1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <sys/capability.h>
11 #include <sys/eventfd.h>
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
16 #include <sys/socket.h>
18 #include <sys/types.h>
24 #include <security/pam_appl.h>
28 #include <selinux/selinux.h>
36 #include <sys/apparmor.h>
39 #include "sd-messages.h"
42 #include "alloc-util.h"
44 #include "apparmor-util.h"
49 #include "capability-util.h"
50 #include "chown-recursive.h"
51 #include "cpu-set-util.h"
55 #include "errno-list.h"
57 #include "exit-status.h"
59 #include "format-util.h"
61 #include "glob-util.h"
70 #include "namespace.h"
71 #include "parse-util.h"
72 #include "path-util.h"
73 #include "process-util.h"
74 #include "rlimit-util.h"
77 #include "seccomp-util.h"
79 #include "securebits-util.h"
80 #include "selinux-util.h"
81 #include "signal-util.h"
82 #include "smack-util.h"
83 #include "socket-util.h"
85 #include "stat-util.h"
86 #include "string-table.h"
87 #include "string-util.h"
89 #include "syslog-util.h"
90 #include "terminal-util.h"
91 #include "umask-util.h"
93 #include "user-util.h"
95 #include "utmp-wtmp.h"
97 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
98 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
100 /* This assumes there is a 'tty' group */
101 #define TTY_MODE 0620
103 #define SNDBUF_SIZE (8*1024*1024)
105 static int shift_fds(int fds
[], size_t n_fds
) {
106 int start
, restart_from
;
111 /* Modifies the fds array! (sorts it) */
121 for (i
= start
; i
< (int) n_fds
; i
++) {
124 /* Already at right index? */
128 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
135 /* Hmm, the fd we wanted isn't free? Then
136 * let's remember that and try again from here */
137 if (nfd
!= i
+3 && restart_from
< 0)
141 if (restart_from
< 0)
144 start
= restart_from
;
150 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
154 n_fds
= n_socket_fds
+ n_storage_fds
;
160 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
161 * O_NONBLOCK only applies to socket activation though. */
163 for (i
= 0; i
< n_fds
; i
++) {
165 if (i
< n_socket_fds
) {
166 r
= fd_nonblock(fds
[i
], nonblock
);
171 /* We unconditionally drop FD_CLOEXEC from the fds,
172 * since after all we want to pass these fds to our
175 r
= fd_cloexec(fds
[i
], false);
183 static const char *exec_context_tty_path(const ExecContext
*context
) {
186 if (context
->stdio_as_fds
)
189 if (context
->tty_path
)
190 return context
->tty_path
;
192 return "/dev/console";
195 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
200 path
= exec_context_tty_path(context
);
202 if (context
->tty_vhangup
) {
203 if (p
&& p
->stdin_fd
>= 0)
204 (void) terminal_vhangup_fd(p
->stdin_fd
);
206 (void) terminal_vhangup(path
);
209 if (context
->tty_reset
) {
210 if (p
&& p
->stdin_fd
>= 0)
211 (void) reset_terminal_fd(p
->stdin_fd
, true);
213 (void) reset_terminal(path
);
216 if (context
->tty_vt_disallocate
&& path
)
217 (void) vt_disallocate(path
);
220 static bool is_terminal_input(ExecInput i
) {
223 EXEC_INPUT_TTY_FORCE
,
224 EXEC_INPUT_TTY_FAIL
);
227 static bool is_terminal_output(ExecOutput o
) {
230 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
231 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
232 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
235 static bool is_syslog_output(ExecOutput o
) {
238 EXEC_OUTPUT_SYSLOG_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(int fd
, uid_t uid
, gid_t gid
) {
277 static const union sockaddr_union sa
= {
278 .un
.sun_family
= AF_UNIX
,
279 .un
.sun_path
= "/run/systemd/journal/stdout",
281 uid_t olduid
= UID_INVALID
;
282 gid_t oldgid
= GID_INVALID
;
285 if (gid_is_valid(gid
)) {
288 if (setegid(gid
) < 0)
292 if (uid_is_valid(uid
)) {
295 if (seteuid(uid
) < 0) {
301 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
303 /* If we fail to restore the uid or gid, things will likely
304 fail later on. This should only happen if an LSM interferes. */
306 if (uid_is_valid(uid
))
307 (void) seteuid(olduid
);
310 if (gid_is_valid(gid
))
311 (void) setegid(oldgid
);
316 static int connect_logger_as(
318 const ExecContext
*context
,
319 const ExecParameters
*params
,
326 _cleanup_close_
int fd
= -1;
331 assert(output
< _EXEC_OUTPUT_MAX
);
335 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
339 r
= connect_journal_socket(fd
, uid
, gid
);
343 if (shutdown(fd
, SHUT_RD
) < 0)
346 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
356 context
->syslog_identifier
?: ident
,
357 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
358 context
->syslog_priority
,
359 !!context
->syslog_level_prefix
,
360 is_syslog_output(output
),
361 is_kmsg_output(output
),
362 is_terminal_output(output
)) < 0)
365 return move_fd(TAKE_FD(fd
), nfd
, false);
368 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
374 fd
= open_terminal(path
, flags
| O_NOCTTY
);
378 return move_fd(fd
, nfd
, false);
381 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
382 union sockaddr_union sa
= {};
383 _cleanup_close_
int fd
= -1;
388 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
391 fd
= open(path
, flags
|O_NOCTTY
, mode
);
395 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
397 if (strlen(path
) >= sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
400 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
402 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
406 salen
= sockaddr_un_set_path(&sa
.un
, path
);
410 if (connect(fd
, &sa
.sa
, salen
) < 0)
411 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
412 * indication that his wasn't an AF_UNIX socket after all */
414 if ((flags
& O_ACCMODE
) == O_RDONLY
)
415 r
= shutdown(fd
, SHUT_WR
);
416 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
417 r
= shutdown(fd
, SHUT_RD
);
426 static int fixup_input(
427 const ExecContext
*context
,
429 bool apply_tty_stdin
) {
435 std_input
= context
->std_input
;
437 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
438 return EXEC_INPUT_NULL
;
440 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
441 return EXEC_INPUT_NULL
;
443 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
444 return EXEC_INPUT_NULL
;
449 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
451 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
452 return EXEC_OUTPUT_INHERIT
;
457 static int setup_input(
458 const ExecContext
*context
,
459 const ExecParameters
*params
,
461 int named_iofds
[3]) {
468 if (params
->stdin_fd
>= 0) {
469 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
472 /* Try to make this the controlling tty, if it is a tty, and reset it */
473 if (isatty(STDIN_FILENO
)) {
474 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
475 (void) reset_terminal_fd(STDIN_FILENO
, true);
481 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
485 case EXEC_INPUT_NULL
:
486 return open_null_as(O_RDONLY
, STDIN_FILENO
);
489 case EXEC_INPUT_TTY_FORCE
:
490 case EXEC_INPUT_TTY_FAIL
: {
493 fd
= acquire_terminal(exec_context_tty_path(context
),
494 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
495 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
496 ACQUIRE_TERMINAL_WAIT
,
501 return move_fd(fd
, STDIN_FILENO
, false);
504 case EXEC_INPUT_SOCKET
:
505 assert(socket_fd
>= 0);
507 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
509 case EXEC_INPUT_NAMED_FD
:
510 assert(named_iofds
[STDIN_FILENO
] >= 0);
512 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
513 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
515 case EXEC_INPUT_DATA
: {
518 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
522 return move_fd(fd
, STDIN_FILENO
, false);
525 case EXEC_INPUT_FILE
: {
529 assert(context
->stdio_file
[STDIN_FILENO
]);
531 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
532 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
534 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
538 return move_fd(fd
, STDIN_FILENO
, false);
542 assert_not_reached("Unknown input type");
546 static bool can_inherit_stderr_from_stdout(
547 const ExecContext
*context
,
553 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
556 if (e
== EXEC_OUTPUT_INHERIT
)
561 if (e
== EXEC_OUTPUT_NAMED_FD
)
562 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
564 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
565 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
570 static int setup_output(
572 const ExecContext
*context
,
573 const ExecParameters
*params
,
580 dev_t
*journal_stream_dev
,
581 ino_t
*journal_stream_ino
) {
591 assert(journal_stream_dev
);
592 assert(journal_stream_ino
);
594 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
596 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
599 return STDOUT_FILENO
;
602 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
603 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
606 return STDERR_FILENO
;
609 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
610 o
= fixup_output(context
->std_output
, socket_fd
);
612 if (fileno
== STDERR_FILENO
) {
614 e
= fixup_output(context
->std_error
, socket_fd
);
616 /* This expects the input and output are already set up */
618 /* Don't change the stderr file descriptor if we inherit all
619 * the way and are not on a tty */
620 if (e
== EXEC_OUTPUT_INHERIT
&&
621 o
== EXEC_OUTPUT_INHERIT
&&
622 i
== EXEC_INPUT_NULL
&&
623 !is_terminal_input(context
->std_input
) &&
627 /* Duplicate from stdout if possible */
628 if (can_inherit_stderr_from_stdout(context
, o
, e
))
629 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
633 } else if (o
== EXEC_OUTPUT_INHERIT
) {
634 /* If input got downgraded, inherit the original value */
635 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
636 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
638 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
639 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
640 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
642 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
646 /* We need to open /dev/null here anew, to get the right access mode. */
647 return open_null_as(O_WRONLY
, fileno
);
652 case EXEC_OUTPUT_NULL
:
653 return open_null_as(O_WRONLY
, fileno
);
655 case EXEC_OUTPUT_TTY
:
656 if (is_terminal_input(i
))
657 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
659 /* We don't reset the terminal if this is just about output */
660 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
662 case EXEC_OUTPUT_SYSLOG
:
663 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
664 case EXEC_OUTPUT_KMSG
:
665 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
666 case EXEC_OUTPUT_JOURNAL
:
667 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
668 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
670 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
671 r
= open_null_as(O_WRONLY
, fileno
);
675 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
676 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
677 * services to detect whether they are connected to the journal or not.
