%sMountImages: %s%s:%s", prefix
,
6204 c
->mount_images
[i
].ignore_enoent
? "-": "",
6205 c
->mount_images
[i
].source
,
6206 c
->mount_images
[i
].destination
);
6207 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
6208 fprintf(f
, ":%s:%s",
6209 partition_designator_to_string(o
->partition_designator
),
6210 strempty(o
->options
));
6214 for (size_t i
= 0; i
< c
->n_extension_images
; i
++) {
6215 fprintf(f
, "%sExtensionImages: %s%s", prefix
,
6216 c
->extension_images
[i
].ignore_enoent
? "-": "",
6217 c
->extension_images
[i
].source
);
6218 LIST_FOREACH(mount_options
, o
, c
->extension_images
[i
].mount_options
)
6219 fprintf(f
, ":%s:%s",
6220 partition_designator_to_string(o
->partition_designator
),
6221 strempty(o
->options
));
6225 strv_dump(f
, prefix
, "ExtensionDirectories", c
->extension_directories
);
6228 bool exec_context_maintains_privileges(const ExecContext
*c
) {
6231 /* Returns true if the process forked off would run under
6232 * an unchanged UID or as root. */
6237 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
6243 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
6251 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
6253 return IOPRIO_DEFAULT_CLASS_AND_PRIO
;
6255 return ioprio_normalize(p
);
6258 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
6261 /* Explicit setting wins */
6262 if (c
->mount_apivfs_set
)
6263 return c
->mount_apivfs
;
6265 /* Default to "yes" if root directory or image are specified */
6266 if (exec_context_with_rootfs(c
))
6272 void exec_context_free_log_extra_fields(ExecContext
*c
) {
6275 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
6276 free(c
->log_extra_fields
[l
].iov_base
);
6277 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
6278 c
->n_log_extra_fields
= 0;
6281 void exec_context_revert_tty(ExecContext
*c
) {
6282 _cleanup_close_
int fd
= -EBADF
;
6289 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
6290 exec_context_tty_reset(c
, NULL
);
6292 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
6293 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
6294 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
6295 if (!exec_context_may_touch_tty(c
))
6298 path
= exec_context_tty_path(c
);
6302 fd
= open(path
, O_PATH
|O_CLOEXEC
);
6304 return (void) log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
6305 "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m",
6308 if (fstat(fd
, &st
) < 0)
6309 return (void) log_warning_errno(errno
, "Failed to stat TTY '%s', ignoring: %m", path
);
6311 /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check
6312 * if things are a character device, since a proper check either means we'd have to open the TTY and
6313 * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects
6314 * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother
6315 * with this at all? → https://github.com/systemd/systemd/issues/19213 */
6316 if (!S_ISCHR(st
.st_mode
))
6317 return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path
);
6319 r
= fchmod_and_chown(fd
, TTY_MODE
, 0, TTY_GID
);
6321 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
6324 int exec_context_get_clean_directories(
6330 _cleanup_strv_free_
char **l
= NULL
;
6337 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
6338 if (!FLAGS_SET(mask
, 1U << t
))
6344 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
6347 j
= path_join(prefix
[t
], c
->directories
[t
].items
[i
].path
);
6351 r
= strv_consume(&l
, j
);
6355 /* Also remove private directories unconditionally. */
6356 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
6357 j
= path_join(prefix
[t
], "private", c
->directories
[t
].items
[i
].path
);
6361 r
= strv_consume(&l
, j
);
6366 STRV_FOREACH(symlink
, c
->directories
[t
].items
[i
].symlinks
) {
6367 j
= path_join(prefix
[t
], *symlink
);
6371 r
= strv_consume(&l
, j
);
6382 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
6383 ExecCleanMask mask
= 0;
6388 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
6389 if (c
->directories
[t
].n_items
> 0)
6396 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
6403 dual_timestamp_get(&s
->start_timestamp
);
6406 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
6414 dual_timestamp_get(&s
->exit_timestamp
);
6419 if (context
&& context
->utmp_id
)
6420 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
