1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
28 #include <sys/capability.h>
29 #include <sys/eventfd.h>
31 #include <sys/personality.h>
32 #include <sys/prctl.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
42 #include <security/pam_appl.h>
46 #include <selinux/selinux.h>
54 #include <sys/apparmor.h>
57 #include "sd-messages.h"
60 #include "alloc-util.h"
62 #include "apparmor-util.h"
67 #include "capability-util.h"
68 #include "chown-recursive.h"
69 #include "cpu-set-util.h"
72 #include "errno-list.h"
74 #include "exit-status.h"
77 #include "format-util.h"
79 #include "glob-util.h"
87 #include "namespace.h"
88 #include "parse-util.h"
89 #include "path-util.h"
90 #include "process-util.h"
91 #include "rlimit-util.h"
94 #include "seccomp-util.h"
96 #include "securebits.h"
97 #include "securebits-util.h"
98 #include "selinux-util.h"
99 #include "signal-util.h"
100 #include "smack-util.h"
102 #include "stat-util.h"
103 #include "string-table.h"
104 #include "string-util.h"
106 #include "syslog-util.h"
107 #include "terminal-util.h"
109 #include "user-util.h"
111 #include "utmp-wtmp.h"
113 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
114 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
116 /* This assumes there is a 'tty' group */
117 #define TTY_MODE 0620
119 #define SNDBUF_SIZE (8*1024*1024)
121 static int shift_fds(int fds
[], unsigned n_fds
) {
122 int start
, restart_from
;
127 /* Modifies the fds array! (sorts it) */
137 for (i
= start
; i
< (int) n_fds
; i
++) {
140 /* Already at right index? */
144 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
151 /* Hmm, the fd we wanted isn't free? Then
152 * let's remember that and try again from here */
153 if (nfd
!= i
+3 && restart_from
< 0)
157 if (restart_from
< 0)
160 start
= restart_from
;
166 static int flags_fds(const int fds
[], unsigned n_storage_fds
, unsigned n_socket_fds
, bool nonblock
) {
170 n_fds
= n_storage_fds
+ n_socket_fds
;
176 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
177 * O_NONBLOCK only applies to socket activation though. */
179 for (i
= 0; i
< n_fds
; i
++) {
181 if (i
< n_socket_fds
) {
182 r
= fd_nonblock(fds
[i
], nonblock
);
187 /* We unconditionally drop FD_CLOEXEC from the fds,
188 * since after all we want to pass these fds to our
191 r
= fd_cloexec(fds
[i
], false);
199 static const char *exec_context_tty_path(const ExecContext
*context
) {
202 if (context
->stdio_as_fds
)
205 if (context
->tty_path
)
206 return context
->tty_path
;
208 return "/dev/console";
211 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
216 path
= exec_context_tty_path(context
);
218 if (context
->tty_vhangup
) {
219 if (p
&& p
->stdin_fd
>= 0)
220 (void) terminal_vhangup_fd(p
->stdin_fd
);
222 (void) terminal_vhangup(path
);
225 if (context
->tty_reset
) {
226 if (p
&& p
->stdin_fd
>= 0)
227 (void) reset_terminal_fd(p
->stdin_fd
, true);
229 (void) reset_terminal(path
);
232 if (context
->tty_vt_disallocate
&& path
)
233 (void) vt_disallocate(path
);
236 static bool is_terminal_input(ExecInput i
) {
239 EXEC_INPUT_TTY_FORCE
,
240 EXEC_INPUT_TTY_FAIL
);
243 static bool is_terminal_output(ExecOutput o
) {
246 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
247 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
248 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
251 static bool is_syslog_output(ExecOutput o
) {
254 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
257 static bool is_kmsg_output(ExecOutput o
) {
260 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
263 static bool exec_context_needs_term(const ExecContext
*c
) {
266 /* Return true if the execution context suggests we should set $TERM to something useful. */
268 if (is_terminal_input(c
->std_input
))
271 if (is_terminal_output(c
->std_output
))
274 if (is_terminal_output(c
->std_error
))
277 return !!c
->tty_path
;
280 static int open_null_as(int flags
, int nfd
) {
285 fd
= open("/dev/null", flags
|O_NOCTTY
);
289 return move_fd(fd
, nfd
, false);
292 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
293 static const union sockaddr_union sa
= {
294 .un
.sun_family
= AF_UNIX
,
295 .un
.sun_path
= "/run/systemd/journal/stdout",
297 uid_t olduid
= UID_INVALID
;
298 gid_t oldgid
= GID_INVALID
;
301 if (gid_is_valid(gid
)) {
304 if (setegid(gid
) < 0)
308 if (uid_is_valid(uid
)) {
311 if (seteuid(uid
) < 0) {
317 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
319 /* If we fail to restore the uid or gid, things will likely
320 fail later on. This should only happen if an LSM interferes. */
322 if (uid_is_valid(uid
))
323 (void) seteuid(olduid
);
326 if (gid_is_valid(gid
))
327 (void) setegid(oldgid
);
332 static int connect_logger_as(
334 const ExecContext
*context
,
335 const ExecParameters
*params
,
346 assert(output
< _EXEC_OUTPUT_MAX
);
350 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
354 r
= connect_journal_socket(fd
, uid
, gid
);
358 if (shutdown(fd
, SHUT_RD
) < 0) {
363 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
373 context
->syslog_identifier
?: ident
,
374 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
375 context
->syslog_priority
,
376 !!context
->syslog_level_prefix
,
377 is_syslog_output(output
),
378 is_kmsg_output(output
),
379 is_terminal_output(output
));
381 return move_fd(fd
, nfd
, false);
383 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
389 fd
= open_terminal(path
, flags
| O_NOCTTY
);
393 return move_fd(fd
, nfd
, false);
396 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
397 union sockaddr_union sa
= {
398 .sa
.sa_family
= AF_UNIX
,
404 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
407 fd
= open(path
, flags
|O_NOCTTY
, mode
);
411 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
413 if (strlen(path
) > sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
416 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
418 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
422 strncpy(sa
.un
.sun_path
, path
, sizeof(sa
.un
.sun_path
));
423 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
425 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
426 * indication that his wasn't an AF_UNIX socket after all */
429 if ((flags
& O_ACCMODE
) == O_RDONLY
)
430 r
= shutdown(fd
, SHUT_WR
);
431 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
432 r
= shutdown(fd
, SHUT_RD
);
443 static int fixup_input(
444 const ExecContext
*context
,
446 bool apply_tty_stdin
) {
452 std_input
= context
->std_input
;
454 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
455 return EXEC_INPUT_NULL
;
457 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
458 return EXEC_INPUT_NULL
;
460 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
461 return EXEC_INPUT_NULL
;
466 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
468 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
469 return EXEC_OUTPUT_INHERIT
;
474 static int setup_input(
475 const ExecContext
*context
,
476 const ExecParameters
*params
,
478 int named_iofds
[3]) {
485 if (params
->stdin_fd
>= 0) {
486 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
489 /* Try to make this the controlling tty, if it is a tty, and reset it */
490 if (isatty(STDIN_FILENO
)) {
491 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
492 (void) reset_terminal_fd(STDIN_FILENO
, true);
498 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
502 case EXEC_INPUT_NULL
:
503 return open_null_as(O_RDONLY
, STDIN_FILENO
);
506 case EXEC_INPUT_TTY_FORCE
:
507 case EXEC_INPUT_TTY_FAIL
: {
510 fd
= acquire_terminal(exec_context_tty_path(context
),
511 i
== EXEC_INPUT_TTY_FAIL
,
512 i
== EXEC_INPUT_TTY_FORCE
,
518 return move_fd(fd
, STDIN_FILENO
, false);
521 case EXEC_INPUT_SOCKET
:
522 assert(socket_fd
>= 0);
524 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
526 case EXEC_INPUT_NAMED_FD
:
527 assert(named_iofds
[STDIN_FILENO
] >= 0);
529 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
530 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
532 case EXEC_INPUT_DATA
: {
535 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
539 return move_fd(fd
, STDIN_FILENO
, false);
542 case EXEC_INPUT_FILE
: {
546 assert(context
->stdio_file
[STDIN_FILENO
]);
548 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
549 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
551 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
555 return move_fd(fd
, STDIN_FILENO
, false);
559 assert_not_reached("Unknown input type");
563 static int setup_output(
565 const ExecContext
*context
,
566 const ExecParameters
*params
,
573 dev_t
*journal_stream_dev
,
574 ino_t
*journal_stream_ino
) {
584 assert(journal_stream_dev
);
585 assert(journal_stream_ino
);
587 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
589 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
592 return STDOUT_FILENO
;
595 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
596 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
599 return STDERR_FILENO
;
602 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
603 o
= fixup_output(context
->std_output
, socket_fd
);
605 if (fileno
== STDERR_FILENO
) {
607 e
= fixup_output(context
->std_error
, socket_fd
);
609 /* This expects the input and output are already set up */
611 /* Don't change the stderr file descriptor if we inherit all
612 * the way and are not on a tty */
613 if (e
== EXEC_OUTPUT_INHERIT
&&
614 o
== EXEC_OUTPUT_INHERIT
&&
615 i
== EXEC_INPUT_NULL
&&
616 !is_terminal_input(context
->std_input
) &&
620 /* Duplicate from stdout if possible */
621 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
622 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
626 } else if (o
== EXEC_OUTPUT_INHERIT
) {
627 /* If input got downgraded, inherit the original value */
628 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
629 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
631 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
632 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
633 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
635 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
639 /* We need to open /dev/null here anew, to get the right access mode. */
640 return open_null_as(O_WRONLY
, fileno
);
645 case EXEC_OUTPUT_NULL
:
646 return open_null_as(O_WRONLY
, fileno
);
648 case EXEC_OUTPUT_TTY
:
649 if (is_terminal_input(i
))
650 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
652 /* We don't reset the terminal if this is just about output */
653 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
655 case EXEC_OUTPUT_SYSLOG
:
656 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
657 case EXEC_OUTPUT_KMSG
:
658 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
659 case EXEC_OUTPUT_JOURNAL
:
660 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
661 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
663 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
664 r
= open_null_as(O_WRONLY
, fileno
);
668 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
669 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
670 * services to detect whether they are connected to the journal or not.