679 * If both stdout and stderr are connected to a stream then let's make sure to store the data
680 * about STDERR as that's usually the best way to do logging. */
682 if (fstat(fileno
, &st
) >= 0 &&
683 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
684 *journal_stream_dev
= st
.st_dev
;
685 *journal_stream_ino
= st
.st_ino
;
690 case EXEC_OUTPUT_SOCKET
:
691 assert(socket_fd
>= 0);
693 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
695 case EXEC_OUTPUT_NAMED_FD
:
696 assert(named_iofds
[fileno
] >= 0);
698 (void) fd_nonblock(named_iofds
[fileno
], false);
699 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
701 case EXEC_OUTPUT_FILE
:
702 case EXEC_OUTPUT_FILE_APPEND
: {
706 assert(context
->stdio_file
[fileno
]);
708 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
709 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
712 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
715 if (o
== EXEC_OUTPUT_FILE_APPEND
)
718 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
722 return move_fd(fd
, fileno
, 0);
726 assert_not_reached("Unknown error type");
730 static int chown_terminal(int fd
, uid_t uid
) {
735 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
739 /* This might fail. What matters are the results. */
740 (void) fchown(fd
, uid
, -1);
741 (void) fchmod(fd
, TTY_MODE
);
743 if (fstat(fd
, &st
) < 0)
746 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
752 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
753 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
756 assert(_saved_stdin
);
757 assert(_saved_stdout
);
759 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
763 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
764 if (saved_stdout
< 0)
767 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
771 r
= chown_terminal(fd
, getuid());
775 r
= reset_terminal_fd(fd
, true);
779 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
784 *_saved_stdin
= saved_stdin
;
785 *_saved_stdout
= saved_stdout
;
787 saved_stdin
= saved_stdout
= -1;
792 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
795 if (err
== -ETIMEDOUT
)
796 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
799 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
803 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
804 _cleanup_close_
int fd
= -1;
808 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
812 write_confirm_error_fd(err
, fd
, u
);
815 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
819 assert(saved_stdout
);
823 if (*saved_stdin
>= 0)
824 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
827 if (*saved_stdout
>= 0)
828 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
831 *saved_stdin
= safe_close(*saved_stdin
);
832 *saved_stdout
= safe_close(*saved_stdout
);
838 CONFIRM_PRETEND_FAILURE
= -1,
839 CONFIRM_PRETEND_SUCCESS
= 0,
843 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
844 int saved_stdout
= -1, saved_stdin
= -1, r
;
845 _cleanup_free_
char *e
= NULL
;
848 /* For any internal errors, assume a positive response. */
849 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
851 write_confirm_error(r
, vc
, u
);
852 return CONFIRM_EXECUTE
;
855 /* confirm_spawn might have been disabled while we were sleeping. */
856 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
861 e
= ellipsize(cmdline
, 60, 100);
869 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
871 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
878 printf("Resuming normal execution.\n");
879 manager_disable_confirm_spawn();
883 unit_dump(u
, stdout
, " ");
884 continue; /* ask again */
886 printf("Failing execution.\n");
887 r
= CONFIRM_PRETEND_FAILURE
;
890 printf(" c - continue, proceed without asking anymore\n"
891 " D - dump, show the state of the unit\n"
892 " f - fail, don't execute the command and pretend it failed\n"
894 " i - info, show a short summary of the unit\n"
895 " j - jobs, show jobs that are in progress\n"
896 " s - skip, don't execute the command and pretend it succeeded\n"
897 " y - yes, execute the command\n");
898 continue; /* ask again */
900 printf(" Description: %s\n"
903 u
->id
, u
->description
, cmdline
);
904 continue; /* ask again */
906 manager_dump_jobs(u
->manager
, stdout
, " ");
907 continue; /* ask again */
909 /* 'n' was removed in favor of 'f'. */
910 printf("Didn't understand 'n', did you mean 'f'?\n");
911 continue; /* ask again */
913 printf("Skipping execution.\n");
914 r
= CONFIRM_PRETEND_SUCCESS
;
920 assert_not_reached("Unhandled choice");
926 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
930 static int get_fixed_user(const ExecContext
*c
, const char **user
,
931 uid_t
*uid
, gid_t
*gid
,
932 const char **home
, const char **shell
) {
941 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
942 * (i.e. are "/" or "/bin/nologin"). */
945 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
953 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
963 r
= get_group_creds(&name
, gid
, 0);
971 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
972 const char *group
, gid_t gid
,
973 gid_t
**supplementary_gids
, int *ngids
) {
977 bool keep_groups
= false;
978 gid_t
*groups
= NULL
;
979 _cleanup_free_ gid_t
*l_gids
= NULL
;
984 * If user is given, then lookup GID and supplementary groups list.
985 * We avoid NSS lookups for gid=0. Also we have to initialize groups
986 * here and as early as possible so we keep the list of supplementary
987 * groups of the caller.
989 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
990 /* First step, initialize groups from /etc/groups */
991 if (initgroups(user
, gid
) < 0)
997 if (strv_isempty(c
->supplementary_groups
))
1001 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1002 * be positive, otherwise fail.
1005 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1006 if (ngroups_max
<= 0) {
1010 return -EOPNOTSUPP
; /* For all other values */
1013 l_gids
= new(gid_t
, ngroups_max
);
1019 * Lookup the list of groups that the user belongs to, we
1020 * avoid NSS lookups here too for gid=0.
1023 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1028 STRV_FOREACH(i
, c
->supplementary_groups
) {
1031 if (k
>= ngroups_max
)
1035 r
= get_group_creds(&g
, l_gids
+k
, 0);
1043 * Sets ngids to zero to drop all supplementary groups, happens
1044 * when we are under root and SupplementaryGroups= is empty.
1051 /* Otherwise get the final list of supplementary groups */
1052 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1056 *supplementary_gids
= groups
;
1064 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1067 /* Handle SupplementaryGroups= if it is not empty */
1069 r
= maybe_setgroups(ngids
, supplementary_gids
);
1074 if (gid_is_valid(gid
)) {
1075 /* Then set our gids */
1076 if (setresgid(gid
, gid
, gid
) < 0)
1083 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1086 if (!uid_is_valid(uid
))
1089 /* Sets (but doesn't look up) the uid and make sure we keep the
1090 * capabilities while doing so. */
1092 if (context
->capability_ambient_set
!= 0) {
1094 /* First step: If we need to keep capabilities but
1095 * drop privileges we need to make sure we keep our
1096 * caps, while we drop privileges. */
1098 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1100 if (prctl(PR_GET_SECUREBITS
) != sb
)
1101 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1106 /* Second step: actually set the uids */
1107 if (setresuid(uid
, uid
, uid
) < 0)
1110 /* At this point we should have all necessary capabilities but
1111 are otherwise a normal user. However, the caps might got
1112 corrupted due to the setresuid() so we need clean them up
1113 later. This is done outside of this call. */
1120 static int null_conv(
1122 const struct pam_message
**msg
,
1123 struct pam_response
**resp
,
1124 void *appdata_ptr
) {
1126 /* We don't support conversations */
1128 return PAM_CONV_ERR
;
1133 static int setup_pam(
1140 int fds
[], size_t n_fds
) {
1144 static const struct pam_conv conv
= {
1149 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1150 pam_handle_t
*handle
= NULL
;
1152 int pam_code
= PAM_SUCCESS
, r
;
1153 char **nv
, **e
= NULL
;
1154 bool close_session
= false;
1155 pid_t pam_pid
= 0, parent_pid
;
1162 /* We set up PAM in the parent process, then fork. The child
1163 * will then stay around until killed via PR_GET_PDEATHSIG or
1164 * systemd via the cgroup logic. It will then remove the PAM
1165 * session again. The parent process will exec() the actual
1166 * daemon. We do things this way to ensure that the main PID
1167 * of the daemon is the one we initially fork()ed. */
1169 r
= barrier_create(&barrier
);
1173 if (log_get_max_level() < LOG_DEBUG
)
1174 flags
|= PAM_SILENT
;
1176 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1177 if (pam_code
!= PAM_SUCCESS
) {
1183 _cleanup_free_
char *q
= NULL
;
1185 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1186 * out if that's the case, and read the TTY off it. */
1188 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1189 tty
= strjoina("/dev/", q
);
1193 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1194 if (pam_code
!= PAM_SUCCESS
)
1198 STRV_FOREACH(nv
, *env
) {
1199 pam_code
= pam_putenv(handle
, *nv
);
1200 if (pam_code
!= PAM_SUCCESS
)
1204 pam_code
= pam_acct_mgmt(handle
, flags
);
1205 if (pam_code
!= PAM_SUCCESS
)
1208 pam_code
= pam_open_session(handle
, flags
);
1209 if (pam_code
!= PAM_SUCCESS
)
1212 close_session
= true;
1214 e
= pam_getenvlist(handle
);
1216 pam_code
= PAM_BUF_ERR
;
1220 /* Block SIGTERM, so that we know that it won't get lost in
1223 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1225 parent_pid
= getpid_cached();
1227 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1231 int sig
, ret
= EXIT_PAM
;
1233 /* The child's job is to reset the PAM session on
1235 barrier_set_role(&barrier
, BARRIER_CHILD
);
1237 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1238 * are open here that have been opened by PAM. */
1239 (void) close_many(fds
, n_fds
);
1241 /* Drop privileges - we don't need any to pam_close_session
1242 * and this will make PR_SET_PDEATHSIG work in most cases.
1243 * If this fails, ignore the error - but expect sd-pam threads
1244 * to fail to exit normally */
1246 r
= maybe_setgroups(0, NULL
);
1248 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1249 if (setresgid(gid
, gid
, gid
) < 0)
1250 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1251 if (setresuid(uid
, uid
, uid
) < 0)
1252 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1254 (void) ignore_signals(SIGPIPE
, -1);
1256 /* Wait until our parent died. This will only work if
1257 * the above setresuid() succeeds, otherwise the kernel
1258 * will not allow unprivileged parents kill their privileged
1259 * children this way. We rely on the control groups kill logic
1260 * to do the rest for us. */
1261 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1264 /* Tell the parent that our setup is done. This is especially
1265 * important regarding dropping privileges. Otherwise, unit
1266 * setup might race against our setresuid(2) call.