6423 void exec_status_reset(ExecStatus
*s
) {
6426 *s
= (ExecStatus
) {};
6429 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
6436 prefix
= strempty(prefix
);
6439 "%sPID: "PID_FMT
"\n",
6442 if (dual_timestamp_is_set(&s
->start_timestamp
))
6444 "%sStart Timestamp: %s\n",
6445 prefix
, FORMAT_TIMESTAMP(s
->start_timestamp
.realtime
));
6447 if (dual_timestamp_is_set(&s
->exit_timestamp
))
6449 "%sExit Timestamp: %s\n"
6451 "%sExit Status: %i\n",
6452 prefix
, FORMAT_TIMESTAMP(s
->exit_timestamp
.realtime
),
6453 prefix
, sigchld_code_to_string(s
->code
),
6457 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6458 _cleanup_free_
char *cmd
= NULL
;
6459 const char *prefix2
;
6464 prefix
= strempty(prefix
);
6465 prefix2
= strjoina(prefix
, "\t");
6467 cmd
= quote_command_line(c
->argv
, SHELL_ESCAPE_EMPTY
);
6470 "%sCommand Line: %s\n",
6471 prefix
, strnull(cmd
));
6473 exec_status_dump(&c
->exec_status
, f
, prefix2
);
6476 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6479 prefix
= strempty(prefix
);
6481 LIST_FOREACH(command
, i
, c
)
6482 exec_command_dump(i
, f
, prefix
);
6485 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
6492 /* It's kind of important, that we keep the order here */
6493 LIST_FIND_TAIL(command
, *l
, end
);
6494 LIST_INSERT_AFTER(command
, *l
, end
, e
);
6499 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
6507 l
= strv_new_ap(path
, ap
);
6519 free_and_replace(c
->path
, p
);
6521 return strv_free_and_replace(c
->argv
, l
);
6524 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
6525 _cleanup_strv_free_
char **l
= NULL
;
6533 l
= strv_new_ap(path
, ap
);
6539 r
= strv_extend_strv(&c
->argv
, l
, false);
6546 static void *remove_tmpdir_thread(void *p
) {
6547 _cleanup_free_
char *path
= p
;
6549 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
6553 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
6560 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
6562 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
6564 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6565 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
6567 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
6569 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
6574 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6575 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
6577 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
6579 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
6581 rt
->var_tmp_dir
= NULL
;
6584 rt
->id
= mfree(rt
->id
);
6585 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
6586 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
6587 safe_close_pair(rt
->netns_storage_socket
);
6588 safe_close_pair(rt
->ipcns_storage_socket
);
6592 static void exec_runtime_freep(ExecRuntime
**rt
) {
6593 (void) exec_runtime_free(*rt
, false);
6596 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
6597 _cleanup_free_
char *id_copy
= NULL
;
6602 id_copy
= strdup(id
);
6606 n
= new(ExecRuntime
, 1);
6610 *n
= (ExecRuntime
) {
6611 .id
= TAKE_PTR(id_copy
),
6612 .netns_storage_socket
= PIPE_EBADF
,
6613 .ipcns_storage_socket
= PIPE_EBADF
,
6620 static int exec_runtime_add(
6625 int netns_storage_socket
[2],
6626 int ipcns_storage_socket
[2],
6627 ExecRuntime
**ret
) {
6629 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6635 /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */
6637 r
= exec_runtime_allocate(&rt
, id
);
6641 r
= hashmap_ensure_put(&m
->exec_runtime_by_id
, &string_hash_ops
, rt
->id
, rt
);
6645 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6646 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6647 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6649 if (netns_storage_socket
) {
6650 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6651 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6654 if (ipcns_storage_socket
) {
6655 rt
->ipcns_storage_socket
[0] = TAKE_FD(ipcns_storage_socket
[0]);
6656 rt
->ipcns_storage_socket
[1] = TAKE_FD(ipcns_storage_socket
[1]);
6663 /* do not remove created ExecRuntime object when the operation succeeds. */
6668 static int exec_runtime_make(
6670 const ExecContext
*c
,
6672 ExecRuntime
**ret
) {
6674 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6675 _cleanup_close_pair_
int netns_storage_socket
[2] = PIPE_EBADF
, ipcns_storage_socket