672 * If both stdout and stderr are connected to a stream then let's make sure to store the data
673 * about STDERR as that's usually the best way to do logging. */
675 if (fstat(fileno
, &st
) >= 0 &&
676 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
677 *journal_stream_dev
= st
.st_dev
;
678 *journal_stream_ino
= st
.st_ino
;
683 case EXEC_OUTPUT_SOCKET
:
684 assert(socket_fd
>= 0);
686 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
688 case EXEC_OUTPUT_NAMED_FD
:
689 assert(named_iofds
[fileno
] >= 0);
691 (void) fd_nonblock(named_iofds
[fileno
], false);
692 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
694 case EXEC_OUTPUT_FILE
: {
698 assert(context
->stdio_file
[fileno
]);
700 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
701 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
704 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
706 fd
= acquire_path(context
->stdio_file
[fileno
], O_WRONLY
, 0666 & ~context
->umask
);
710 return move_fd(fd
, fileno
, false);
714 assert_not_reached("Unknown error type");
718 static int chown_terminal(int fd
, uid_t uid
) {
723 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
727 /* This might fail. What matters are the results. */
728 (void) fchown(fd
, uid
, -1);
729 (void) fchmod(fd
, TTY_MODE
);
731 if (fstat(fd
, &st
) < 0)
734 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
740 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
741 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
744 assert(_saved_stdin
);
745 assert(_saved_stdout
);
747 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
751 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
752 if (saved_stdout
< 0)
755 fd
= acquire_terminal(vc
, false, false, false, DEFAULT_CONFIRM_USEC
);
759 r
= chown_terminal(fd
, getuid());
763 r
= reset_terminal_fd(fd
, true);
767 if (dup2(fd
, STDIN_FILENO
) < 0)
770 if (dup2(fd
, STDOUT_FILENO
) < 0)
777 *_saved_stdin
= saved_stdin
;
778 *_saved_stdout
= saved_stdout
;
780 saved_stdin
= saved_stdout
= -1;
785 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
788 if (err
== -ETIMEDOUT
)
789 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
792 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
796 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
797 _cleanup_close_
int fd
= -1;
801 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
805 write_confirm_error_fd(err
, fd
, u
);
808 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
812 assert(saved_stdout
);
816 if (*saved_stdin
>= 0)
817 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
820 if (*saved_stdout
>= 0)
821 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
824 *saved_stdin
= safe_close(*saved_stdin
);
825 *saved_stdout
= safe_close(*saved_stdout
);
831 CONFIRM_PRETEND_FAILURE
= -1,
832 CONFIRM_PRETEND_SUCCESS
= 0,
836 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
837 int saved_stdout
= -1, saved_stdin
= -1, r
;
838 _cleanup_free_
char *e
= NULL
;
841 /* For any internal errors, assume a positive response. */
842 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
844 write_confirm_error(r
, vc
, u
);
845 return CONFIRM_EXECUTE
;
848 /* confirm_spawn might have been disabled while we were sleeping. */
849 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
854 e
= ellipsize(cmdline
, 60, 100);
862 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
864 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
871 printf("Resuming normal execution.\n");
872 manager_disable_confirm_spawn();
876 unit_dump(u
, stdout
, " ");
877 continue; /* ask again */
879 printf("Failing execution.\n");
880 r
= CONFIRM_PRETEND_FAILURE
;
883 printf(" c - continue, proceed without asking anymore\n"
884 " D - dump, show the state of the unit\n"
885 " f - fail, don't execute the command and pretend it failed\n"
887 " i - info, show a short summary of the unit\n"
888 " j - jobs, show jobs that are in progress\n"
889 " s - skip, don't execute the command and pretend it succeeded\n"
890 " y - yes, execute the command\n");
891 continue; /* ask again */
893 printf(" Description: %s\n"
896 u
->id
, u
->description
, cmdline
);
897 continue; /* ask again */
899 manager_dump_jobs(u
->manager
, stdout
, " ");
900 continue; /* ask again */
902 /* 'n' was removed in favor of 'f'. */
903 printf("Didn't understand 'n', did you mean 'f'?\n");
904 continue; /* ask again */
906 printf("Skipping execution.\n");
907 r
= CONFIRM_PRETEND_SUCCESS
;
913 assert_not_reached("Unhandled choice");
919 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
923 static int get_fixed_user(const ExecContext
*c
, const char **user
,
924 uid_t
*uid
, gid_t
*gid
,
925 const char **home
, const char **shell
) {
934 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
935 * (i.e. are "/" or "/bin/nologin"). */
938 r
= get_user_creds_clean(&name
, uid
, gid
, home
, shell
);
946 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
956 r
= get_group_creds(&name
, gid
);
964 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
965 const char *group
, gid_t gid
,
966 gid_t
**supplementary_gids
, int *ngids
) {
970 bool keep_groups
= false;
971 gid_t
*groups
= NULL
;
972 _cleanup_free_ gid_t
*l_gids
= NULL
;
977 * If user is given, then lookup GID and supplementary groups list.
978 * We avoid NSS lookups for gid=0. Also we have to initialize groups
979 * here and as early as possible so we keep the list of supplementary
980 * groups of the caller.
982 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
983 /* First step, initialize groups from /etc/groups */
984 if (initgroups(user
, gid
) < 0)
990 if (strv_isempty(c
->supplementary_groups
))
994 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
995 * be positive, otherwise fail.
998 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
999 if (ngroups_max
<= 0) {
1003 return -EOPNOTSUPP
; /* For all other values */
1006 l_gids
= new(gid_t
, ngroups_max
);
1012 * Lookup the list of groups that the user belongs to, we
1013 * avoid NSS lookups here too for gid=0.
1016 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1021 STRV_FOREACH(i
, c
->supplementary_groups
) {
1024 if (k
>= ngroups_max
)
1028 r
= get_group_creds(&g
, l_gids
+k
);
1036 * Sets ngids to zero to drop all supplementary groups, happens
1037 * when we are under root and SupplementaryGroups= is empty.
1044 /* Otherwise get the final list of supplementary groups */
1045 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1049 *supplementary_gids
= groups
;
1057 static int enforce_groups(gid_t gid
, gid_t
*supplementary_gids
, int ngids
) {
1060 /* Handle SupplementaryGroups= if it is not empty */
1062 r
= maybe_setgroups(ngids
, supplementary_gids
);
1067 if (gid_is_valid(gid
)) {
1068 /* Then set our gids */
1069 if (setresgid(gid
, gid
, gid
) < 0)
1076 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1079 if (!uid_is_valid(uid
))
1082 /* Sets (but doesn't look up) the uid and make sure we keep the
1083 * capabilities while doing so. */
1085 if (context
->capability_ambient_set
!= 0) {
1087 /* First step: If we need to keep capabilities but
1088 * drop privileges we need to make sure we keep our
1089 * caps, while we drop privileges. */
1091 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1093 if (prctl(PR_GET_SECUREBITS
) != sb
)
1094 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1099 /* Second step: actually set the uids */
1100 if (setresuid(uid
, uid
, uid
) < 0)
1103 /* At this point we should have all necessary capabilities but
1104 are otherwise a normal user. However, the caps might got
1105 corrupted due to the setresuid() so we need clean them up
1106 later. This is done outside of this call. */
1113 static int null_conv(
1115 const struct pam_message
**msg
,
1116 struct pam_response
**resp
,
1117 void *appdata_ptr
) {
1119 /* We don't support conversations */
1121 return PAM_CONV_ERR
;
1126 static int setup_pam(
1133 int fds
[], unsigned n_fds
) {
1137 static const struct pam_conv conv
= {
1142 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1143 pam_handle_t
*handle
= NULL
;
1145 int pam_code
= PAM_SUCCESS
, r
;
1146 char **nv
, **e
= NULL
;
1147 bool close_session
= false;
1148 pid_t pam_pid
= 0, parent_pid
;
1155 /* We set up PAM in the parent process, then fork. The child
1156 * will then stay around until killed via PR_GET_PDEATHSIG or
1157 * systemd via the cgroup logic. It will then remove the PAM
1158 * session again. The parent process will exec() the actual
1159 * daemon. We do things this way to ensure that the main PID
1160 * of the daemon is the one we initially fork()ed. */
1162 r
= barrier_create(&barrier
);
1166 if (log_get_max_level() < LOG_DEBUG
)
1167 flags
|= PAM_SILENT
;
1169 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1170 if (pam_code
!= PAM_SUCCESS
) {
1176 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1177 if (pam_code
!= PAM_SUCCESS
)
1181 STRV_FOREACH(nv
, *env
) {
1182 pam_code
= pam_putenv(handle
, *nv
);
1183 if (pam_code
!= PAM_SUCCESS
)
1187 pam_code
= pam_acct_mgmt(handle
, flags
);
1188 if (pam_code
!= PAM_SUCCESS
)
1191 pam_code
= pam_open_session(handle
, flags
);
1192 if (pam_code
!= PAM_SUCCESS
)
1195 close_session
= true;
1197 e
= pam_getenvlist(handle
);
1199 pam_code
= PAM_BUF_ERR
;
1203 /* Block SIGTERM, so that we know that it won't get lost in
1206 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1208 parent_pid
= getpid_cached();
1217 int sig
, ret
= EXIT_PAM
;
1219 /* The child's job is to reset the PAM session on
1221 barrier_set_role(&barrier
, BARRIER_CHILD
);
1223 /* This string must fit in 10 chars (i.e. the length
1224 * of "/sbin/init"), to look pretty in /bin/ps */
1225 rename_process("(sd-pam)");
1227 /* Make sure we don't keep open the passed fds in this
1228 child. We assume that otherwise only those fds are
1229 open here that have been opened by PAM. */
1230 close_many(fds
, n_fds
);
1232 /* Drop privileges - we don't need any to pam_close_session
1233 * and this will make PR_SET_PDEATHSIG work in most cases.