1268 * If the parent aborted, we'll detect this below, hence ignore
1269 * return failure here. */
1270 (void) barrier_place(&barrier
);
1272 /* Check if our parent process might already have died? */
1273 if (getppid() == parent_pid
) {
1276 assert_se(sigemptyset(&ss
) >= 0);
1277 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1280 if (sigwait(&ss
, &sig
) < 0) {
1287 assert(sig
== SIGTERM
);
1292 /* If our parent died we'll end the session */
1293 if (getppid() != parent_pid
) {
1294 pam_code
= pam_close_session(handle
, flags
);
1295 if (pam_code
!= PAM_SUCCESS
)
1302 pam_end(handle
, pam_code
| flags
);
1306 barrier_set_role(&barrier
, BARRIER_PARENT
);
1308 /* If the child was forked off successfully it will do all the
1309 * cleanups, so forget about the handle here. */
1312 /* Unblock SIGTERM again in the parent */
1313 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1315 /* We close the log explicitly here, since the PAM modules
1316 * might have opened it, but we don't want this fd around. */
1319 /* Synchronously wait for the child to initialize. We don't care for
1320 * errors as we cannot recover. However, warn loudly if it happens. */
1321 if (!barrier_place_and_sync(&barrier
))
1322 log_error("PAM initialization failed");
1324 return strv_free_and_replace(*env
, e
);
1327 if (pam_code
!= PAM_SUCCESS
) {
1328 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1329 r
= -EPERM
; /* PAM errors do not map to errno */
1331 log_error_errno(r
, "PAM failed: %m");
1335 pam_code
= pam_close_session(handle
, flags
);
1337 pam_end(handle
, pam_code
| flags
);
1349 static void rename_process_from_path(const char *path
) {
1350 char process_name
[11];
1354 /* This resulting string must fit in 10 chars (i.e. the length
1355 * of "/sbin/init") to look pretty in /bin/ps */
1359 rename_process("(...)");
1365 /* The end of the process name is usually more
1366 * interesting, since the first bit might just be
1372 process_name
[0] = '(';
1373 memcpy(process_name
+1, p
, l
);
1374 process_name
[1+l
] = ')';
1375 process_name
[1+l
+1] = 0;
1377 rename_process(process_name
);
1380 static bool context_has_address_families(const ExecContext
*c
) {
1383 return c
->address_families_whitelist
||
1384 !set_isempty(c
->address_families
);
1387 static bool context_has_syscall_filters(const ExecContext
*c
) {
1390 return c
->syscall_whitelist
||
1391 !hashmap_isempty(c
->syscall_filter
);
1394 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1397 if (c
->no_new_privileges
)
1400 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1403 /* We need NNP if we have any form of seccomp and are unprivileged */
1404 return context_has_address_families(c
) ||
1405 c
->memory_deny_write_execute
||
1406 c
->restrict_realtime
||
1407 exec_context_restrict_namespaces_set(c
) ||
1408 c
->protect_kernel_tunables
||
1409 c
->protect_kernel_modules
||
1410 c
->private_devices
||
1411 context_has_syscall_filters(c
) ||
1412 !set_isempty(c
->syscall_archs
) ||
1413 c
->lock_personality
;
1418 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1420 if (is_seccomp_available())
1423 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1427 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1428 uint32_t negative_action
, default_action
, action
;
1434 if (!context_has_syscall_filters(c
))
1437 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1440 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1442 if (c
->syscall_whitelist
) {
1443 default_action
= negative_action
;
1444 action
= SCMP_ACT_ALLOW
;
1446 default_action
= SCMP_ACT_ALLOW
;
1447 action
= negative_action
;
1450 if (needs_ambient_hack
) {
1451 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1456 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1459 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1463 if (set_isempty(c
->syscall_archs
))
1466 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1469 return seccomp_restrict_archs(c
->syscall_archs
);
1472 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1476 if (!context_has_address_families(c
))
1479 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1482 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1485 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1489 if (!c
->memory_deny_write_execute
)
1492 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1495 return seccomp_memory_deny_write_execute();
1498 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1502 if (!c
->restrict_realtime
)
1505 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1508 return seccomp_restrict_realtime();
1511 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1515 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1516 * let's protect even those systems where this is left on in the kernel. */
1518 if (!c
->protect_kernel_tunables
)
1521 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1524 return seccomp_protect_sysctl();
1527 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1531 /* Turn off module syscalls on ProtectKernelModules=yes */
1533 if (!c
->protect_kernel_modules
)
1536 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1539 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1542 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1546 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1548 if (!c
->private_devices
)
1551 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1554 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1557 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1561 if (!exec_context_restrict_namespaces_set(c
))
1564 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1567 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1570 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1571 unsigned long personality
;
1577 if (!c
->lock_personality
)
1580 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1583 personality
= c
->personality
;
1585 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1586 if (personality
== PERSONALITY_INVALID
) {
1588 r
= opinionated_personality(&personality
);
1593 return seccomp_lock_personality(personality
);
1598 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1601 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1602 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1604 if (idle_pipe
[0] >= 0) {
1607 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1609 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1612 /* Signal systemd that we are bored and want to continue. */
1613 n
= write(idle_pipe
[3], "x", 1);
1615 /* Wait for systemd to react to the signal above. */
1616 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1619 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1623 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1626 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1628 static int build_environment(
1630 const ExecContext
*c
,
1631 const ExecParameters
*p
,
1634 const char *username
,
1636 dev_t journal_stream_dev
,
1637 ino_t journal_stream_ino
,
1640 _cleanup_strv_free_
char **our_env
= NULL
;
1641 ExecDirectoryType t
;
1650 our_env
= new0(char*, 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1655 _cleanup_free_
char *joined
= NULL
;
1657 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1659 our_env
[n_env
++] = x
;
1661 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1663 our_env
[n_env
++] = x
;
1665 joined
= strv_join(p
->fd_names
, ":");
1669 x
= strjoin("LISTEN_FDNAMES=", joined
);
1672 our_env
[n_env
++] = x
;
1675 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1676 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1678 our_env
[n_env
++] = x
;
1680 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1682 our_env
[n_env
++] = x
;
1685 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1686 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1687 * check the database directly. */
1688 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1689 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1692 our_env
[n_env
++] = x
;
1696 x
= strappend("HOME=", home
);
1699 our_env
[n_env
++] = x
;
1703 x
= strappend("LOGNAME=", username
);
1706 our_env
[n_env
++] = x
;
1708 x
= strappend("USER=", username
);
1711 our_env
[n_env
++] = x
;
1715 x
= strappend("SHELL=", shell
);
1718 our_env
[n_env
++] = x
;
1721 if (!sd_id128_is_null(u
->invocation_id
)) {
1722 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1725 our_env
[n_env
++] = x
;
1728 if (exec_context_needs_term(c
)) {
1729 const char *tty_path
, *term
= NULL
;
1731 tty_path
= exec_context_tty_path(c
);
1733 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1734 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1735 * passes to PID 1 ends up all the way in the console login shown. */
1737 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1738 term
= getenv("TERM");
1740 term
= default_term_for_tty(tty_path
);
1742 x
= strappend("TERM=", term
);
1745 our_env
[n_env
++] = x
;
1748 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1749 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1752 our_env
[n_env
++] = x
;
1755 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1756 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1762 if (strv_isempty(c
->directories
[t
].paths
))
1765 n
= exec_directory_env_name_to_string(t
);
1769 pre
= strjoin(p
->prefix
[t
], "/");
1773 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1777 x
= strjoin(n
, "=", joined
);
1781 our_env
[n_env
++] = x
;
1784 our_env
[n_env
++] = NULL
;
1785 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1787 *ret
= TAKE_PTR(our_env
);
1792 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1793 _cleanup_strv_free_
char **pass_env
= NULL
;
1794 size_t n_env
= 0, n_bufsize
= 0;
1797 STRV_FOREACH(i
, c
->pass_environment
) {
1798 _cleanup_free_
char *x
= NULL
;
1804 x
= strjoin(*i
, "=", v
);
1808 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1811 pass_env
[n_env
++] = TAKE_PTR(x
);
1812 pass_env
[n_env
] = NULL
;
1815 *ret
= TAKE_PTR(pass_env
);
1820 static bool exec_needs_mount_namespace(
1821 const ExecContext
*context
,
1822 const ExecParameters
*params
,
1823 const ExecRuntime
*runtime
) {
1828 if (context
->root_image
)
1831 if (!strv_isempty(context
->read_write_paths
) ||
1832 !strv_isempty(context
->read_only_paths
) ||
1833 !strv_isempty(context
->inaccessible_paths
))
1836 if (context
->n_bind_mounts
> 0)
1839 if (context
->n_temporary_filesystems
> 0)
1842 if (context
->mount_flags
!= 0)
1845 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1848 if (context
->private_devices
||
1849 context
->private_mounts
||
1850 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1851 context
->protect_home
!= PROTECT_HOME_NO
||
1852 context
->protect_kernel_tunables
||
1853 context
->protect_kernel_modules
||
1854 context
->protect_control_groups
)
1857 if (context
->root_directory
) {
1858 ExecDirectoryType t
;
1860 if (context
->mount_apivfs
)
1863 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1864 if (!params
->prefix
[t
])
1867 if (!strv_isempty(context
->directories
[t
].paths
))
1872 if (context
->dynamic_user
&&
1873 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1874 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1875 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1881 static int setup_private_users(uid_t uid
, gid_t gid
) {
1882 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1883 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1884 _cleanup_close_
int unshare_ready_fd
= -1;
1885 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1890 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1891 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1892 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1893 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1894 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1895 * continues execution normally. */
1897 if (uid
!= 0 && uid_is_valid(uid
)) {
1898 r
= asprintf(&uid_map
,
1899 "0 0 1\n" /* Map root → root */
1900 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1905 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1910 if (gid
!= 0 && gid_is_valid(gid
)) {
1911 r
= asprintf(&gid_map
,
1912 "0 0 1\n" /* Map root → root */
1913 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1918 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1923 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1925 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1926 if (unshare_ready_fd
< 0)
1929 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1931 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1934 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1938 _cleanup_close_
int fd
= -1;
1942 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1943 * here, after the parent opened its own user namespace. */
1946 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1948 /* Wait until the parent unshared the user namespace */
1949 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1954 /* Disable the setgroups() system call in the child user namespace, for good. */
1955 a
= procfs_file_alloca(ppid
, "setgroups");
1956 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1958 if (errno
!= ENOENT
) {
1963 /* If the file is missing the kernel is too old, let's continue anyway. */
1965 if (write(fd
, "deny\n", 5) < 0) {
1970 fd
= safe_close(fd
);
1973 /* First write the GID map */
1974 a
= procfs_file_alloca(ppid
, "gid_map");
1975 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1980 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1984 fd
= safe_close(fd
);
1986 /* The write the UID map */
1987 a
= procfs_file_alloca(ppid
, "uid_map");
1988 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1993 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1998 _exit(EXIT_SUCCESS
);
2001 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2002 _exit(EXIT_FAILURE
);
2005 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2007 if (unshare(CLONE_NEWUSER
) < 0)
2010 /* Let the child know that the namespace is ready now */
2011 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2014 /* Try to read an error code from the child */
2015 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2018 if (n
== sizeof(r
)) { /* an error code was sent to us */
2023 if (n
!= 0) /* on success we should have read 0 bytes */
2026 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2030 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2036 static int setup_exec_directory(
2037 const ExecContext
*context
,
2038 const ExecParameters
*params
,
2041 ExecDirectoryType type
,
2044 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2045 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2046 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2047 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2048 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2049 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2056 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2057 assert(exit_status
);
2059 if (!params
->prefix
[type
])
2062 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2063 if (!uid_is_valid(uid
))
2065 if (!gid_is_valid(gid
))
2069 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2070 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2072 p
= strjoin(params
->prefix
[type
], "/", *rt
);
2078 r
= mkdir_parents_label(p
, 0755);
2082 if (context
->dynamic_user
&&
2083 !IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2084 _cleanup_free_
char *private_root
= NULL
;
2086 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2087 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2088 * whose UID is later on reused. To lock this down we use the same trick used by container
2089 * managers to prohibit host users to get access to files of the same UID in containers: we
2090 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2091 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2092 * to make this directory permeable for the service itself.