[2] = PIPE_EBADF
;
6682 /* It is not necessary to create ExecRuntime object. */
6683 if (!c
->private_network
&& !c
->private_ipc
&& !c
->private_tmp
&& !c
->network_namespace_path
) {
6688 if (c
->private_tmp
&&
6689 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6690 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6691 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6692 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6697 if (c
->private_network
|| c
->network_namespace_path
) {
6698 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6702 if (c
->private_ipc
|| c
->ipc_namespace_path
) {
6703 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, ipcns_storage_socket
) < 0)
6707 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ipcns_storage_socket
, ret
);
6714 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6722 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6724 /* We already have an ExecRuntime object, let's increase the ref count and reuse it */
6732 /* If not found, then create a new object. */
6733 r
= exec_runtime_make(m
, c
, id
, &rt
);
6737 /* When r == 0, it is not necessary to create ExecRuntime object. */
6743 /* increment reference counter. */
6749 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6753 assert(rt
->n_ref
> 0);
6759 return exec_runtime_free(rt
, destroy
);
6762 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6769 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6770 fprintf(f
, "exec-runtime=%s", rt
->id
);
6773 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6775 if (rt
->var_tmp_dir
)
6776 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6778 if (rt
->netns_storage_socket
[0] >= 0) {
6781 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6785 fprintf(f
, " netns-socket-0=%i", copy
);
6788 if (rt
->netns_storage_socket
[1] >= 0) {
6791 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6795 fprintf(f
, " netns-socket-1=%i", copy
);
6798 if (rt
->ipcns_storage_socket
[0] >= 0) {
6801 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[0]);
6805 fprintf(f
, " ipcns-socket-0=%i", copy
);
6808 if (rt
->ipcns_storage_socket
[1] >= 0) {
6811 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[1]);
6815 fprintf(f
, " ipcns-socket-1=%i", copy
);
6824 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6825 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6829 /* This is for the migration from old (v237 or earlier) deserialization text.
6830 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6831 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6832 * so or not from the serialized text, then we always creates a new object owned by this. */
6838 /* Manager manages ExecRuntime objects by the unit id.
6839 * So, we omit the serialized text when the unit does not have id (yet?)... */
6840 if (isempty(u
->id
)) {
6841 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6845 if (hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
) < 0)
6848 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6850 if (exec_runtime_allocate(&rt_create
, u
->id
) < 0)
6856 if (streq(key
, "tmp-dir")) {
6857 if (free_and_strdup_warn(&rt
->tmp_dir
, value
) < 0)
6860 } else if (streq(key
, "var-tmp-dir")) {
6861 if (free_and_strdup_warn(&rt
->var_tmp_dir
, value
) < 0)
6864 } else if (streq(key
, "netns-socket-0")) {
6867 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6868 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6872 safe_close(rt
->netns_storage_socket
[0]);
6873 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6875 } else if (streq(key
, "netns-socket-1")) {
6878 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6879 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6883 safe_close(rt
->netns_storage_socket
[1]);
6884 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6889 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6891 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6893 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6897 rt_create
->manager
= u
->manager
;
6900 TAKE_PTR(rt_create
);
6906 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6907 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6909 int r
, netns_fdpair
[] = {-1, -1}, ipcns_fdpair
[] = {-1, -1};
6910 const char *p
, *v
= ASSERT_PTR(value
);
6916 n
= strcspn(v
, " ");
6917 id
= strndupa_safe(v
, n
);