1234 * If this fails, ignore the error - but expect sd-pam threads
1235 * to fail to exit normally */
1237 r
= maybe_setgroups(0, NULL
);
1239 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1240 if (setresgid(gid
, gid
, gid
) < 0)
1241 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1242 if (setresuid(uid
, uid
, uid
) < 0)
1243 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1245 (void) ignore_signals(SIGPIPE
, -1);
1247 /* Wait until our parent died. This will only work if
1248 * the above setresuid() succeeds, otherwise the kernel
1249 * will not allow unprivileged parents kill their privileged
1250 * children this way. We rely on the control groups kill logic
1251 * to do the rest for us. */
1252 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1255 /* Tell the parent that our setup is done. This is especially
1256 * important regarding dropping privileges. Otherwise, unit
1257 * setup might race against our setresuid(2) call.
1259 * If the parent aborted, we'll detect this below, hence ignore
1260 * return failure here. */
1261 (void) barrier_place(&barrier
);
1263 /* Check if our parent process might already have died? */
1264 if (getppid() == parent_pid
) {
1267 assert_se(sigemptyset(&ss
) >= 0);
1268 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1271 if (sigwait(&ss
, &sig
) < 0) {
1278 assert(sig
== SIGTERM
);
1283 /* If our parent died we'll end the session */
1284 if (getppid() != parent_pid
) {
1285 pam_code
= pam_close_session(handle
, flags
);
1286 if (pam_code
!= PAM_SUCCESS
)
1293 pam_end(handle
, pam_code
| flags
);
1297 barrier_set_role(&barrier
, BARRIER_PARENT
);
1299 /* If the child was forked off successfully it will do all the
1300 * cleanups, so forget about the handle here. */
1303 /* Unblock SIGTERM again in the parent */
1304 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1306 /* We close the log explicitly here, since the PAM modules
1307 * might have opened it, but we don't want this fd around. */
1310 /* Synchronously wait for the child to initialize. We don't care for
1311 * errors as we cannot recover. However, warn loudly if it happens. */
1312 if (!barrier_place_and_sync(&barrier
))
1313 log_error("PAM initialization failed");
1321 if (pam_code
!= PAM_SUCCESS
) {
1322 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1323 r
= -EPERM
; /* PAM errors do not map to errno */
1325 log_error_errno(r
, "PAM failed: %m");
1329 pam_code
= pam_close_session(handle
, flags
);
1331 pam_end(handle
, pam_code
| flags
);
1343 static void rename_process_from_path(const char *path
) {
1344 char process_name
[11];
1348 /* This resulting string must fit in 10 chars (i.e. the length
1349 * of "/sbin/init") to look pretty in /bin/ps */
1353 rename_process("(...)");
1359 /* The end of the process name is usually more
1360 * interesting, since the first bit might just be
1366 process_name
[0] = '(';
1367 memcpy(process_name
+1, p
, l
);
1368 process_name
[1+l
] = ')';
1369 process_name
[1+l
+1] = 0;
1371 rename_process(process_name
);
1374 static bool context_has_address_families(const ExecContext
*c
) {
1377 return c
->address_families_whitelist
||
1378 !set_isempty(c
->address_families
);
1381 static bool context_has_syscall_filters(const ExecContext
*c
) {
1384 return c
->syscall_whitelist
||
1385 !hashmap_isempty(c
->syscall_filter
);
1388 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1391 if (c
->no_new_privileges
)
1394 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1397 /* We need NNP if we have any form of seccomp and are unprivileged */
1398 return context_has_address_families(c
) ||
1399 c
->memory_deny_write_execute
||
1400 c
->restrict_realtime
||
1401 exec_context_restrict_namespaces_set(c
) ||
1402 c
->protect_kernel_tunables
||
1403 c
->protect_kernel_modules
||
1404 c
->private_devices
||
1405 context_has_syscall_filters(c
) ||
1406 !set_isempty(c
->syscall_archs
) ||
1407 c
->lock_personality
;
1412 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1414 if (is_seccomp_available())
1417 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1421 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1422 uint32_t negative_action
, default_action
, action
;
1428 if (!context_has_syscall_filters(c
))
1431 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1434 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1436 if (c
->syscall_whitelist
) {
1437 default_action
= negative_action
;
1438 action
= SCMP_ACT_ALLOW
;
1440 default_action
= SCMP_ACT_ALLOW
;
1441 action
= negative_action
;
1444 if (needs_ambient_hack
) {
1445 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1450 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
);
1453 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1457 if (set_isempty(c
->syscall_archs
))
1460 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1463 return seccomp_restrict_archs(c
->syscall_archs
);
1466 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1470 if (!context_has_address_families(c
))
1473 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1476 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1479 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1483 if (!c
->memory_deny_write_execute
)
1486 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1489 return seccomp_memory_deny_write_execute();
1492 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1496 if (!c
->restrict_realtime
)
1499 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1502 return seccomp_restrict_realtime();
1505 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1509 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1510 * let's protect even those systems where this is left on in the kernel. */
1512 if (!c
->protect_kernel_tunables
)
1515 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1518 return seccomp_protect_sysctl();
1521 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1525 /* Turn off module syscalls on ProtectKernelModules=yes */
1527 if (!c
->protect_kernel_modules
)
1530 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1533 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
));
1536 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1540 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1542 if (!c
->private_devices
)
1545 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1548 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
));
1551 static int apply_restrict_namespaces(Unit
*u
, const ExecContext
*c
) {
1555 if (!exec_context_restrict_namespaces_set(c
))
1558 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1561 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1564 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1565 unsigned long personality
;
1571 if (!c
->lock_personality
)
1574 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1577 personality
= c
->personality
;
1579 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1580 if (personality
== PERSONALITY_INVALID
) {
1582 r
= opinionated_personality(&personality
);
1587 return seccomp_lock_personality(personality
);
1592 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1595 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1596 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1598 if (idle_pipe
[0] >= 0) {
1601 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1603 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1606 /* Signal systemd that we are bored and want to continue. */
1607 n
= write(idle_pipe
[3], "x", 1);
1609 /* Wait for systemd to react to the signal above. */
1610 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1613 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1617 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1620 static int build_environment(
1622 const ExecContext
*c
,
1623 const ExecParameters
*p
,
1626 const char *username
,
1628 dev_t journal_stream_dev
,
1629 ino_t journal_stream_ino
,
1632 _cleanup_strv_free_
char **our_env
= NULL
;
1640 our_env
= new0(char*, 14);
1645 _cleanup_free_
char *joined
= NULL
;
1647 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1649 our_env
[n_env
++] = x
;
1651 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1653 our_env
[n_env
++] = x
;
1655 joined
= strv_join(p
->fd_names
, ":");
1659 x
= strjoin("LISTEN_FDNAMES=", joined
);
1662 our_env
[n_env
++] = x
;
1665 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1666 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1668 our_env
[n_env
++] = x
;
1670 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1672 our_env
[n_env
++] = x
;
1675 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1676 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1677 * check the database directly. */
1678 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1679 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1682 our_env
[n_env
++] = x
;
1686 x
= strappend("HOME=", home
);
1689 our_env
[n_env
++] = x
;
1693 x
= strappend("LOGNAME=", username
);
1696 our_env
[n_env
++] = x
;
1698 x
= strappend("USER=", username
);
1701 our_env
[n_env
++] = x
;
1705 x
= strappend("SHELL=", shell
);
1708 our_env
[n_env
++] = x
;
1711 if (!sd_id128_is_null(u
->invocation_id
)) {
1712 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1715 our_env
[n_env
++] = x
;
1718 if (exec_context_needs_term(c
)) {
1719 const char *tty_path
, *term
= NULL
;
1721 tty_path
= exec_context_tty_path(c
);
1723 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1724 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1725 * passes to PID 1 ends up all the way in the console login shown. */
1727 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1728 term
= getenv("TERM");
1730 term
= default_term_for_tty(tty_path
);
1732 x
= strappend("TERM=", term
);
1735 our_env
[n_env
++] = x
;
1738 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1739 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1742 our_env
[n_env
++] = x
;
1745 our_env
[n_env
++] = NULL
;
1746 assert(n_env
<= 12);
1754 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1755 _cleanup_strv_free_
char **pass_env
= NULL
;
1756 size_t n_env
= 0, n_bufsize
= 0;
1759 STRV_FOREACH(i
, c
->pass_environment
) {
1760 _cleanup_free_
char *x
= NULL
;
1766 x
= strjoin(*i
, "=", v
);
1770 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1773 pass_env
[n_env
++] = x
;
1774 pass_env
[n_env
] = NULL
;
1784 static bool exec_needs_mount_namespace(
1785 const ExecContext
*context
,
1786 const ExecParameters
*params
,
1787 ExecRuntime
*runtime
) {
1792 if (context
->root_image
)
1795 if (!strv_isempty(context
->read_write_paths
) ||
1796 !strv_isempty(context
->read_only_paths
) ||
1797 !strv_isempty(context
->inaccessible_paths
))
1800 if (context
->n_bind_mounts
> 0 ||
1801 !strv_isempty(context
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) ||
1802 !strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1803 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1804 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
) ||
1805 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CONFIGURATION
].paths
))
1808 if (context
->mount_flags
!= 0)
1811 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1814 if (context
->private_devices
||
1815 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1816 context
->protect_home
!= PROTECT_HOME_NO
||
1817 context
->protect_kernel_tunables
||
1818 context
->protect_kernel_modules
||
1819 context
->protect_control_groups
)
1822 if (context
->mount_apivfs
&& (context
->root_image
|| context
->root_directory
))
1828 static int setup_private_users(uid_t uid
, gid_t gid
) {
1829 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1830 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1831 _cleanup_close_
int unshare_ready_fd
= -1;
1832 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1838 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1839 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1840 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1841 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1842 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1843 * continues execution normally. */
1845 if (uid
!= 0 && uid_is_valid(uid
)) {
1846 r
= asprintf(&uid_map
,
1847 "0 0 1\n" /* Map root → root */
1848 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1853 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1858 if (gid
!= 0 && gid_is_valid(gid
)) {
1859 r
= asprintf(&gid_map
,
1860 "0 0 1\n" /* Map root → root */
1861 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1866 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1871 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1873 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1874 if (unshare_ready_fd
< 0)
1877 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1879 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1887 _cleanup_close_
int fd
= -1;
1891 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1892 * here, after the parent opened its own user namespace. */
1895 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1897 /* Wait until the parent unshared the user namespace */
1898 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1903 /* Disable the setgroups() system call in the child user namespace, for good. */
1904 a
= procfs_file_alloca(ppid
, "setgroups");
1905 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1907 if (errno
!= ENOENT
) {
1912 /* If the file is missing the kernel is too old, let's continue anyway. */
1914 if (write(fd
, "deny\n", 5) < 0) {
1919 fd
= safe_close(fd
);
1922 /* First write the GID map */
1923 a
= procfs_file_alloca(ppid
, "gid_map");
1924 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1929 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1933 fd
= safe_close(fd
);
1935 /* The write the UID map */
1936 a
= procfs_file_alloca(ppid
, "uid_map");
1937 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1942 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1947 _exit(EXIT_SUCCESS
);
1950 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1951 _exit(EXIT_FAILURE
);
1954 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1956 if (unshare(CLONE_NEWUSER
) < 0)
1959 /* Let the child know that the namespace is ready now */
1960 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1963 /* Try to read an error code from the child */
1964 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1967 if (n
== sizeof(r
)) { /* an error code was sent to us */
1972 if (n
!= 0) /* on success we should have read 0 bytes */
1975 r
= wait_for_terminate(pid
, &si
);
1980 /* If something strange happened with the child, let's consider this fatal, too */
1981 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1987 static int setup_exec_directory(
1988 const ExecContext
*context
,
1989 const ExecParameters
*params
,
1992 ExecDirectoryType type
,
1995 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
1996 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
1997 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
1998 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
1999 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2000 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2007 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2008 assert(exit_status
);
2010 if (!params
->prefix
[type
])
2013 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2014 if (!uid_is_valid(uid
))
2016 if (!gid_is_valid(gid
))
2020 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2021 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2022 const char *effective
;
2024 p
= strjoin(params
->prefix
[type
], "/", *rt
);
2030 r
= mkdir_parents_label(p
, 0755);
2034 if (context
->dynamic_user
&&
2035 !IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2036 _cleanup_free_
char *private_root
= NULL
, *relative
= NULL
, *parent
= NULL
;
2038 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2039 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2040 * whose UID is later on reused. To lock this down we use the same trick used by container
2041 * managers to prohibit host users to get access to files of the same UID in containers: we
2042 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2043 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2044 * to make this directory permeable for the service itself.