2094 * Specifically: for a service which wants a special directory "foo/" we first create a
2095 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2096 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2097 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2098 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2099 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2100 * disabling the access boundary for the service and making sure it only gets access to the
2101 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2103 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2104 * owned by the service itself.
2105 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2106 * files or sockets with other services. */
2108 private_root
= strjoin(params
->prefix
[type
], "/private");
2109 if (!private_root
) {
2114 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2115 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2119 pp
= strjoin(private_root
, "/", *rt
);
2125 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2126 r
= mkdir_parents_label(pp
, 0755);
2130 if (is_dir(p
, false) > 0 &&
2131 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2133 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2134 * it over. Most likely the service has been upgraded from one that didn't use
2135 * DynamicUser=1, to one that does. */
2137 if (rename(p
, pp
) < 0) {
2142 /* Otherwise, create the actual directory for the service */
2144 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2145 if (r
< 0 && r
!= -EEXIST
)
2149 /* And link it up from the original place */
2150 r
= symlink_idempotent(pp
, p
, true);
2154 /* Lock down the access mode */
2155 if (chmod(pp
, context
->directories
[type
].mode
) < 0) {
2160 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2161 if (r
< 0 && r
!= -EEXIST
)
2163 if (r
== -EEXIST
&& !context
->dynamic_user
)
2167 /* Don't change the owner of the configuration directory, as in the common case it is not written to by
2168 * a service, and shall not be writable. */
2169 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2172 /* Then, change the ownership of the whole tree, if necessary */
2173 r
= path_chown_recursive(pp
?: p
, uid
, gid
);
2181 *exit_status
= exit_status_table
[type
];
2186 static int setup_smack(
2187 const ExecContext
*context
,
2188 const ExecCommand
*command
) {
2195 if (context
->smack_process_label
) {
2196 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2200 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2202 _cleanup_free_
char *exec_label
= NULL
;
2204 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2205 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2208 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2218 static int compile_bind_mounts(
2219 const ExecContext
*context
,
2220 const ExecParameters
*params
,
2221 BindMount
**ret_bind_mounts
,
2222 size_t *ret_n_bind_mounts
,
2223 char ***ret_empty_directories
) {
2225 _cleanup_strv_free_
char **empty_directories
= NULL
;
2226 BindMount
*bind_mounts
;
2228 ExecDirectoryType t
;
2233 assert(ret_bind_mounts
);
2234 assert(ret_n_bind_mounts
);
2235 assert(ret_empty_directories
);
2237 n
= context
->n_bind_mounts
;
2238 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2239 if (!params
->prefix
[t
])
2242 n
+= strv_length(context
->directories
[t
].paths
);
2246 *ret_bind_mounts
= NULL
;
2247 *ret_n_bind_mounts
= 0;
2248 *ret_empty_directories
= NULL
;
2252 bind_mounts
= new(BindMount
, n
);
2256 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2257 BindMount
*item
= context
->bind_mounts
+ i
;
2260 s
= strdup(item
->source
);
2266 d
= strdup(item
->destination
);
2273 bind_mounts
[h
++] = (BindMount
) {
2276 .read_only
= item
->read_only
,
2277 .recursive
= item
->recursive
,
2278 .ignore_enoent
= item
->ignore_enoent
,
2282 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2285 if (!params
->prefix
[t
])
2288 if (strv_isempty(context
->directories
[t
].paths
))
2291 if (context
->dynamic_user
&&
2292 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2293 !(context
->root_directory
|| context
->root_image
)) {
2296 /* So this is for a dynamic user, and we need to make sure the process can access its own
2297 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2298 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2300 private_root
= strjoin(params
->prefix
[t
], "/private");
2301 if (!private_root
) {
2306 r
= strv_consume(&empty_directories
, private_root
);
2311 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2314 if (context
->dynamic_user
&&
2315 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2316 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2318 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2324 if (context
->dynamic_user
&&
2325 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2326 (context
->root_directory
|| context
->root_image
))
2327 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2328 * directory is not created on the root directory. So, let's bind-mount the directory
2329 * on the 'non-private' place. */
2330 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2339 bind_mounts
[h
++] = (BindMount
) {
2344 .ignore_enoent
= false,
2351 *ret_bind_mounts
= bind_mounts
;
2352 *ret_n_bind_mounts
= n
;
2353 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2358 bind_mount_free_many(bind_mounts
, h
);
2362 static int apply_mount_namespace(
2364 const ExecCommand
*command
,
2365 const ExecContext
*context
,
2366 const ExecParameters
*params
,
2367 const ExecRuntime
*runtime
) {
2369 _cleanup_strv_free_
char **empty_directories
= NULL
;
2370 char *tmp
= NULL
, *var
= NULL
;
2371 const char *root_dir
= NULL
, *root_image
= NULL
;
2372 NamespaceInfo ns_info
;
2373 bool needs_sandboxing
;
2374 BindMount
*bind_mounts
= NULL
;
2375 size_t n_bind_mounts
= 0;
2380 /* The runtime struct only contains the parent of the private /tmp,
2381 * which is non-accessible to world users. Inside of it there's a /tmp
2382 * that is sticky, and that's the one we want to use here. */
2384 if (context
->private_tmp
&& runtime
) {
2385 if (runtime
->tmp_dir
)
2386 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2387 if (runtime
->var_tmp_dir
)
2388 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2391 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2392 root_image
= context
->root_image
;
2395 root_dir
= context
->root_directory
;
2398 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2402 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2403 if (needs_sandboxing
)
2404 ns_info
= (NamespaceInfo
) {
2405 .ignore_protect_paths
= false,
2406 .private_dev
= context
->private_devices
,
2407 .protect_control_groups
= context
->protect_control_groups
,
2408 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2409 .protect_kernel_modules
= context
->protect_kernel_modules
,
2410 .mount_apivfs
= context
->mount_apivfs
,
2411 .private_mounts
= context
->private_mounts
,
2413 else if (!context
->dynamic_user
&& root_dir
)
2415 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2416 * sandbox info, otherwise enforce it, don't ignore protected paths and
2417 * fail if we are enable to apply the sandbox inside the mount namespace.
2419 ns_info
= (NamespaceInfo
) {
2420 .ignore_protect_paths
= true,
2423 ns_info
= (NamespaceInfo
) {};
2425 r
= setup_namespace(root_dir
, root_image
,
2426 &ns_info
, context
->read_write_paths
,
2427 needs_sandboxing
? context
->read_only_paths
: NULL
,
2428 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2432 context
->temporary_filesystems
,
2433 context
->n_temporary_filesystems
,
2436 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2437 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2438 context
->mount_flags
,
2439 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2441 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2443 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2444 * that with a special, recognizable error ENOANO. In this case, silently proceeed, but only if exclusively
2445 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2446 * completely different execution environment. */
2448 if (n_bind_mounts
== 0 &&
2449 context
->n_temporary_filesystems
== 0 &&
2450 !root_dir
&& !root_image
&&
2451 !context
->dynamic_user
) {
2452 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2456 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2457 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2458 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2466 static int apply_working_directory(
2467 const ExecContext
*context
,
2468 const ExecParameters
*params
,
2470 const bool needs_mount_ns
,
2476 assert(exit_status
);
2478 if (context
->working_directory_home
) {
2481 *exit_status
= EXIT_CHDIR
;
2487 } else if (context
->working_directory
)
2488 wd
= context
->working_directory
;
2492 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2493 if (!needs_mount_ns
&& context
->root_directory
)
2494 if (chroot(context
->root_directory
) < 0) {
2495 *exit_status
= EXIT_CHROOT
;
2501 d
= prefix_roota(context
->root_directory
, wd
);
2503 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2504 *exit_status
= EXIT_CHDIR
;
2511 static int setup_keyring(
2513 const ExecContext
*context
,
2514 const ExecParameters
*p
,
2515 uid_t uid
, gid_t gid
) {
2517 key_serial_t keyring
;
2526 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2527 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2528 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2529 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2530 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2531 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2533 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2536 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2539 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2540 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2541 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2542 * & group is just as nasty as acquiring a reference to the user keyring. */
2544 saved_uid
= getuid();
2545 saved_gid
= getgid();
2547 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2548 if (setregid(gid
, -1) < 0)
2549 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2552 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2553 if (setreuid(uid
, -1) < 0) {
2554 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2559 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2560 if (keyring
== -1) {
2561 if (errno
== ENOSYS
)
2562 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2563 else if (IN_SET(errno
, EACCES
, EPERM
))
2564 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2565 else if (errno
== EDQUOT
)
2566 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2568 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2573 /* When requested link the user keyring into the session keyring. */
2574 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2576 if (keyctl(KEYCTL_LINK
,
2577 KEY_SPEC_USER_KEYRING
,
2578 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2579 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2584 /* Restore uid/gid back */
2585 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2586 if (setreuid(saved_uid
, -1) < 0) {
2587 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2592 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2593 if (setregid(saved_gid
, -1) < 0)
2594 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2597 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2598 if (!sd_id128_is_null(u
->invocation_id
)) {
2601 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2603 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2605 if (keyctl(KEYCTL_SETPERM
, key
,
2606 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2607 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2608 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2613 /* Revert back uid & gid for the the last time, and exit */
2614 /* no extra logging, as only the first already reported error matters */
2615 if (getuid() != saved_uid
)
2616 (void) setreuid(saved_uid
, -1);
2618 if (getgid() != saved_gid
)
2619 (void) setregid(saved_gid
, -1);
2624 static void append_socket_pair(int *array
, size_t *n
, const int pair
[2]) {
2632 array
[(*n
)++] = pair
[0];
2634 array
[(*n
)++] = pair
[1];
2637 static int close_remaining_fds(
2638 const ExecParameters
*params
,
2639 const ExecRuntime
*runtime
,
2640 const DynamicCreds
*dcreds
,
2644 int *fds
, size_t n_fds
) {
2646 size_t n_dont_close
= 0;
2647 int dont_close
[n_fds
+ 12];
2651 if (params
->stdin_fd
>= 0)
2652 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2653 if (params
->stdout_fd
>= 0)
2654 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2655 if (params
->stderr_fd
>= 0)
2656 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2659 dont_close
[n_dont_close
++] = socket_fd
;
2661 dont_close
[n_dont_close
++] = exec_fd
;
2663 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2664 n_dont_close
+= n_fds
;
2668 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2672 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2674 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2677 if (user_lookup_fd
>= 0)
2678 dont_close
[n_dont_close
++] = user_lookup_fd
;
2680 return close_all_fds(dont_close
, n_dont_close
);
2683 static int send_user_lookup(
2691 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2692 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2695 if (user_lookup_fd
< 0)
2698 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2701 if (writev(user_lookup_fd
,
2703 IOVEC_INIT(&uid
, sizeof(uid
)),
2704 IOVEC_INIT(&gid
, sizeof(gid
)),
2705 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2711 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2718 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2723 if (!c