6922 v
= startswith(p
, "tmp-dir=");
6924 n
= strcspn(v
, " ");
6925 tmp_dir
= strndup(v
, n
);
6933 v
= startswith(p
, "var-tmp-dir=");
6935 n
= strcspn(v
, " ");
6936 var_tmp_dir
= strndup(v
, n
);
6944 v
= startswith(p
, "netns-socket-0=");
6948 n
= strcspn(v
, " ");
6949 buf
= strndupa_safe(v
, n
);
6951 r
= safe_atoi(buf
, &netns_fdpair
[0]);
6953 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6954 if (!fdset_contains(fds
, netns_fdpair
[0]))
6955 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6956 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair
[0]);
6957 netns_fdpair
[0] = fdset_remove(fds
, netns_fdpair
[0]);
6963 v
= startswith(p
, "netns-socket-1=");
6967 n
= strcspn(v
, " ");
6968 buf
= strndupa_safe(v
, n
);
6970 r
= safe_atoi(buf
, &netns_fdpair
[1]);
6972 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6973 if (!fdset_contains(fds
, netns_fdpair
[1]))
6974 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6975 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair
[1]);
6976 netns_fdpair
[1] = fdset_remove(fds
, netns_fdpair
[1]);
6982 v
= startswith(p
, "ipcns-socket-0=");
6986 n
= strcspn(v
, " ");
6987 buf
= strndupa_safe(v
, n
);
6989 r
= safe_atoi(buf
, &ipcns_fdpair
[0]);
6991 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf
);
6992 if (!fdset_contains(fds
, ipcns_fdpair
[0]))
6993 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6994 "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair
[0]);
6995 ipcns_fdpair
[0] = fdset_remove(fds
, ipcns_fdpair
[0]);
7001 v
= startswith(p
, "ipcns-socket-1=");
7005 n
= strcspn(v
, " ");
7006 buf
= strndupa_safe(v
, n
);
7008 r
= safe_atoi(buf
, &ipcns_fdpair
[1]);
7010 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf
);
7011 if (!fdset_contains(fds
, ipcns_fdpair
[1]))
7012 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7013 "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair
[1]);
7014 ipcns_fdpair
[1] = fdset_remove(fds
, ipcns_fdpair
[1]);
7018 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_fdpair
, ipcns_fdpair
, NULL
);
7020 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
7024 void exec_runtime_vacuum(Manager
*m
) {
7029 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
7031 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
7035 (void) exec_runtime_free(rt
, false);
7039 void exec_params_clear(ExecParameters
*p
) {
7043 p
->environment
= strv_free(p
->environment
);
7044 p
->fd_names
= strv_free(p
->fd_names
);
7045 p
->fds
= mfree(p
->fds
);
7046 p
->exec_fd
= safe_close(p
->exec_fd
);
7049 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
7058 ExecLoadCredential
*exec_load_credential_free(ExecLoadCredential
*lc
) {
7067 void exec_directory_done(ExecDirectory
*d
) {
7071 for (size_t i
= 0; i
< d
->n_items
; i
++) {
7072 free(d
->items
[i
].path
);
7073 strv_free(d
->items
[i
].symlinks
);
7076 d
->items
= mfree(d
->items
);
7081 static ExecDirectoryItem
*exec_directory_find(ExecDirectory
*d
, const char *path
) {
7085 for (size_t i
= 0; i
< d
->n_items
; i
++)
7086 if (path_equal(d
->items
[i
].path
, path
))
7087 return &d
->items
[i
];
7092 int exec_directory_add(ExecDirectory
*d
, const char *path
, const char *symlink
) {
7093 _cleanup_strv_free_
char **s
= NULL
;
7094 _cleanup_free_
char *p
= NULL
;
7095 ExecDirectoryItem
*existing
;
7101 existing
= exec_directory_find(d
, path
);
7103 r
= strv_extend(&existing
->symlinks
, symlink
);
7107 return 0; /* existing item is updated */
7115 s
= strv_new(symlink
);
7120 if (!GREEDY_REALLOC(d
->items
, d
->n_items
+ 1))
7123 d
->items
[d
->n_items
++] = (ExecDirectoryItem
) {
7124 .path
= TAKE_PTR(p
),
7125 .symlinks
= TAKE_PTR(s
),
7128 return 1; /* new item is added */
7131 static int exec_directory_item_compare_func(const ExecDirectoryItem
*a
, const ExecDirectoryItem
*b
) {
7135 return path_compare(a
->path
, b
->path
);
7138 void exec_directory_sort(ExecDirectory
*d
) {
7141 /* Sort the exec directories to make always parent directories processed at first in
7142 * setup_exec_directory(), e.g., even if StateDirectory=foo/bar foo, we need to create foo at first,
7143 * then foo/bar. Also, set .only_create flag if one of the parent directories is contained in the
7144 * list. See also comments in setup_exec_directory() and issue #24783. */
7146 if (d
->n_items
<= 1)
7149 typesafe_qsort(d
->items
, d