2046 * Specifically: for a service which wants a special directory "foo/" we first create a
2047 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2048 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2049 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2050 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2051 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2052 * disabling the access boundary for the service and making sure it only gets access to the
2053 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2055 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2056 * owned by the service itself.
2057 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2058 * files or sockets with other services. */
2060 private_root
= strjoin(params
->prefix
[type
], "/private");
2061 if (!private_root
) {
2066 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2067 r
= mkdir_safe_label(private_root
, 0700, 0, 0, false);
2071 pp
= strjoin(private_root
, "/", *rt
);
2077 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2078 r
= mkdir_parents_label(pp
, 0755);
2082 if (is_dir(p
, false) > 0 &&
2083 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2085 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2086 * it over. Most likely the service has been upgraded from one that didn't use
2087 * DynamicUser=1, to one that does. */
2089 if (rename(p
, pp
) < 0) {
2094 /* Otherwise, create the actual directory for the service */
2096 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2097 if (r
< 0 && r
!= -EEXIST
)
2101 parent
= dirname_malloc(p
);
2107 r
= path_make_relative(parent
, pp
, &relative
);
2111 /* And link it up from the original place */
2112 r
= symlink_idempotent(relative
, p
);
2119 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2120 if (r
< 0 && r
!= -EEXIST
)
2126 /* First lock down the access mode */
2127 if (chmod(effective
, context
->directories
[type
].mode
) < 0) {
2132 /* Don't change the owner of the configuration directory, as in the common case it is not written to by
2133 * a service, and shall not be writable. */
2134 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2137 /* Then, change the ownership of the whole tree, if necessary */
2138 r
= path_chown_recursive(effective
, uid
, gid
);
2146 *exit_status
= exit_status_table
[type
];
2151 static int setup_smack(
2152 const ExecContext
*context
,
2153 const ExecCommand
*command
) {
2160 if (context
->smack_process_label
) {
2161 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2165 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2167 _cleanup_free_
char *exec_label
= NULL
;
2169 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2170 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2173 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2183 static int compile_bind_mounts(
2184 const ExecContext
*context
,
2185 const ExecParameters
*params
,
2186 BindMount
**ret_bind_mounts
,
2187 unsigned *ret_n_bind_mounts
,
2188 char ***ret_empty_directories
) {
2190 _cleanup_strv_free_
char **empty_directories
= NULL
;
2191 BindMount
*bind_mounts
;
2192 unsigned n
, h
= 0, i
;
2193 ExecDirectoryType t
;
2198 assert(ret_bind_mounts
);
2199 assert(ret_n_bind_mounts
);
2200 assert(ret_empty_directories
);
2202 n
= context
->n_bind_mounts
;
2203 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2204 if (!params
->prefix
[t
])
2207 n
+= strv_length(context
->directories
[t
].paths
);
2211 *ret_bind_mounts
= NULL
;
2212 *ret_n_bind_mounts
= 0;
2213 *ret_empty_directories
= NULL
;
2217 bind_mounts
= new(BindMount
, n
);
2221 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2222 BindMount
*item
= context
->bind_mounts
+ i
;
2225 s
= strdup(item
->source
);
2231 d
= strdup(item
->destination
);
2238 bind_mounts
[h
++] = (BindMount
) {
2241 .read_only
= item
->read_only
,
2242 .recursive
= item
->recursive
,
2243 .ignore_enoent
= item
->ignore_enoent
,
2247 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2250 if (!params
->prefix
[t
])
2253 if (strv_isempty(context
->directories
[t
].paths
))
2256 if (context
->dynamic_user
&&
2257 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2260 /* So this is for a dynamic user, and we need to make sure the process can access its own
2261 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2262 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2264 private_root
= strjoin(params
->prefix
[t
], "/private");
2265 if (!private_root
) {
2270 r
= strv_consume(&empty_directories
, private_root
);
2277 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2280 if (context
->dynamic_user
&&
2281 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2282 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2284 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2297 bind_mounts
[h
++] = (BindMount
) {
2302 .ignore_enoent
= false,
2309 *ret_bind_mounts
= bind_mounts
;
2310 *ret_n_bind_mounts
= n
;
2311 *ret_empty_directories
= empty_directories
;
2313 empty_directories
= NULL
;
2318 bind_mount_free_many(bind_mounts
, h
);
2322 static int apply_mount_namespace(
2324 ExecCommand
*command
,
2325 const ExecContext
*context
,
2326 const ExecParameters
*params
,
2327 ExecRuntime
*runtime
) {
2329 _cleanup_strv_free_
char **empty_directories
= NULL
;
2330 char *tmp
= NULL
, *var
= NULL
;
2331 const char *root_dir
= NULL
, *root_image
= NULL
;
2332 NamespaceInfo ns_info
= {
2333 .ignore_protect_paths
= false,
2334 .private_dev
= context
->private_devices
,
2335 .protect_control_groups
= context
->protect_control_groups
,
2336 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2337 .protect_kernel_modules
= context
->protect_kernel_modules
,
2338 .mount_apivfs
= context
->mount_apivfs
,
2340 bool needs_sandboxing
;
2341 BindMount
*bind_mounts
= NULL
;
2342 unsigned n_bind_mounts
= 0;
2347 /* The runtime struct only contains the parent of the private /tmp,
2348 * which is non-accessible to world users. Inside of it there's a /tmp
2349 * that is sticky, and that's the one we want to use here. */
2351 if (context
->private_tmp
&& runtime
) {
2352 if (runtime
->tmp_dir
)
2353 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2354 if (runtime
->var_tmp_dir
)
2355 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2358 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2359 root_image
= context
->root_image
;
2362 root_dir
= context
->root_directory
;
2365 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2370 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2371 * sandbox info, otherwise enforce it, don't ignore protected paths and
2372 * fail if we are enable to apply the sandbox inside the mount namespace.