->working_directory_home
)
2727 /* Hardcode /root as home directory for UID 0 */
2732 r
= get_home_dir(buf
);
2740 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2741 _cleanup_strv_free_
char ** list
= NULL
;
2742 ExecDirectoryType t
;
2749 assert(c
->dynamic_user
);
2751 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2752 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2755 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2758 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2764 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2767 if (t
== EXEC_DIRECTORY_RUNTIME
)
2768 e
= strjoin(p
->prefix
[t
], "/", *i
);
2770 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2774 r
= strv_consume(&list
, e
);
2780 *ret
= TAKE_PTR(list
);
2785 static char *exec_command_line(char **argv
);
2787 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2788 bool using_subcgroup
;
2794 if (!params
->cgroup_path
)
2797 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2798 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2799 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2800 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2801 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2802 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2803 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2804 * flag, which is only passed for the former statements, not for the latter. */
2806 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2807 if (using_subcgroup
)
2808 p
= strjoin(params
->cgroup_path
, "/.control");
2810 p
= strdup(params
->cgroup_path
);
2815 return using_subcgroup
;
2818 static int exec_child(
2820 const ExecCommand
*command
,
2821 const ExecContext
*context
,
2822 const ExecParameters
*params
,
2823 ExecRuntime
*runtime
,
2824 DynamicCreds
*dcreds
,
2828 size_t n_socket_fds
,
2829 size_t n_storage_fds
,
2834 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2835 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2836 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2837 const char *username
= NULL
, *groupname
= NULL
;
2838 _cleanup_free_
char *home_buffer
= NULL
;
2839 const char *home
= NULL
, *shell
= NULL
;
2840 dev_t journal_stream_dev
= 0;
2841 ino_t journal_stream_ino
= 0;
2842 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2843 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2844 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2845 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2847 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2848 bool use_selinux
= false;
2851 bool use_smack
= false;
2854 bool use_apparmor
= false;
2856 uid_t uid
= UID_INVALID
;
2857 gid_t gid
= GID_INVALID
;
2859 ExecDirectoryType dt
;
2866 assert(exit_status
);
2868 rename_process_from_path(command
->path
);
2870 /* We reset exactly these signals, since they are the
2871 * only ones we set to SIG_IGN in the main daemon. All
2872 * others we leave untouched because we set them to
2873 * SIG_DFL or a valid handler initially, both of which
2874 * will be demoted to SIG_DFL. */
2875 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2876 SIGNALS_IGNORE
, -1);
2878 if (context
->ignore_sigpipe
)
2879 (void) ignore_signals(SIGPIPE
, -1);
2881 r
= reset_signal_mask();
2883 *exit_status
= EXIT_SIGNAL_MASK
;
2884 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2887 if (params
->idle_pipe
)
2888 do_idle_pipe_dance(params
->idle_pipe
);
2890 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2891 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2892 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2893 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2896 log_set_open_when_needed(true);
2898 /* In case anything used libc syslog(), close this here, too */
2901 n_fds
= n_socket_fds
+ n_storage_fds
;
2902 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2904 *exit_status
= EXIT_FDS
;
2905 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2908 if (!context
->same_pgrp
)
2910 *exit_status
= EXIT_SETSID
;
2911 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2914 exec_context_tty_reset(context
, params
);
2916 if (unit_shall_confirm_spawn(unit
)) {
2917 const char *vc
= params
->confirm_spawn
;
2918 _cleanup_free_
char *cmdline
= NULL
;
2920 cmdline
= exec_command_line(command
->argv
);
2922 *exit_status
= EXIT_MEMORY
;
2926 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2927 if (r
!= CONFIRM_EXECUTE
) {
2928 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2929 *exit_status
= EXIT_SUCCESS
;
2932 *exit_status
= EXIT_CONFIRM
;
2933 log_unit_error(unit
, "Execution cancelled by the user");
2938 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2939 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2940 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2941 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2942 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2943 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2944 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2945 *exit_status
= EXIT_MEMORY
;
2946 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2949 if (context
->dynamic_user
&& dcreds
) {
2950 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2952 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2953 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2954 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2955 *exit_status
= EXIT_USER
;
2956 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2959 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2961 *exit_status
= EXIT_MEMORY
;
2965 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2967 *exit_status
= EXIT_USER
;
2969 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2972 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2975 if (!uid_is_valid(uid
)) {
2976 *exit_status
= EXIT_USER
;
2977 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
2981 if (!gid_is_valid(gid
)) {
2982 *exit_status
= EXIT_USER
;
2983 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
2988 username
= dcreds
->user
->name
;
2991 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2993 *exit_status
= EXIT_USER
;
2994 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
2997 r
= get_fixed_group(context
, &groupname
, &gid
);
2999 *exit_status
= EXIT_GROUP
;
3000 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3004 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3005 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3006 &supplementary_gids
, &ngids
);
3008 *exit_status
= EXIT_GROUP
;
3009 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3012 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3014 *exit_status
= EXIT_USER
;
3015 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3018 user_lookup_fd
= safe_close(user_lookup_fd
);
3020 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3022 *exit_status
= EXIT_CHDIR
;
3023 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3026 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3027 * must sure to drop O_NONBLOCK */
3029 (void) fd_nonblock(socket_fd
, false);
3031 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3033 *exit_status
= EXIT_STDIN
;
3034 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3037 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3039 *exit_status
= EXIT_STDOUT
;
3040 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3043 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3045 *exit_status
= EXIT_STDERR
;
3046 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3049 if (params
->cgroup_path
) {
3050 _cleanup_free_
char *p
= NULL
;
3052 r
= exec_parameters_get_cgroup_path(params
, &p
);
3054 *exit_status
= EXIT_CGROUP
;
3055 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3058 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3060 *exit_status
= EXIT_CGROUP
;
3061 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3065 if (context
->oom_score_adjust_set
) {
3066 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3067 * prohibit write access to this file, and we shouldn't trip up over that. */
3068 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3069 if (IN_SET(r
, -EPERM
, -EACCES
))
3070 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3072 *exit_status
= EXIT_OOM_ADJUST
;
3073 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3077 if (context
->nice_set
)
3078 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3079 *exit_status
= EXIT_NICE
;
3080 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3083 if (context
->cpu_sched_set
) {
3084 struct sched_param param
= {
3085 .sched_priority
= context
->cpu_sched_priority
,
3088 r
= sched_setscheduler(0,
3089 context
->cpu_sched_policy
|
3090 (context
->cpu_sched_reset_on_fork
?
3091 SCHED_RESET_ON_FORK
: 0),
3094 *exit_status
= EXIT_SETSCHEDULER
;
3095 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3099 if (context
->cpuset
)
3100 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3101 *exit_status
= EXIT_CPUAFFINITY
;
3102 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3105 if (context
->ioprio_set
)
3106 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3107 *exit_status
= EXIT_IOPRIO
;
3108 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3111 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3112 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3113 *exit_status
= EXIT_TIMERSLACK
;
3114 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3117 if (context
->personality
!= PERSONALITY_INVALID
) {
3118 r
= safe_personality(context
->personality
);
3120 *exit_status
= EXIT_PERSONALITY
;
3121 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3125 if (context
->utmp_id
)
3126 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3128 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3129 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3133 if (context
->user
) {
3134 r
= chown_terminal(STDIN_FILENO
, uid
);
3136 *exit_status
= EXIT_STDIN
;
3137 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3141 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroupsv1
3142 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3143 * safe. On cgroupsv2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3144 * touch a single hierarchy too. */
3145 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3146 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3148 *exit_status
= EXIT_CGROUP
;
3149 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3153 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3154 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3156 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3159 r
= build_environment(
3171 *exit_status
= EXIT_MEMORY
;
3175 r
= build_pass_environment(context
, &pass_env
);
3177 *exit_status
= EXIT_MEMORY
;
3181 accum_env
= strv_env_merge(5,
3182 params
->environment
,
3185 context
->environment
,
3189 *exit_status
= EXIT_MEMORY
;
3192 accum_env
= strv_env_clean(accum_env
);
3194 (void) umask(context
->umask
);
3196 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3198 *exit_status
= EXIT_KEYRING
;
3199 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3202 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3203 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3205 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked for it, and the kernel doesn't actually support ambient caps */
3206 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3208 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not desired */
3209 if (needs_ambient_hack
)
3210 needs_setuid
= false;
3212 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3214 if (needs_sandboxing
) {
3215 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3216 * present. The actual MAC context application will happen later, as late as possible, to avoid
3217 * impacting our own code paths. */
3220 use_selinux
= mac_selinux_use();
3223 use_smack
= mac_smack_use();
3226 use_apparmor
= mac_apparmor_use();
3231 if (context
->pam_name
&& username
) {
3232 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3234 *exit_status
= EXIT_PAM
;
3235 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3240 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3241 if (ns_type_supported(NAMESPACE_NET
)) {
3242 r
= setup_netns(runtime
->netns_storage_socket
);
3244 *exit_status
= EXIT_NETWORK
;
3245 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3248 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3251 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3252 if (needs_mount_namespace
) {
3253 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3255 *exit_status
= EXIT_NAMESPACE
;
3256 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3260 /* Drop groups as early as possbile */
3262 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3264 *exit_status
= EXIT_GROUP
;
3265 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3269 if (needs_sandboxing
) {
3271 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3272 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3274 *exit_status
= EXIT_SELINUX_CONTEXT
;
3275 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3280 if (context
->private_users
) {
3281 r
= setup_private_users(uid
, gid
);
3283 *exit_status
= EXIT_USER
;
3284 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3289 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3290 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3291 * however if we have it as we want to keep it open until the final execve(). */
3293 if (params
->exec_fd
>= 0) {
3294 exec_fd
= params
->exec_fd
;
3296 if (exec_fd
< 3 + (int) n_fds
) {
3299 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3300 * process we are about to execute. */