->n_items
, exec_directory_item_compare_func
);
7151 for (size_t i
= 1; i
< d
->n_items
; i
++)
7152 for (size_t j
= 0; j
< i
; j
++)
7153 if (path_startswith(d
->items
[i
].path
, d
->items
[j
].path
)) {
7154 d
->items
[i
].only_create
= true;
7159 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
7160 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_load_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecLoadCredential
, exec_load_credential_free
);
7162 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
7163 [EXEC_INPUT_NULL
] = "null",
7164 [EXEC_INPUT_TTY
] = "tty",
7165 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
7166 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
7167 [EXEC_INPUT_SOCKET
] = "socket",
7168 [EXEC_INPUT_NAMED_FD
] = "fd",
7169 [EXEC_INPUT_DATA
] = "data",
7170 [EXEC_INPUT_FILE
] = "file",
7173 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
7175 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
7176 [EXEC_OUTPUT_INHERIT
] = "inherit",
7177 [EXEC_OUTPUT_NULL
] = "null",
7178 [EXEC_OUTPUT_TTY
] = "tty",
7179 [EXEC_OUTPUT_KMSG
] = "kmsg",
7180 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
7181 [EXEC_OUTPUT_JOURNAL
] = "journal",
7182 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
7183 [EXEC_OUTPUT_SOCKET
] = "socket",
7184 [EXEC_OUTPUT_NAMED_FD
] = "fd",
7185 [EXEC_OUTPUT_FILE
] = "file",
7186 [EXEC_OUTPUT_FILE_APPEND
] = "append",
7187 [EXEC_OUTPUT_FILE_TRUNCATE
] = "truncate",
7190 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
7192 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
7193 [EXEC_UTMP_INIT
] = "init",
7194 [EXEC_UTMP_LOGIN
] = "login",
7195 [EXEC_UTMP_USER
] = "user",
7198 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
7200 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
7201 [EXEC_PRESERVE_NO
] = "no",
7202 [EXEC_PRESERVE_YES
] = "yes",
7203 [EXEC_PRESERVE_RESTART
] = "restart",
7206 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
7208 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
7209 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7210 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
7211 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
7212 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
7213 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
7214 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
7217 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
7219 /* This table maps ExecDirectoryType to the symlink setting it is configured with in the unit */
7220 static const char* const exec_directory_type_symlink_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7221 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectorySymlink",
7222 [EXEC_DIRECTORY_STATE
] = "StateDirectorySymlink",
7223 [EXEC_DIRECTORY_CACHE
] = "CacheDirectorySymlink",
7224 [EXEC_DIRECTORY_LOGS
] = "LogsDirectorySymlink",
7225 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectorySymlink",
7228 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type_symlink
, ExecDirectoryType
);
7230 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
7231 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
7232 * directories, specifically .timer units with their timestamp touch file. */
7233 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7234 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
7235 [EXEC_DIRECTORY_STATE
] = "state",
7236 [EXEC_DIRECTORY_CACHE
] = "cache",
7237 [EXEC_DIRECTORY_LOGS
] = "logs",
7238 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
7241 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
7243 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
7244 * the service payload in. */
7245 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
7246 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
7247 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
7248 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
7249 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
7250 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
7253 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
7255 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
7256 [EXEC_KEYRING_INHERIT
] = "inherit",
7257 [EXEC_KEYRING_PRIVATE
] = "private",
7258 [EXEC_KEYRING_SHARED
] = "shared",
7261 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);