2374 if (!context
->dynamic_user
&& root_dir
)
2375 ns_info
.ignore_protect_paths
= true;
2377 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2379 r
= setup_namespace(root_dir
, root_image
,
2380 &ns_info
, context
->read_write_paths
,
2381 needs_sandboxing
? context
->read_only_paths
: NULL
,
2382 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2388 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2389 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2390 context
->mount_flags
,
2391 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2393 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2395 /* If we couldn't set up the namespace this is probably due to a
2396 * missing capability. In this case, silently proceeed. */
2397 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2398 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2405 static int apply_working_directory(
2406 const ExecContext
*context
,
2407 const ExecParameters
*params
,
2409 const bool needs_mount_ns
,
2415 assert(exit_status
);
2417 if (context
->working_directory_home
) {
2420 *exit_status
= EXIT_CHDIR
;
2426 } else if (context
->working_directory
)
2427 wd
= context
->working_directory
;
2431 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2432 if (!needs_mount_ns
&& context
->root_directory
)
2433 if (chroot(context
->root_directory
) < 0) {
2434 *exit_status
= EXIT_CHROOT
;
2440 d
= prefix_roota(context
->root_directory
, wd
);
2442 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2443 *exit_status
= EXIT_CHDIR
;
2450 static int setup_keyring(
2452 const ExecContext
*context
,
2453 const ExecParameters
*p
,
2454 uid_t uid
, gid_t gid
) {
2456 key_serial_t keyring
;
2463 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2464 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2465 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2466 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2467 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2468 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2470 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2473 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2476 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2477 if (keyring
== -1) {
2478 if (errno
== ENOSYS
)
2479 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2480 else if (IN_SET(errno
, EACCES
, EPERM
))
2481 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2482 else if (errno
== EDQUOT
)
2483 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2485 return log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2490 /* Populate they keyring with the invocation ID by default. */
2491 if (!sd_id128_is_null(u
->invocation_id
)) {
2494 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2496 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2498 if (keyctl(KEYCTL_SETPERM
, key
,
2499 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2500 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2501 return log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2505 /* And now, make the keyring owned by the service's user */
2506 if (uid_is_valid(uid
) || gid_is_valid(gid
))
2507 if (keyctl(KEYCTL_CHOWN
, keyring
, uid
, gid
, 0) < 0)
2508 return log_unit_error_errno(u
, errno
, "Failed to change ownership of session keyring: %m");
2510 /* When requested link the user keyring into the session keyring. */
2511 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2515 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things
2516 * set up properly by the kernel. If we don't do that then we can't create it atomically, and that
2517 * sucks for parallel execution. This mimics what pam_keyinit does, too.*/
2519 saved_uid
= getuid();
2520 saved_gid
= getgid();
2522 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2523 if (setregid(gid
, -1) < 0)
2524 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2527 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2528 if (setreuid(uid
, -1) < 0) {
2529 (void) setregid(saved_gid
, -1);
2530 return log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2534 if (keyctl(KEYCTL_LINK
,
2535 KEY_SPEC_USER_KEYRING
,
2536 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2540 (void) setreuid(saved_uid
, -1);
2541 (void) setregid(saved_gid
, -1);
2543 return log_unit_error_errno(u
, r
, "Failed to link user keyring into session keyring: %m");
2546 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2547 if (setreuid(saved_uid
, -1) < 0) {
2548 (void) setregid(saved_gid
, -1);
2549 return log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2553 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2554 if (setregid(saved_gid
, -1) < 0)
2555 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2562 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
2570 array
[(*n
)++] = pair
[0];
2572 array
[(*n
)++] = pair
[1];
2575 static int close_remaining_fds(
2576 const ExecParameters
*params
,
2577 ExecRuntime
*runtime
,
2578 DynamicCreds
*dcreds
,
2581 int *fds
, unsigned n_fds
) {
2583 unsigned n_dont_close
= 0;
2584 int dont_close
[n_fds
+ 12];
2588 if (params
->stdin_fd
>= 0)
2589 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2590 if (params
->stdout_fd
>= 0)
2591 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2592 if (params
->stderr_fd
>= 0)
2593 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2596 dont_close
[n_dont_close
++] = socket_fd
;
2598 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2599 n_dont_close
+= n_fds
;
2603 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2607 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2609 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2612 if (user_lookup_fd
>= 0)
2613 dont_close
[n_dont_close
++] = user_lookup_fd
;
2615 return close_all_fds(dont_close
, n_dont_close
);
2618 static int send_user_lookup(
2626 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2627 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2630 if (user_lookup_fd
< 0)
2633 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2636 if (writev(user_lookup_fd
,
2638 IOVEC_INIT(&uid
, sizeof(uid
)),
2639 IOVEC_INIT(&gid
, sizeof(gid
)),
2640 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2646 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2653 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2658 if (!c
->working_directory_home
)
2662 /* Hardcode /root as home directory for UID 0 */
2667 r
= get_home_dir(buf
);
2675 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2676 _cleanup_strv_free_
char ** list
= NULL
;
2677 ExecDirectoryType t
;
2684 assert(c
->dynamic_user
);
2686 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2687 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2690 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2693 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2699 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2702 if (t
== EXEC_DIRECTORY_RUNTIME
)
2703 e
= strjoin(p
->prefix
[t
], "/", *i
);
2705 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2709 r
= strv_consume(&list
, e
);
2721 static int exec_child(
2723 ExecCommand
*command
,
2724 const ExecContext
*context
,
2725 const ExecParameters
*params
,
2726 ExecRuntime
*runtime
,
2727 DynamicCreds
*dcreds
,
2732 unsigned n_storage_fds
,
2733 unsigned n_socket_fds
,
2738 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2739 _cleanup_free_
char *home_buffer
= NULL
;
2740 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2741 const char *username
= NULL
, *groupname
= NULL
;
2742 const char *home
= NULL
, *shell
= NULL
;
2743 dev_t journal_stream_dev
= 0;
2744 ino_t journal_stream_ino
= 0;
2745 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2746 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2747 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2748 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2750 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2751 bool use_selinux
= false;
2754 bool use_smack
= false;
2757 bool use_apparmor
= false;
2759 uid_t uid
= UID_INVALID
;
2760 gid_t gid
= GID_INVALID
;
2761 int i
, r
, ngids
= 0;
2763 ExecDirectoryType dt
;
2770 assert(exit_status
);
2772 rename_process_from_path(command
->path
);
2774 /* We reset exactly these signals, since they are the
2775 * only ones we set to SIG_IGN in the main daemon. All
2776 * others we leave untouched because we set them to
2777 * SIG_DFL or a valid handler initially, both of which
2778 * will be demoted to SIG_DFL. */
2779 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2780 SIGNALS_IGNORE
, -1);
2782 if (context
->ignore_sigpipe
)
2783 (void) ignore_signals(SIGPIPE
, -1);
2785 r
= reset_signal_mask();
2787 *exit_status
= EXIT_SIGNAL_MASK
;
2788 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2791 if (params
->idle_pipe
)
2792 do_idle_pipe_dance(params
->idle_pipe
);
2794 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2795 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2796 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2797 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2800 log_set_open_when_needed(true);
2802 /* In case anything used libc syslog(), close this here, too */
2805 n_fds
= n_storage_fds
+ n_socket_fds
;
2806 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2808 *exit_status
= EXIT_FDS
;
2809 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2812 if (!context
->same_pgrp
)
2814 *exit_status
= EXIT_SETSID
;
2815 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2818 exec_context_tty_reset(context
, params
);
2820 if (unit_shall_confirm_spawn(unit
)) {
2821 const char *vc
= params
->confirm_spawn
;
2822 _cleanup_free_
char *cmdline
= NULL
;
2824 cmdline
= exec_command_line(argv
);
2826 *exit_status
= EXIT_MEMORY
;
2830 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2831 if (r
!= CONFIRM_EXECUTE
) {
2832 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2833 *exit_status
= EXIT_SUCCESS
;
2836 *exit_status
= EXIT_CONFIRM
;
2837 log_unit_error(unit
, "Execution cancelled by the user");
2842 if (context
->dynamic_user
&& dcreds
) {
2843 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2845 /* Make sure we bypass our own NSS module for any NSS checks */
2846 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2847 *exit_status
= EXIT_USER
;
2848 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2851 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2853 *exit_status
= EXIT_MEMORY
;
2857 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2859 *exit_status
= EXIT_USER
;
2861 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2864 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2867 if (!uid_is_valid(uid
)) {
2868 *exit_status
= EXIT_USER
;
2869 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
2873 if (!gid_is_valid(gid
)) {
2874 *exit_status
= EXIT_USER
;
2875 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
2880 username
= dcreds
->user
->name
;
2883 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2885 *exit_status
= EXIT_USER
;
2886 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
2889 r
= get_fixed_group(context
, &groupname
, &gid
);
2891 *exit_status
= EXIT_GROUP
;
2892 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
2896 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2897 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2898 &supplementary_gids
, &ngids
);
2900 *exit_status
= EXIT_GROUP
;
2901 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
2904 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2906 *exit_status
= EXIT_USER
;
2907 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
2910 user_lookup_fd
= safe_close(user_lookup_fd
);
2912 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2914 *exit_status
= EXIT_CHDIR
;
2915 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
2918 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2919 * must sure to drop O_NONBLOCK */
2921 (void) fd_nonblock(socket_fd
, false);
2923 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2925 *exit_status
= EXIT_STDIN
;
2926 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
2929 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2931 *exit_status
= EXIT_STDOUT
;
2932 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
2935 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2937 *exit_status
= EXIT_STDERR
;
2938 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
2941 if (params
->cgroup_path
) {
2942 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2944 *exit_status
= EXIT_CGROUP
;
2945 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", params
->cgroup_path
);
2949 if (context
->oom_score_adjust_set
) {
2950 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2952 /* When we can't make this change due to EPERM, then
2953 * let's silently skip over it. User namespaces
2954 * prohibit write access to this file, and we
2955 * shouldn't trip up over that. */
2957 sprintf(t
, "%i", context
->oom_score_adjust
);
2958 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2959 if (IN_SET(r
, -EPERM
, -EACCES
))
2960 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2962 *exit_status
= EXIT_OOM_ADJUST
;
2963 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
2967 if (context
->nice_set
)
2968 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2969 *exit_status
= EXIT_NICE
;
2970 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
2973 if (context
->cpu_sched_set
) {
2974 struct sched_param param
= {
2975 .sched_priority
= context
->cpu_sched_priority
,
2978 r
= sched_setscheduler(0,
2979 context
->cpu_sched_policy
|
2980 (context
->cpu_sched_reset_on_fork
?