3302 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3304 *exit_status
= EXIT_FDS
;
3305 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3308 safe_close(exec_fd
);
3311 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3312 r
= fd_cloexec(exec_fd
, true);
3314 *exit_status
= EXIT_FDS
;
3315 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3319 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3320 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3321 fds_with_exec_fd
[n_fds
] = exec_fd
;
3322 n_fds_with_exec_fd
= n_fds
+ 1;
3324 fds_with_exec_fd
= fds
;
3325 n_fds_with_exec_fd
= n_fds
;
3328 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3330 r
= shift_fds(fds
, n_fds
);
3332 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3334 *exit_status
= EXIT_FDS
;
3335 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3338 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3339 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3340 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3343 secure_bits
= context
->secure_bits
;
3345 if (needs_sandboxing
) {
3349 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3351 *exit_status
= EXIT_LIMITS
;
3352 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3355 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
3356 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3357 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3358 *exit_status
= EXIT_LIMITS
;
3359 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3364 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3365 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3367 r
= setup_smack(context
, command
);
3369 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3370 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3375 bset
= context
->capability_bounding_set
;
3376 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3377 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3378 * instead of us doing that */
3379 if (needs_ambient_hack
)
3380 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3381 (UINT64_C(1) << CAP_SETUID
) |
3382 (UINT64_C(1) << CAP_SETGID
);
3384 if (!cap_test_all(bset
)) {
3385 r
= capability_bounding_set_drop(bset
, false);
3387 *exit_status
= EXIT_CAPABILITIES
;
3388 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3392 /* This is done before enforce_user, but ambient set
3393 * does not survive over setresuid() if keep_caps is not set. */
3394 if (!needs_ambient_hack
&&
3395 context
->capability_ambient_set
!= 0) {
3396 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3398 *exit_status
= EXIT_CAPABILITIES
;
3399 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3405 if (context
->user
) {
3406 r
= enforce_user(context
, uid
);
3408 *exit_status
= EXIT_USER
;
3409 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3412 if (!needs_ambient_hack
&&
3413 context
->capability_ambient_set
!= 0) {
3415 /* Fix the ambient capabilities after user change. */
3416 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3418 *exit_status
= EXIT_CAPABILITIES
;
3419 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3422 /* If we were asked to change user and ambient capabilities
3423 * were requested, we had to add keep-caps to the securebits
3424 * so that we would maintain the inherited capability set
3425 * through the setresuid(). Make sure that the bit is added
3426 * also to the context secure_bits so that we don't try to
3427 * drop the bit away next. */
3429 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3434 /* Apply working directory here, because the working directory might be on NFS and only the user running
3435 * this service might have the correct privilege to change to the working directory */
3436 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3438 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3440 if (needs_sandboxing
) {
3441 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3442 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3443 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3444 * are restricted. */
3448 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3451 r
= setexeccon(exec_context
);
3453 *exit_status
= EXIT_SELINUX_CONTEXT
;
3454 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3461 if (use_apparmor
&& context
->apparmor_profile
) {
3462 r
= aa_change_onexec(context
->apparmor_profile
);
3463 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3464 *exit_status
= EXIT_APPARMOR_PROFILE
;
3465 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3470 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3471 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3472 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3473 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3474 *exit_status
= EXIT_SECUREBITS
;
3475 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3478 if (context_has_no_new_privileges(context
))
3479 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3480 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3481 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3485 r
= apply_address_families(unit
, context
);
3487 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3488 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3491 r
= apply_memory_deny_write_execute(unit
, context
);
3493 *exit_status
= EXIT_SECCOMP
;
3494 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3497 r
= apply_restrict_realtime(unit
, context
);
3499 *exit_status
= EXIT_SECCOMP
;
3500 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3503 r
= apply_restrict_namespaces(unit
, context
);
3505 *exit_status
= EXIT_SECCOMP
;
3506 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3509 r
= apply_protect_sysctl(unit
, context
);
3511 *exit_status
= EXIT_SECCOMP
;
3512 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3515 r
= apply_protect_kernel_modules(unit
, context
);
3517 *exit_status
= EXIT_SECCOMP
;
3518 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3521 r
= apply_private_devices(unit
, context
);
3523 *exit_status
= EXIT_SECCOMP
;
3524 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3527 r
= apply_syscall_archs(unit
, context
);
3529 *exit_status
= EXIT_SECCOMP
;
3530 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3533 r
= apply_lock_personality(unit
, context
);
3535 *exit_status
= EXIT_SECCOMP
;
3536 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3539 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3540 * by the filter as little as possible. */
3541 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3543 *exit_status
= EXIT_SECCOMP
;
3544 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3549 if (!strv_isempty(context
->unset_environment
)) {
3552 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3554 *exit_status
= EXIT_MEMORY
;
3558 strv_free_and_replace(accum_env
, ee
);
3561 final_argv
= replace_env_argv(command
->argv
, accum_env
);
3563 *exit_status
= EXIT_MEMORY
;
3567 if (DEBUG_LOGGING
) {
3568 _cleanup_free_
char *line
;
3570 line
= exec_command_line(final_argv
);
3572 log_struct(LOG_DEBUG
,
3573 "EXECUTABLE=%s", command
->path
,
3574 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3576 LOG_UNIT_INVOCATION_ID(unit
));
3582 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3583 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3585 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3586 *exit_status
= EXIT_EXEC
;
3587 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3591 execve(command
->path
, final_argv
, accum_env
);
3597 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3598 * that POLLHUP on it no longer means execve() succeeded. */
3600 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3601 *exit_status
= EXIT_EXEC
;
3602 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3606 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3607 log_struct_errno(LOG_INFO
, r
,
3608 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3610 LOG_UNIT_INVOCATION_ID(unit
),
3611 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3613 "EXECUTABLE=%s", command
->path
);
3617 *exit_status
= EXIT_EXEC
;
3618 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3621 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3622 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3624 int exec_spawn(Unit
*unit
,
3625 ExecCommand
*command
,
3626 const ExecContext
*context
,
3627 const ExecParameters
*params
,
3628 ExecRuntime
*runtime
,
3629 DynamicCreds
*dcreds
,
3632 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3633 _cleanup_free_
char *subcgroup_path
= NULL
;
3634 _cleanup_strv_free_
char **files_env
= NULL
;
3635 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3636 _cleanup_free_
char *line
= NULL
;
3644 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3646 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3647 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3648 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3650 if (params
->n_socket_fds
> 1) {
3651 log_unit_error(unit
, "Got more than one socket.");
3655 if (params
->n_socket_fds
== 0) {
3656 log_unit_error(unit
, "Got no socket.");
3660 socket_fd
= params
->fds
[0];
3664 n_socket_fds
= params
->n_socket_fds
;
3665 n_storage_fds
= params
->n_storage_fds
;
3668 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3670 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3672 r
= exec_context_load_environment(unit
, context
, &files_env
);
3674 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3676 line
= exec_command_line(command
->argv
);
3680 log_struct(LOG_DEBUG
,
3681 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3682 "EXECUTABLE=%s", command
->path
,
3684 LOG_UNIT_INVOCATION_ID(unit
));
3686 if (params
->cgroup_path
) {
3687 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3689 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3690 if (r
> 0) { /* We are using a child cgroup */
3691 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3693 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3699 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3702 int exit_status
= EXIT_SUCCESS
;
3704 r
= exec_child(unit
,
3716 unit
->manager
->user_lookup_fds
[1],
3720 log_struct_errno(LOG_ERR
, r
,
3721 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3723 LOG_UNIT_INVOCATION_ID(unit
),
3724 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3725 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3727 "EXECUTABLE=%s", command
->path
);
3732 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3734 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3735 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3736 * process will be killed too). */
3738 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3740 exec_status_start(&command
->exec_status
, pid
);
3746 void exec_context_init(ExecContext
*c
) {
3747 ExecDirectoryType i
;
3752 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3753 c
->cpu_sched_policy
= SCHED_OTHER
;
3754 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3755 c
->syslog_level_prefix
= true;
3756 c
->ignore_sigpipe
= true;
3757 c
->timer_slack_nsec
= NSEC_INFINITY
;
3758 c
->personality
= PERSONALITY_INVALID
;
3759 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3760 c
->directories
[i
].mode
= 0755;
3761 c
->capability_bounding_set
= CAP_ALL
;
3762 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3763 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3764 c
->log_level_max
= -1;
3767 void exec_context_done(ExecContext
*c
) {
3768 ExecDirectoryType i
;
3773 c
->environment
= strv_free(c
->environment
);
3774 c
->environment_files
= strv_free(c
->environment_files
);
3775 c
->pass_environment
= strv_free(c
->pass_environment
);
3776 c
->unset_environment
= strv_free(c
->unset_environment
);
3778 rlimit_free_all(c
->rlimit
);
3780 for (l
= 0; l
< 3; l
++) {
3781 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3782 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3785 c
->working_directory
= mfree(c
->working_directory
);
3786 c
->root_directory
= mfree(c
->root_directory
);
3787 c
->root_image
= mfree(c
->root_image
);
3788 c
->tty_path
= mfree(c
->tty_path
);
3789 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3790 c
->user
= mfree(c
->user
);
3791 c
->group
= mfree(c
->group
);
3793 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3795 c
->pam_name
= mfree(c
->pam_name
);
3797 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3798 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3799 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3801 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3802 c
->bind_mounts
= NULL
;
3803 c
->n_bind_mounts
= 0;
3804 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3805 c
->temporary_filesystems
= NULL
;
3806 c
->n_temporary_filesystems
= 0;
3808 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3810 c
->utmp_id
= mfree(c
->utmp_id
);
3811 c
->selinux_context
= mfree(c
->selinux_context
);
3812 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3813 c
->smack_process_label
= mfree(c
->smack_process_label
);
3815 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3816 c
->syscall_archs
= set_free(c
->syscall_archs
);
3817 c
->address_families
= set_free(c
->address_families
);
3819 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3820 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3822 c
->log_level_max
= -1;
3824 exec_context_free_log_extra_fields(c
);
3826 c
->log_rate_limit_interval_usec
= 0;
3827 c
->log_rate_limit_burst
= 0;
3829 c
->stdin_data
= mfree(c
->stdin_data
);
3830 c
->stdin_data_size
= 0;
3833 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3838 if (!runtime_prefix
)
3841 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3842 _cleanup_free_
char *p
;
3844 p
= strjoin(runtime_prefix
, "/", *i
);
3848 /* We execute this synchronously, since we need to be sure this is gone when we start the service
3850 (void) rm_rf(p
, REMOVE_ROOT
);
3856 static void exec_command_done(ExecCommand
*c
) {
3859 c
->path
= mfree(c
->path
);
3860 c
->argv
= strv_free(c
->argv
);
3863 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3866 for (i
= 0; i
< n
; i
++)
3867 exec_command_done(c
+i
);
3870 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3874 LIST_REMOVE(command
, c
, i
);
3875 exec_command_done(i
);
3882 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3885 for (i
= 0; i
< n
; i
++)
3886 c
[i
] = exec_command_free_list(c
[i
]);
3889 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3892 for (i
= 0; i
< n
; i
++)
3893 exec_status_reset(&c
[i
].exec_status
);
3896 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3899 for (i
= 0; i
< n
; i
++) {
3902 LIST_FOREACH(command
, z
, c
[i
])
3903 exec_status_reset(&z
->exec_status
);
3907 typedef struct InvalidEnvInfo
{
3912 static void invalid_env(const char *p
, void *userdata
) {
3913 InvalidEnvInfo
*info
= userdata
;
3915 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3918 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3924 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3927 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3930 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3933 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3936 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3939 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3946 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3948 const char* stdio_fdname
[3];
3954 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3955 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3956 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3958 for (i
= 0; i
< 3; i
++)
3959 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3961 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3963 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3964 if (named_iofds
[STDIN_FILENO
] < 0 &&
3965 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3966 stdio_fdname
[STDIN_FILENO
] &&
3967 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3969 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3972 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3973 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3974 stdio_fdname
[STDOUT_FILENO
] &&
3975 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3977 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3980 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3981 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3982 stdio_fdname
[STDERR_FILENO
] &&
3983 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3985 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3989 return targets
== 0 ? 0 : -ENOENT
;
3992 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
3993 char **i
, **r
= NULL
;
3998 STRV_FOREACH(i
, c
->environment_files
) {
4002 bool ignore
= false;
4004 _cleanup_globfree_ glob_t pglob
= {};
4013 if (!path_is_absolute(fn
)) {
4021 /* Filename supports globbing, take all matching files */
4022 k
= safe_glob(fn
, 0, &pglob
);
4031 /* When we don't match anything, -ENOENT should be returned */
4032 assert(pglob
.gl_pathc
> 0);
4034 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4035 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4043 /* Log invalid environment variables with filename */
4045 InvalidEnvInfo info
= {
4047 .path
= pglob
.gl_pathv
[n
]
4050 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4058 m
= strv_env_merge(2, r
, p
);
4074 static bool tty_may_match_dev_console(const char *tty
) {
4075 _cleanup_free_
char *resolved
= NULL
;
4080 tty
= skip_dev_prefix(tty
);
4082 /* trivial identity? */
4083 if (streq(tty
, "console"))
4086 if (resolve_dev_console(&resolved
) < 0)
4087 return true; /* if we could not resolve, assume it may */
4089 /* "tty0" means the active VC, so it may be the same sometimes */
4090 return streq(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4093 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4095 return (ec
->tty_reset
||
4097 ec
->tty_vt_disallocate
||
4098 is_terminal_input(ec
->std_input
) ||
4099 is_terminal_output(ec
->std_output
) ||
4100 is_terminal_output(ec
->std_error
)) &&
4101 tty_may_match_dev_console(exec_context_tty_path(ec
));
4104 static void strv_fprintf(FILE *f
, char **l
) {
4110 fprintf(f
, " %s", *g
);
4113 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4114 ExecDirectoryType dt
;
4122 prefix
= strempty(prefix
);
4126 "%sWorkingDirectory: %s\n"
4127 "%sRootDirectory: %s\n"
4128 "%sNonBlocking: %s\n"
4129 "%sPrivateTmp: %s\n"
4130 "%sPrivateDevices: %s\n"
4131 "%sProtectKernelTunables: %s\n"
4132 "%sProtectKernelModules: %s\n"
4133 "%sProtectControlGroups: %s\n"
4134 "%sPrivateNetwork: %s\n"
4135 "%sPrivateUsers: %s\n"
4136 "%sProtectHome: %s\n"
4137 "%sProtectSystem: %s\n"
4138 "%sMountAPIVFS: %s\n"
4139 "%sIgnoreSIGPIPE: %s\n"
4140 "%sMemoryDenyWriteExecute: %s\n"
4141 "%sRestrictRealtime: %s\n"
4142 "%sKeyringMode: %s\n",
4144 prefix
, c
->working_directory
? c
->working_directory
: "/",
4145 prefix
, c
->root_directory
? c
->root_directory
: "/",
4146 prefix
, yes_no(c
->non_blocking
),
4147 prefix
, yes_no(c
->private_tmp
),
4148 prefix
, yes_no(c
->private_devices
),
4149 prefix
, yes_no(c
->protect_kernel_tunables
),
4150 prefix
, yes_no(c
->protect_kernel_modules
),
4151 prefix
, yes_no(c
->protect_control_groups
),
4152 prefix
, yes_no(c
->private_network
),
4153 prefix
, yes_no(c
->private_users
),
4154 prefix
, protect_home_to_string(c
->protect_home
),
4155 prefix
, protect_system_to_string(c
->protect_system
),
4156 prefix
, yes_no(c
->mount_apivfs
),
4157 prefix
, yes_no(c
->ignore_sigpipe
),
4158 prefix
, yes_no(c
->memory_deny_write_execute
),
4159 prefix
, yes_no(c
->restrict_realtime
),
4160 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
));
4163 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4165 STRV_FOREACH(e
, c
->environment
)
4166 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4168 STRV_FOREACH(e
, c
->environment_files
)
4169 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4171 STRV_FOREACH(e
, c
->pass_environment
)
4172 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4174 STRV_FOREACH(e
, c
->unset_environment
)
4175 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4177 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4179 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4180 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4182 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4183 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4191 if (c
->oom_score_adjust_set
)
4193 "%sOOMScoreAdjust: %i\n",
4194 prefix
, c
->oom_score_adjust
);
4196 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4198 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4199 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4200 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4201 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4204 if (c
->ioprio_set
) {
4205 _cleanup_free_
char *class_str
= NULL
;
4207 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4209 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4211 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4214 if (c
->cpu_sched_set
) {
4215 _cleanup_free_
char *policy_str
= NULL
;
4217 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4219 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4222 "%sCPUSchedulingPriority: %i\n"
4223 "%sCPUSchedulingResetOnFork: %s\n",
4224 prefix
, c
->cpu_sched_priority
,
4225 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4229 fprintf(f
, "%sCPUAffinity:", prefix
);
4230 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4231 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4232 fprintf(f
, " %u", i
);
4236 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4237 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4240 "%sStandardInput: %s\n"
4241 "%sStandardOutput: %s\n"
4242 "%sStandardError: %s\n",
4243 prefix
, exec_input_to_string(c
->std_input
),
4244 prefix
, exec_output_to_string(c
->std_output
),
4245 prefix
, exec_output_to_string(c
->std_error
));
4247 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4248 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4249 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4250 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4251 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4252 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4254 if (c
->std_input
== EXEC_INPUT_FILE
)
4255 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4256 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4257 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4258 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4259 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4260 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4261 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4262 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4263 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4269 "%sTTYVHangup: %s\n"
4270 "%sTTYVTDisallocate: %s\n",
4271 prefix
, c
->tty_path
,
4272 prefix
, yes_no(c
->tty_reset
),
4273 prefix
, yes_no(c
->tty_vhangup
),
4274 prefix
, yes_no(c
->tty_vt_disallocate
));
4276 if (IN_SET(c
->std_output
,
4279 EXEC_OUTPUT_JOURNAL
,
4280 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4281 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4282 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4283 IN_SET(c
->std_error
,
4286 EXEC_OUTPUT_JOURNAL
,
4287 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4288 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4289 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4291 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4293 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4295 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4297 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4299 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4302 if (c
->log_level_max
>= 0) {
4303 _cleanup_free_
char *t
= NULL
;
4305 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4307 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4310 if (c
->log_rate_limit_interval_usec
> 0) {
4311 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4314 "%sLogRateLimitIntervalSec: %s\n",
4315 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4318 if (c
->log_rate_limit_burst
> 0)
4319 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4321 if (c
->n_log_extra_fields
> 0) {
4324 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4325 fprintf(f
, "%sLogExtraFields: ", prefix
);
4326 fwrite(c
->log_extra_fields
[j
].iov_base
,
4327 1, c
->log_extra_fields
[j
].iov_len
,
4333 if (c
->secure_bits
) {
4334 _cleanup_free_
char *str
= NULL
;
4336 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4338 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4341 if (c
->capability_bounding_set
!= CAP_ALL
) {
4342 _cleanup_free_
char *str
= NULL
;
4344 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4346 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4349 if (c
->capability_ambient_set
!= 0) {
4350 _cleanup_free_
char *str
= NULL
;
4352 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4354 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4358 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4360 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4362 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4364 if (!strv_isempty(c
->supplementary_groups
)) {
4365 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4366 strv_fprintf(f
, c
->supplementary_groups
);
4371 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4373 if (!strv_isempty(c
->read_write_paths
)) {
4374 fprintf(f
, "%sReadWritePaths:", prefix
);
4375 strv_fprintf(f
, c
->read_write_paths
);
4379 if (!strv_isempty(c
->read_only_paths
)) {
4380 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4381 strv_fprintf(f
, c
->read_only_paths
);
4385 if (!strv_isempty(c
->inaccessible_paths
)) {
4386 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4387 strv_fprintf(f
, c
->inaccessible_paths
);
4391 if (c
->n_bind_mounts
> 0)
4392 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4393 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4394 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4395 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4396 c
->bind_mounts
[i
].source
,
4397 c
->bind_mounts
[i
].destination
,
4398 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4400 if (c
->n_temporary_filesystems
> 0)
4401 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4402 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4404 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4406 isempty(t
->options
) ? "" : ":",
4407 strempty(t
->options
));
4412 "%sUtmpIdentifier: %s\n",
4413 prefix
, c
->utmp_id
);
4415 if (c
->selinux_context
)
4417 "%sSELinuxContext: %s%s\n",
4418 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4420 if (c
->apparmor_profile
)
4422 "%sAppArmorProfile: %s%s\n",
4423 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4425 if (c
->smack_process_label
)
4427 "%sSmackProcessLabel: %s%s\n",
4428 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4430 if (c
->personality
!= PERSONALITY_INVALID
)
4432 "%sPersonality: %s\n",
4433 prefix
, strna(personality_to_string(c
->personality
)));
4436 "%sLockPersonality: %s\n",
4437 prefix
, yes_no(c
->lock_personality
));
4439 if (c
->syscall_filter
) {
4447 "%sSystemCallFilter: ",
4450 if (!c
->syscall_whitelist
)