2981 SCHED_RESET_ON_FORK
: 0),
2984 *exit_status
= EXIT_SETSCHEDULER
;
2985 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
2989 if (context
->cpuset
)
2990 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2991 *exit_status
= EXIT_CPUAFFINITY
;
2992 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
2995 if (context
->ioprio_set
)
2996 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2997 *exit_status
= EXIT_IOPRIO
;
2998 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3001 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3002 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3003 *exit_status
= EXIT_TIMERSLACK
;
3004 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3007 if (context
->personality
!= PERSONALITY_INVALID
) {
3008 r
= safe_personality(context
->personality
);
3010 *exit_status
= EXIT_PERSONALITY
;
3011 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3015 if (context
->utmp_id
)
3016 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3018 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3019 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3023 if (context
->user
) {
3024 r
= chown_terminal(STDIN_FILENO
, uid
);
3026 *exit_status
= EXIT_STDIN
;
3027 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3031 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroupsv1
3032 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3033 * safe. On cgroupsv2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3034 * touch a single hierarchy too. */
3035 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3036 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3038 *exit_status
= EXIT_CGROUP
;
3039 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3043 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3044 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3046 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3049 r
= build_environment(
3061 *exit_status
= EXIT_MEMORY
;
3065 r
= build_pass_environment(context
, &pass_env
);
3067 *exit_status
= EXIT_MEMORY
;
3071 accum_env
= strv_env_merge(5,
3072 params
->environment
,
3075 context
->environment
,
3079 *exit_status
= EXIT_MEMORY
;
3082 accum_env
= strv_env_clean(accum_env
);
3084 (void) umask(context
->umask
);
3086 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3088 *exit_status
= EXIT_KEYRING
;
3089 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3092 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3093 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3095 /* 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 */
3096 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3098 /* 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 */
3099 if (needs_ambient_hack
)
3100 needs_setuid
= false;
3102 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3104 if (needs_sandboxing
) {
3105 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3106 * present. The actual MAC context application will happen later, as late as possible, to avoid
3107 * impacting our own code paths. */
3110 use_selinux
= mac_selinux_use();
3113 use_smack
= mac_smack_use();
3116 use_apparmor
= mac_apparmor_use();
3121 if (context
->pam_name
&& username
) {
3122 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3124 *exit_status
= EXIT_PAM
;
3125 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3130 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3131 if (ns_type_supported(NAMESPACE_NET
)) {
3132 r
= setup_netns(runtime
->netns_storage_socket
);
3134 *exit_status
= EXIT_NETWORK
;
3135 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3138 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3141 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3142 if (needs_mount_namespace
) {
3143 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3145 *exit_status
= EXIT_NAMESPACE
;
3146 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3150 /* Apply just after mount namespace setup */
3151 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3153 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3155 /* Drop groups as early as possbile */
3157 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3159 *exit_status
= EXIT_GROUP
;
3160 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3164 if (needs_sandboxing
) {
3166 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3167 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3169 *exit_status
= EXIT_SELINUX_CONTEXT
;
3170 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3175 if (context
->private_users
) {
3176 r
= setup_private_users(uid
, gid
);
3178 *exit_status
= EXIT_USER
;
3179 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3184 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3185 * more aggressive this time since socket_fd and the netns fds we don't need anymore. The custom endpoint fd
3186 * was needed to upload the policy and can now be closed as well. */
3187 r
= close_all_fds(fds
, n_fds
);
3189 r
= shift_fds(fds
, n_fds
);
3191 r
= flags_fds(fds
, n_storage_fds
, n_socket_fds
, context
->non_blocking
);
3193 *exit_status
= EXIT_FDS
;
3194 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3197 secure_bits
= context
->secure_bits
;
3199 if (needs_sandboxing
) {
3202 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3204 if (!context
->rlimit
[i
])
3207 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
3209 *exit_status
= EXIT_LIMITS
;
3210 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit %s: %m", rlimit_to_string(i
));
3214 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
3215 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3216 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3217 *exit_status
= EXIT_LIMITS
;
3218 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3223 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3224 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3226 r
= setup_smack(context
, command
);
3228 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3229 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3234 bset
= context
->capability_bounding_set
;
3235 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3236 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3237 * instead of us doing that */
3238 if (needs_ambient_hack
)
3239 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3240 (UINT64_C(1) << CAP_SETUID
) |
3241 (UINT64_C(1) << CAP_SETGID
);
3243 if (!cap_test_all(bset
)) {
3244 r
= capability_bounding_set_drop(bset
, false);
3246 *exit_status
= EXIT_CAPABILITIES
;
3247 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3251 /* This is done before enforce_user, but ambient set
3252 * does not survive over setresuid() if keep_caps is not set. */
3253 if (!needs_ambient_hack
&&
3254 context
->capability_ambient_set
!= 0) {
3255 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3257 *exit_status
= EXIT_CAPABILITIES
;
3258 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3264 if (context
->user
) {
3265 r
= enforce_user(context
, uid
);
3267 *exit_status
= EXIT_USER
;
3268 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3271 if (!needs_ambient_hack
&&
3272 context
->capability_ambient_set
!= 0) {
3274 /* Fix the ambient capabilities after user change. */
3275 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3277 *exit_status
= EXIT_CAPABILITIES
;
3278 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3281 /* If we were asked to change user and ambient capabilities
3282 * were requested, we had to add keep-caps to the securebits
3283 * so that we would maintain the inherited capability set
3284 * through the setresuid(). Make sure that the bit is added
3285 * also to the context secure_bits so that we don't try to
3286 * drop the bit away next. */
3288 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3293 if (needs_sandboxing
) {
3294 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3295 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3296 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3297 * are restricted. */
3301 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3304 r
= setexeccon(exec_context
);
3306 *exit_status
= EXIT_SELINUX_CONTEXT
;
3307 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3314 if (use_apparmor
&& context
->apparmor_profile
) {
3315 r
= aa_change_onexec(context
->apparmor_profile
);
3316 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3317 *exit_status
= EXIT_APPARMOR_PROFILE
;
3318 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3323 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3324 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3325 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3326 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3327 *exit_status
= EXIT_SECUREBITS
;
3328 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3331 if (context_has_no_new_privileges(context
))
3332 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3333 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3334 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3338 r
= apply_address_families(unit
, context
);
3340 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3341 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3344 r
= apply_memory_deny_write_execute(unit
, context
);
3346 *exit_status
= EXIT_SECCOMP
;
3347 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3350 r
= apply_restrict_realtime(unit
, context
);
3352 *exit_status
= EXIT_SECCOMP
;
3353 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3356 r
= apply_restrict_namespaces(unit
, context
);
3358 *exit_status
= EXIT_SECCOMP
;
3359 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3362 r
= apply_protect_sysctl(unit
, context
);
3364 *exit_status
= EXIT_SECCOMP
;
3365 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3368 r
= apply_protect_kernel_modules(unit
, context
);
3370 *exit_status
= EXIT_SECCOMP
;
3371 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3374 r
= apply_private_devices(unit
, context
);
3376 *exit_status
= EXIT_SECCOMP
;
3377 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3380 r
= apply_syscall_archs(unit
, context
);
3382 *exit_status
= EXIT_SECCOMP
;
3383 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3386 r
= apply_lock_personality(unit
, context
);
3388 *exit_status
= EXIT_SECCOMP
;
3389 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3392 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3393 * by the filter as little as possible. */
3394 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3396 *exit_status
= EXIT_SECCOMP
;
3397 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3402 if (!strv_isempty(context
->unset_environment
)) {
3405 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3407 *exit_status
= EXIT_MEMORY
;
3411 strv_free(accum_env
);
3415 final_argv
= replace_env_argv(argv
, accum_env
);
3417 *exit_status
= EXIT_MEMORY
;
3421 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
3422 _cleanup_free_
char *line
;
3424 line
= exec_command_line(final_argv
);
3426 log_struct(LOG_DEBUG
,
3427 "EXECUTABLE=%s", command
->path
,
3428 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3430 LOG_UNIT_INVOCATION_ID(unit
),
3435 execve(command
->path
, final_argv
, accum_env
);
3437 if (errno
== ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3439 log_struct_errno(LOG_INFO
, errno
,
3440 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3442 LOG_UNIT_INVOCATION_ID(unit
),
3443 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3445 "EXECUTABLE=%s", command
->path
,
3451 *exit_status
= EXIT_EXEC
;
3452 return log_unit_error_errno(unit
, errno
, "Failed to execute command: %m");
3455 int exec_spawn(Unit
*unit
,
3456 ExecCommand
*command
,
3457 const ExecContext
*context
,
3458 const ExecParameters
*params
,
3459 ExecRuntime
*runtime
,
3460 DynamicCreds
*dcreds
,
3463 _cleanup_strv_free_
char **files_env
= NULL
;
3465 unsigned n_storage_fds
= 0, n_socket_fds
= 0;
3466 _cleanup_free_
char *line
= NULL
;
3468 int named_iofds
[3] = { -1, -1, -1 };
3477 assert(params
->fds
|| (params
->n_storage_fds
+ params
->n_socket_fds
<= 0));
3479 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3480 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3481 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3483 if (params
->n_socket_fds
> 1) {
3484 log_unit_error(unit
, "Got more than one socket.");
3488 if (params
->n_socket_fds
== 0) {
3489 log_unit_error(unit
, "Got no socket.");
3493 socket_fd
= params
->fds
[0];
3497 n_storage_fds
= params
->n_storage_fds
;
3498 n_socket_fds
= params
->n_socket_fds
;
3501 r
= exec_context_named_iofds(unit
, context
, params
, named_iofds
);
3503 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3505 r
= exec_context_load_environment(unit
, context
, &files_env
);
3507 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3509 argv
= params
->argv
?: command
->argv
;
3510 line
= exec_command_line(argv
);
3514 log_struct(LOG_DEBUG
,
3515 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3516 "EXECUTABLE=%s", command
->path
,
3518 LOG_UNIT_INVOCATION_ID(unit
),
3523 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3526 int exit_status
= EXIT_SUCCESS
;
3528 r
= exec_child(unit
,
3541 unit
->manager
->user_lookup_fds
[1],
3545 log_struct_errno(LOG_ERR
, r
,
3546 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3548 LOG_UNIT_INVOCATION_ID(unit
),
3549 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3550 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3552 "EXECUTABLE=%s", command
->path
,
3559 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3561 /* We add the new process to the cgroup both in the child (so
3562 * that we can be sure that no user code is ever executed
3563 * outside of the cgroup) and in the parent (so that we can be
3564 * sure that when we kill the cgroup the process will be
3566 if (params
->cgroup_path
)
3567 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3569 exec_status_start(&command
->exec_status
, pid
);
3575 void exec_context_init(ExecContext
*c
) {
3576 ExecDirectoryType i
;
3581 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3582 c
->cpu_sched_policy
= SCHED_OTHER
;
3583 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3584 c
->syslog_level_prefix
= true;
3585 c
->ignore_sigpipe
= true;
3586 c
->timer_slack_nsec
= NSEC_INFINITY
;
3587 c
->personality
= PERSONALITY_INVALID
;
3588 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3589 c
->directories
[i
].mode
= 0755;
3590 c
->capability_bounding_set
= CAP_ALL
;
3591 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
3592 c
->log_level_max
= -1;
3595 void exec_context_done(ExecContext
*c
) {
3596 ExecDirectoryType i
;
3601 c
->environment
= strv_free(c
->environment
);
3602 c
->environment_files
= strv_free(c
->environment_files
);
3603 c
->pass_environment
= strv_free(c
->pass_environment
);
3604 c
->unset_environment
= strv_free(c
->unset_environment
);
3606 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
3607 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
3609 for (l
= 0; l
< 3; l
++) {
3610 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3611 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3614 c
->working_directory
= mfree(c
->working_directory
);
3615 c
->root_directory
= mfree(c
->root_directory
);
3616 c
->root_image
= mfree(c
->root_image
);
3617 c
->tty_path
= mfree(c
->tty_path
);
3618 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3619 c
->user
= mfree(c
->user
);
3620 c
->group
= mfree(c
->group
);
3622 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3624 c
->pam_name
= mfree(c
->pam_name
);
3626 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3627 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3628 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3630 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3632 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3634 c
->utmp_id
= mfree(c
->utmp_id
);
3635 c
->selinux_context
= mfree(c
->selinux_context
);
3636 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3637 c
->smack_process_label
= mfree(c
->smack_process_label
);
3639 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3640 c
->syscall_archs
= set_free(c
->syscall_archs
);
3641 c
->address_families
= set_free(c
->address_families
);
3643 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3644 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3646 c
->log_level_max
= -1;
3648 exec_context_free_log_extra_fields(c
);
3650 c
->stdin_data
= mfree(c
->stdin_data
);
3651 c
->stdin_data_size
= 0;
3654 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
3659 if (!runtime_prefix
)
3662 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3663 _cleanup_free_
char *p
;
3665 p
= strjoin(runtime_prefix
, "/", *i
);
3669 /* We execute this synchronously, since we need to be sure this is gone when we start the service
3671 (void) rm_rf(p
, REMOVE_ROOT
);
3677 void exec_command_done(ExecCommand
*c
) {
3680 c
->path
= mfree(c
->path
);
3682 c
->argv
= strv_free(c
->argv
);
3685 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3688 for (i
= 0; i
< n
; i
++)
3689 exec_command_done(c
+i
);
3692 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3696 LIST_REMOVE(command
, c
, i
);
3697 exec_command_done(i
);
3704 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3707 for (i
= 0; i
< n
; i
++)
3708 c
[i
] = exec_command_free_list(c
[i
]);
3711 typedef struct InvalidEnvInfo
{
3716 static void invalid_env(const char *p
, void *userdata
) {
3717 InvalidEnvInfo
*info
= userdata
;
3719 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3722 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3728 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3731 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3734 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3737 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3740 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3743 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3750 int exec_context_named_iofds(Unit
*unit
, const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3751 unsigned i
, targets
;
3752 const char* stdio_fdname
[3];
3758 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3759 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3760 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3762 for (i
= 0; i
< 3; i
++)
3763 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3765 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3767 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3768 if (named_iofds
[STDIN_FILENO
] < 0 &&
3769 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3770 stdio_fdname
[STDIN_FILENO
] &&
3771 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3773 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3776 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3777 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3778 stdio_fdname
[STDOUT_FILENO
] &&
3779 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3781 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3784 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3785 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3786 stdio_fdname
[STDERR_FILENO
] &&
3787 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3789 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3793 return targets
== 0 ? 0 : -ENOENT
;
3796 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
3797 char **i
, **r
= NULL
;
3802 STRV_FOREACH(i
, c
->environment_files
) {
3806 bool ignore
= false;
3808 _cleanup_globfree_ glob_t pglob
= {};
3817 if (!path_is_absolute(fn
)) {
3825 /* Filename supports globbing, take all matching files */
3826 k
= safe_glob(fn
, 0, &pglob
);
3835 /* When we don't match anything, -ENOENT should be returned */
3836 assert(pglob
.gl_pathc
> 0);
3838 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
3839 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3847 /* Log invalid environment variables with filename */
3849 InvalidEnvInfo info
= {
3851 .path
= pglob
.gl_pathv
[n
]
3854 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3862 m
= strv_env_merge(2, r
, p
);
3878 static bool tty_may_match_dev_console(const char *tty
) {
3879 _cleanup_free_
char *active
= NULL
;
3885 tty
= skip_dev_prefix(tty
);
3887 /* trivial identity? */
3888 if (streq(tty
, "console"))
3891 console
= resolve_dev_console(&active
);
3892 /* if we could not resolve, assume it may */
3896 /* "tty0" means the active VC, so it may be the same sometimes */
3897 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3900 bool exec_context_may_touch_console(ExecContext
*ec
) {
3902 return (ec
->tty_reset
||
3904 ec
->tty_vt_disallocate
||
3905 is_terminal_input(ec
->std_input
) ||
3906 is_terminal_output(ec
->std_output
) ||
3907 is_terminal_output(ec
->std_error
)) &&
3908 tty_may_match_dev_console(exec_context_tty_path(ec
));
3911 static void strv_fprintf(FILE *f
, char **l
) {
3917 fprintf(f
, " %s", *g
);
3920 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3921 ExecDirectoryType dt
;
3929 prefix
= strempty(prefix
);
3933 "%sWorkingDirectory: %s\n"
3934 "%sRootDirectory: %s\n"
3935 "%sNonBlocking: %s\n"
3936 "%sPrivateTmp: %s\n"
3937 "%sPrivateDevices: %s\n"
3938 "%sProtectKernelTunables: %s\n"
3939 "%sProtectKernelModules: %s\n"
3940 "%sProtectControlGroups: %s\n"
3941 "%sPrivateNetwork: %s\n"
3942 "%sPrivateUsers: %s\n"
3943 "%sProtectHome: %s\n"
3944 "%sProtectSystem: %s\n"
3945 "%sMountAPIVFS: %s\n"
3946 "%sIgnoreSIGPIPE: %s\n"
3947 "%sMemoryDenyWriteExecute: %s\n"
3948 "%sRestrictRealtime: %s\n"
3949 "%sKeyringMode: %s\n",
3951 prefix
, c
->working_directory
? c
->working_directory
: "/",
3952 prefix
, c
->root_directory
? c
->root_directory
: "/",
3953 prefix
, yes_no(c
->non_blocking
),
3954 prefix
, yes_no(c
->private_tmp
),
3955 prefix
, yes_no(c
->private_devices
),
3956 prefix
, yes_no(c
->protect_kernel_tunables
),
3957 prefix
, yes_no(c
->protect_kernel_modules
),
3958 prefix
, yes_no(c
->protect_control_groups
),
3959 prefix
, yes_no(c
->private_network
),
3960 prefix
, yes_no(c
->private_users
),
3961 prefix
, protect_home_to_string(c
->protect_home
),
3962 prefix
, protect_system_to_string(c
->protect_system
),
3963 prefix
, yes_no(c
->mount_apivfs
),
3964 prefix
, yes_no(c
->ignore_sigpipe
),
3965 prefix
, yes_no(c
->memory_deny_write_execute
),
3966 prefix
, yes_no(c
->restrict_realtime
),
3967 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
));
3970 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3972 STRV_FOREACH(e
, c
->environment
)
3973 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3975 STRV_FOREACH(e
, c
->environment_files
)
3976 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3978 STRV_FOREACH(e
, c
->pass_environment
)
3979 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3981 STRV_FOREACH(e
, c
->unset_environment
)
3982 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
3984 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
3986 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3987 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
3989 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
3990 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
3998 if (c
->oom_score_adjust_set
)
4000 "%sOOMScoreAdjust: %i\n",
4001 prefix
, c
->oom_score_adjust
);
4003 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4005 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
4006 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4007 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
4008 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4011 if (c
->ioprio_set
) {
4012 _cleanup_free_
char *class_str
= NULL
;
4014 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4016 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4018 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4021 if (c
->cpu_sched_set
) {
4022 _cleanup_free_
char *policy_str
= NULL
;
4024 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4026 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4029 "%sCPUSchedulingPriority: %i\n"
4030 "%sCPUSchedulingResetOnFork: %s\n",
4031 prefix
, c
->cpu_sched_priority
,
4032 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4036 fprintf(f
, "%sCPUAffinity:", prefix
);
4037 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4038 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4039 fprintf(f
, " %u", i
);
4043 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4044 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4047 "%sStandardInput: %s\n"
4048 "%sStandardOutput: %s\n"
4049 "%sStandardError: %s\n",
4050 prefix
, exec_input_to_string(c
->std_input
),
4051 prefix
, exec_output_to_string(c
->std_output
),
4052 prefix
, exec_output_to_string(c
->std_error
));
4054 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4055 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4056 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4057 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4058 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4059 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4061 if (c
->std_input
== EXEC_INPUT_FILE
)
4062 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4063 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4064 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4065 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4066 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4072 "%sTTYVHangup: %s\n"
4073 "%sTTYVTDisallocate: %s\n",
4074 prefix
, c
->tty_path
,
4075 prefix
, yes_no(c
->tty_reset
),
4076 prefix
, yes_no(c
->tty_vhangup
),
4077 prefix
, yes_no(c
->tty_vt_disallocate
));
4079 if (IN_SET(c
->std_output
,