4454 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4455 _cleanup_free_
char *name
= NULL
;
4456 const char *errno_name
= NULL
;
4457 int num
= PTR_TO_INT(val
);
4464 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4465 fputs(strna(name
), f
);
4468 errno_name
= errno_to_name(num
);
4470 fprintf(f
, ":%s", errno_name
);
4472 fprintf(f
, ":%d", num
);
4480 if (c
->syscall_archs
) {
4487 "%sSystemCallArchitectures:",
4491 SET_FOREACH(id
, c
->syscall_archs
, j
)
4492 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4497 if (exec_context_restrict_namespaces_set(c
)) {
4498 _cleanup_free_
char *s
= NULL
;
4500 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4502 fprintf(f
, "%sRestrictNamespaces: %s\n",
4506 if (c
->syscall_errno
> 0) {
4507 const char *errno_name
;
4509 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4511 errno_name
= errno_to_name(c
->syscall_errno
);
4513 fprintf(f
, "%s\n", errno_name
);
4515 fprintf(f
, "%d\n", c
->syscall_errno
);
4518 if (c
->apparmor_profile
)
4520 "%sAppArmorProfile: %s%s\n",
4521 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4524 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4527 /* Returns true if the process forked off would run under
4528 * an unchanged UID or as root. */
4533 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4539 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4547 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4549 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4554 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4559 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4560 free(c
->log_extra_fields
[l
].iov_base
);
4561 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4562 c
->n_log_extra_fields
= 0;
4565 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4572 dual_timestamp_get(&s
->start_timestamp
);
4575 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4578 if (s
->pid
!= pid
) {
4584 dual_timestamp_get(&s
->exit_timestamp
);
4590 if (context
->utmp_id
)
4591 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4593 exec_context_tty_reset(context
, NULL
);
4597 void exec_status_reset(ExecStatus
*s
) {
4600 *s
= (ExecStatus
) {};
4603 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4604 char buf
[FORMAT_TIMESTAMP_MAX
];
4612 prefix
= strempty(prefix
);
4615 "%sPID: "PID_FMT
"\n",
4618 if (dual_timestamp_is_set(&s
->start_timestamp
))
4620 "%sStart Timestamp: %s\n",
4621 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4623 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4625 "%sExit Timestamp: %s\n"
4627 "%sExit Status: %i\n",
4628 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4629 prefix
, sigchld_code_to_string(s
->code
),
4633 static char *exec_command_line(char **argv
) {
4641 STRV_FOREACH(a
, argv
)
4649 STRV_FOREACH(a
, argv
) {
4656 if (strpbrk(*a
, WHITESPACE
)) {
4667 /* FIXME: this doesn't really handle arguments that have
4668 * spaces and ticks in them */
4673 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4674 _cleanup_free_
char *cmd
= NULL
;
4675 const char *prefix2
;
4680 prefix
= strempty(prefix
);
4681 prefix2
= strjoina(prefix
, "\t");
4683 cmd
= exec_command_line(c
->argv
);
4685 "%sCommand Line: %s\n",
4686 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4688 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4691 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4694 prefix
= strempty(prefix
);
4696 LIST_FOREACH(command
, c
, c
)
4697 exec_command_dump(c
, f
, prefix
);
4700 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4707 /* It's kind of important, that we keep the order here */
4708 LIST_FIND_TAIL(command
, *l
, end
);
4709 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4714 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4722 l
= strv_new_ap(path
, ap
);
4734 free_and_replace(c
->path
, p
);
4736 return strv_free_and_replace(c
->argv
, l
);
4739 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4740 _cleanup_strv_free_
char **l
= NULL
;
4748 l
= strv_new_ap(path
, ap
);
4754 r
= strv_extend_strv(&c
->argv
, l
, false);
4761 static void *remove_tmpdir_thread(void *p
) {
4762 _cleanup_free_
char *path
= p
;
4764 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4768 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4775 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4777 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4778 if (destroy
&& rt
->tmp_dir
) {
4779 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4781 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4783 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4790 if (destroy
&& rt
->var_tmp_dir
) {
4791 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4793 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4795 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4796 free(rt
->var_tmp_dir
);
4799 rt
->var_tmp_dir
= NULL
;
4802 rt
->id
= mfree(rt
->id
);
4803 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4804 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4805 safe_close_pair(rt
->netns_storage_socket
);
4809 static void exec_runtime_freep(ExecRuntime
**rt
) {
4811 (void) exec_runtime_free(*rt
, false);
4814 static int exec_runtime_allocate(ExecRuntime
**rt
) {
4817 *rt
= new0(ExecRuntime
, 1);
4821 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
4825 static int exec_runtime_add(
4828 const char *tmp_dir
,
4829 const char *var_tmp_dir
,
4830 const int netns_storage_socket
[2],
4831 ExecRuntime
**ret
) {
4833 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4839 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4843 r
= exec_runtime_allocate(&rt
);
4847 rt
->id
= strdup(id
);
4852 rt
->tmp_dir
= strdup(tmp_dir
);
4856 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4857 assert(var_tmp_dir
);
4858 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4859 if (!rt
->var_tmp_dir
)
4863 if (netns_storage_socket
) {
4864 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4865 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4868 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4877 /* do not remove created ExecRuntime object when the operation succeeds. */
4882 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4883 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4884 _cleanup_close_pair_
int netns_storage_socket
[2] = {-1, -1};
4891 /* It is not necessary to create ExecRuntime object. */
4892 if (!c
->private_network
&& !c
->private_tmp
)
4895 if (c
->private_tmp
) {
4896 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
4901 if (c
->private_network
) {
4902 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
4906 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
4911 netns_storage_socket
[0] = -1;
4912 netns_storage_socket
[1] = -1;
4916 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
4924 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
4926 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
4932 /* If not found, then create a new object. */
4933 r
= exec_runtime_make(m
, c
, id
, &rt
);
4935 /* When r == 0, it is not necessary to create ExecRuntime object. */
4939 /* increment reference counter. */
4945 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
4949 assert(rt
->n_ref
> 0);
4955 return exec_runtime_free(rt
, destroy
);
4958 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
4966 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
4967 fprintf(f
, "exec-runtime=%s", rt
->id
);
4970 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
4972 if (rt
->var_tmp_dir
)
4973 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
4975 if (rt
->netns_storage_socket
[0] >= 0) {
4978 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4982 fprintf(f
, " netns-socket-0=%i", copy
);
4985 if (rt
->netns_storage_socket
[1] >= 0) {
4988 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
4992 fprintf(f
, " netns-socket-1=%i", copy
);
5001 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5002 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5006 /* This is for the migration from old (v237 or earlier) deserialization text.
5007 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5008 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5009 * so or not from the serialized text, then we always creates a new object owned by this. */
5015 /* Manager manages ExecRuntime objects by the unit id.
5016 * So, we omit the serialized text when the unit does not have id (yet?)... */
5017 if (isempty(u
->id
)) {
5018 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5022 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5024 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5028 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5030 r
= exec_runtime_allocate(&rt_create
);
5034 rt_create
->id
= strdup(u
->id
);
5041 if (streq(key
, "tmp-dir")) {
5044 copy
= strdup(value
);
5048 free_and_replace(rt
->tmp_dir
, copy
);
5050 } else if (streq(key
, "var-tmp-dir")) {
5053 copy
= strdup(value
);
5057 free_and_replace(rt
->var_tmp_dir
, copy
);
5059 } else if (streq(key
, "netns-socket-0")) {
5062 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5063 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5067 safe_close(rt
->netns_storage_socket
[0]);
5068 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5070 } else if (streq(key
, "netns-socket-1")) {
5073 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5074 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5078 safe_close(rt
->netns_storage_socket
[1]);
5079 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5083 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5085 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5087 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5091 rt_create
->manager
= u
->manager
;
5100 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5101 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5102 int r
, fd0
= -1, fd1
= -1;
5103 const char *p
, *v
= value
;
5110 n
= strcspn(v
, " ");
5111 id
= strndupa(v
, n
);
5116 v
= startswith(p
, "tmp-dir=");
5118 n
= strcspn(v
, " ");
5119 tmp_dir
= strndupa(v
, n
);
5125 v
= startswith(p
, "var-tmp-dir=");
5127 n
= strcspn(v
, " ");
5128 var_tmp_dir
= strndupa(v
, n
);
5134 v
= startswith(p
, "netns-socket-0=");
5138 n
= strcspn(v
, " ");
5139 buf
= strndupa(v
, n
);
5140 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5141 log_debug("Unable to process exec-runtime netns fd specification.");
5144 fd0
= fdset_remove(fds
, fd0
);
5150 v
= startswith(p
, "netns-socket-1=");
5154 n
= strcspn(v
, " ");
5155 buf
= strndupa(v
, n
);
5156 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5157 log_debug("Unable to process exec-runtime netns fd specification.");
5160 fd1
= fdset_remove(fds
, fd1
);
5165 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5167 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5170 void exec_runtime_vacuum(Manager
*m
) {
5176 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5178 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5182 (void) exec_runtime_free(rt
, false);
5186 void exec_params_clear(ExecParameters
*p
) {
5190 strv_free(p
->environment
);
5193 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5194 [EXEC_INPUT_NULL
] = "null",
5195 [EXEC_INPUT_TTY
] = "tty",
5196 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5197 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5198 [EXEC_INPUT_SOCKET
] = "socket",
5199 [EXEC_INPUT_NAMED_FD
] = "fd",
5200 [EXEC_INPUT_DATA
] = "data",
5201 [EXEC_INPUT_FILE
] = "file",
5204 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5206 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5207 [EXEC_OUTPUT_INHERIT
] = "inherit",
5208 [EXEC_OUTPUT_NULL
] = "null",
5209 [EXEC_OUTPUT_TTY
] = "tty",
5210 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5211 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5212 [EXEC_OUTPUT_KMSG
] = "kmsg",
5213 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5214 [EXEC_OUTPUT_JOURNAL
] = "journal",
5215 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5216 [EXEC_OUTPUT_SOCKET
] = "socket",
5217 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5218 [EXEC_OUTPUT_FILE
] = "file",
5219 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5222 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5224 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5225 [EXEC_UTMP_INIT
] = "init",
5226 [EXEC_UTMP_LOGIN
] = "login",
5227 [EXEC_UTMP_USER
] = "user",
5230 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5232 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5233 [EXEC_PRESERVE_NO
] = "no",
5234 [EXEC_PRESERVE_YES
] = "yes",
5235 [EXEC_PRESERVE_RESTART
] = "restart",
5238 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5240 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5241 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5242 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5243 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5244 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5245 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5248 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5250 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5251 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5252 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5253 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5254 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5255 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5258 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5260 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5261 [EXEC_KEYRING_INHERIT
] = "inherit",
5262 [EXEC_KEYRING_PRIVATE
] = "private",
5263 [EXEC_KEYRING_SHARED
] = "shared",
5266 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);