4082 EXEC_OUTPUT_JOURNAL
,
4083 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4084 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4085 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4086 IN_SET(c
->std_error
,
4089 EXEC_OUTPUT_JOURNAL
,
4090 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4091 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4092 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4094 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4096 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4098 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4100 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4102 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4105 if (c
->log_level_max
>= 0) {
4106 _cleanup_free_
char *t
= NULL
;
4108 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4110 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4113 if (c
->n_log_extra_fields
> 0) {
4116 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4117 fprintf(f
, "%sLogExtraFields: ", prefix
);
4118 fwrite(c
->log_extra_fields
[j
].iov_base
,
4119 1, c
->log_extra_fields
[j
].iov_len
,
4125 if (c
->secure_bits
) {
4126 _cleanup_free_
char *str
= NULL
;
4128 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4130 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4133 if (c
->capability_bounding_set
!= CAP_ALL
) {
4134 _cleanup_free_
char *str
= NULL
;
4136 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4138 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4141 if (c
->capability_ambient_set
!= 0) {
4142 _cleanup_free_
char *str
= NULL
;
4144 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4146 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4150 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4152 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4154 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4156 if (!strv_isempty(c
->supplementary_groups
)) {
4157 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4158 strv_fprintf(f
, c
->supplementary_groups
);
4163 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4165 if (strv_length(c
->read_write_paths
) > 0) {
4166 fprintf(f
, "%sReadWritePaths:", prefix
);
4167 strv_fprintf(f
, c
->read_write_paths
);
4171 if (strv_length(c
->read_only_paths
) > 0) {
4172 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4173 strv_fprintf(f
, c
->read_only_paths
);
4177 if (strv_length(c
->inaccessible_paths
) > 0) {
4178 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4179 strv_fprintf(f
, c
->inaccessible_paths
);
4183 if (c
->n_bind_mounts
> 0)
4184 for (i
= 0; i
< c
->n_bind_mounts
; i
++) {
4185 fprintf(f
, "%s%s: %s:%s:%s\n", prefix
,
4186 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4187 c
->bind_mounts
[i
].source
,
4188 c
->bind_mounts
[i
].destination
,
4189 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4194 "%sUtmpIdentifier: %s\n",
4195 prefix
, c
->utmp_id
);
4197 if (c
->selinux_context
)
4199 "%sSELinuxContext: %s%s\n",
4200 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4202 if (c
->apparmor_profile
)
4204 "%sAppArmorProfile: %s%s\n",
4205 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4207 if (c
->smack_process_label
)
4209 "%sSmackProcessLabel: %s%s\n",
4210 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4212 if (c
->personality
!= PERSONALITY_INVALID
)
4214 "%sPersonality: %s\n",
4215 prefix
, strna(personality_to_string(c
->personality
)));
4218 "%sLockPersonality: %s\n",
4219 prefix
, yes_no(c
->lock_personality
));
4221 if (c
->syscall_filter
) {
4229 "%sSystemCallFilter: ",
4232 if (!c
->syscall_whitelist
)
4236 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4237 _cleanup_free_
char *name
= NULL
;
4238 const char *errno_name
= NULL
;
4239 int num
= PTR_TO_INT(val
);
4246 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4247 fputs(strna(name
), f
);
4250 errno_name
= errno_to_name(num
);
4252 fprintf(f
, ":%s", errno_name
);
4254 fprintf(f
, ":%d", num
);
4262 if (c
->syscall_archs
) {
4269 "%sSystemCallArchitectures:",
4273 SET_FOREACH(id
, c
->syscall_archs
, j
)
4274 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4279 if (exec_context_restrict_namespaces_set(c
)) {
4280 _cleanup_free_
char *s
= NULL
;
4282 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
4284 fprintf(f
, "%sRestrictNamespaces: %s\n",
4288 if (c
->syscall_errno
> 0) {
4289 const char *errno_name
;
4291 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4293 errno_name
= errno_to_name(c
->syscall_errno
);
4295 fprintf(f
, "%s\n", errno_name
);
4297 fprintf(f
, "%d\n", c
->syscall_errno
);
4300 if (c
->apparmor_profile
)
4302 "%sAppArmorProfile: %s%s\n",
4303 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4306 bool exec_context_maintains_privileges(ExecContext
*c
) {
4309 /* Returns true if the process forked off would run under
4310 * an unchanged UID or as root. */
4315 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4321 int exec_context_get_effective_ioprio(ExecContext
*c
) {
4329 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4331 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4336 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4341 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4342 free(c
->log_extra_fields
[l
].iov_base
);
4343 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4344 c
->n_log_extra_fields
= 0;
4347 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4352 dual_timestamp_get(&s
->start_timestamp
);
4355 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
4358 if (s
->pid
&& s
->pid
!= pid
)
4362 dual_timestamp_get(&s
->exit_timestamp
);
4368 if (context
->utmp_id
)
4369 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4371 exec_context_tty_reset(context
, NULL
);
4375 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
4376 char buf
[FORMAT_TIMESTAMP_MAX
];
4384 prefix
= strempty(prefix
);
4387 "%sPID: "PID_FMT
"\n",
4390 if (dual_timestamp_is_set(&s
->start_timestamp
))
4392 "%sStart Timestamp: %s\n",
4393 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4395 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4397 "%sExit Timestamp: %s\n"
4399 "%sExit Status: %i\n",
4400 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4401 prefix
, sigchld_code_to_string(s
->code
),
4405 char *exec_command_line(char **argv
) {
4413 STRV_FOREACH(a
, argv
)
4421 STRV_FOREACH(a
, argv
) {
4428 if (strpbrk(*a
, WHITESPACE
)) {
4439 /* FIXME: this doesn't really handle arguments that have
4440 * spaces and ticks in them */
4445 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4446 _cleanup_free_
char *cmd
= NULL
;
4447 const char *prefix2
;
4452 prefix
= strempty(prefix
);
4453 prefix2
= strjoina(prefix
, "\t");
4455 cmd
= exec_command_line(c
->argv
);
4457 "%sCommand Line: %s\n",
4458 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4460 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4463 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4466 prefix
= strempty(prefix
);
4468 LIST_FOREACH(command
, c
, c
)
4469 exec_command_dump(c
, f
, prefix
);
4472 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4479 /* It's kind of important, that we keep the order here */
4480 LIST_FIND_TAIL(command
, *l
, end
);
4481 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4486 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4494 l
= strv_new_ap(path
, ap
);
4515 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4516 _cleanup_strv_free_
char **l
= NULL
;
4524 l
= strv_new_ap(path
, ap
);
4530 r
= strv_extend_strv(&c
->argv
, l
, false);
4538 static int exec_runtime_allocate(ExecRuntime
**rt
) {
4543 *rt
= new0(ExecRuntime
, 1);
4548 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
4553 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
4563 if (!c
->private_network
&& !c
->private_tmp
)
4566 r
= exec_runtime_allocate(rt
);
4570 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
4571 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
4575 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
4576 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
4584 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
4586 assert(r
->n_ref
> 0);
4592 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
4597 assert(r
->n_ref
> 0);
4604 free(r
->var_tmp_dir
);
4605 safe_close_pair(r
->netns_storage_socket
);
4609 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
4618 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
4620 if (rt
->var_tmp_dir
)
4621 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
4623 if (rt
->netns_storage_socket
[0] >= 0) {
4626 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4630 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
4633 if (rt
->netns_storage_socket
[1] >= 0) {
4636 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
4640 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
4646 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
4653 if (streq(key
, "tmp-dir")) {
4656 r
= exec_runtime_allocate(rt
);
4660 copy
= strdup(value
);
4664 free((*rt
)->tmp_dir
);
4665 (*rt
)->tmp_dir
= copy
;
4667 } else if (streq(key
, "var-tmp-dir")) {
4670 r
= exec_runtime_allocate(rt
);
4674 copy
= strdup(value
);
4678 free((*rt
)->var_tmp_dir
);
4679 (*rt
)->var_tmp_dir
= copy
;
4681 } else if (streq(key
, "netns-socket-0")) {
4684 r
= exec_runtime_allocate(rt
);
4688 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4689 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4691 safe_close((*rt
)->netns_storage_socket
[0]);
4692 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4694 } else if (streq(key
, "netns-socket-1")) {
4697 r
= exec_runtime_allocate(rt
);
4701 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4702 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4704 safe_close((*rt
)->netns_storage_socket
[1]);
4705 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
4713 static void *remove_tmpdir_thread(void *p
) {
4714 _cleanup_free_
char *path
= p
;
4716 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4720 void exec_runtime_destroy(ExecRuntime
*rt
) {
4726 /* If there are multiple users of this, let's leave the stuff around */
4731 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4733 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4735 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4742 if (rt
->var_tmp_dir
) {
4743 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4745 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4747 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4748 free(rt
->var_tmp_dir
);
4751 rt
->var_tmp_dir
= NULL
;
4754 safe_close_pair(rt
->netns_storage_socket
);
4757 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4758 [EXEC_INPUT_NULL
] = "null",
4759 [EXEC_INPUT_TTY
] = "tty",
4760 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4761 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4762 [EXEC_INPUT_SOCKET
] = "socket",
4763 [EXEC_INPUT_NAMED_FD
] = "fd",
4764 [EXEC_INPUT_DATA
] = "data",
4765 [EXEC_INPUT_FILE
] = "file",
4768 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4770 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4771 [EXEC_OUTPUT_INHERIT
] = "inherit",
4772 [EXEC_OUTPUT_NULL
] = "null",
4773 [EXEC_OUTPUT_TTY
] = "tty",
4774 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4775 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4776 [EXEC_OUTPUT_KMSG
] = "kmsg",
4777 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4778 [EXEC_OUTPUT_JOURNAL
] = "journal",
4779 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4780 [EXEC_OUTPUT_SOCKET
] = "socket",
4781 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4782 [EXEC_OUTPUT_FILE
] = "file",
4785 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4787 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4788 [EXEC_UTMP_INIT
] = "init",
4789 [EXEC_UTMP_LOGIN
] = "login",
4790 [EXEC_UTMP_USER
] = "user",
4793 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
4795 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
4796 [EXEC_PRESERVE_NO
] = "no",
4797 [EXEC_PRESERVE_YES
] = "yes",
4798 [EXEC_PRESERVE_RESTART
] = "restart",
4801 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
4803 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
4804 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
4805 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
4806 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
4807 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
4808 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
4811 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
4813 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
4814 [EXEC_KEYRING_INHERIT
] = "inherit",
4815 [EXEC_KEYRING_PRIVATE
] = "private",
4816 [EXEC_KEYRING_SHARED
] = "shared",
4819 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);