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"
88 #include "namespace.h"
89 #include "parse-util.h"
90 #include "path-util.h"
91 #include "process-util.h"
92 #include "rlimit-util.h"
95 #include "seccomp-util.h"
97 #include "securebits.h"
98 #include "securebits-util.h"
99 #include "selinux-util.h"
100 #include "signal-util.h"
101 #include "smack-util.h"
103 #include "stat-util.h"
104 #include "string-table.h"
105 #include "string-util.h"
107 #include "syslog-util.h"
108 #include "terminal-util.h"
110 #include "user-util.h"
112 #include "utmp-wtmp.h"
114 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
115 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
117 /* This assumes there is a 'tty' group */
118 #define TTY_MODE 0620
120 #define SNDBUF_SIZE (8*1024*1024)
122 static int shift_fds(int fds
[], unsigned n_fds
) {
123 int start
, restart_from
;
128 /* Modifies the fds array! (sorts it) */
138 for (i
= start
; i
< (int) n_fds
; i
++) {
141 /* Already at right index? */
145 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
152 /* Hmm, the fd we wanted isn't free? Then
153 * let's remember that and try again from here */
154 if (nfd
!= i
+3 && restart_from
< 0)
158 if (restart_from
< 0)
161 start
= restart_from
;
167 static int flags_fds(const int fds
[], unsigned n_storage_fds
, unsigned n_socket_fds
, bool nonblock
) {
171 n_fds
= n_storage_fds
+ n_socket_fds
;
177 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
178 * O_NONBLOCK only applies to socket activation though. */
180 for (i
= 0; i
< n_fds
; i
++) {
182 if (i
< n_socket_fds
) {
183 r
= fd_nonblock(fds
[i
], nonblock
);
188 /* We unconditionally drop FD_CLOEXEC from the fds,
189 * since after all we want to pass these fds to our
192 r
= fd_cloexec(fds
[i
], false);
200 static const char *exec_context_tty_path(const ExecContext
*context
) {
203 if (context
->stdio_as_fds
)
206 if (context
->tty_path
)
207 return context
->tty_path
;
209 return "/dev/console";
212 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
217 path
= exec_context_tty_path(context
);
219 if (context
->tty_vhangup
) {
220 if (p
&& p
->stdin_fd
>= 0)
221 (void) terminal_vhangup_fd(p
->stdin_fd
);
223 (void) terminal_vhangup(path
);
226 if (context
->tty_reset
) {
227 if (p
&& p
->stdin_fd
>= 0)
228 (void) reset_terminal_fd(p
->stdin_fd
, true);
230 (void) reset_terminal(path
);
233 if (context
->tty_vt_disallocate
&& path
)
234 (void) vt_disallocate(path
);
237 static bool is_terminal_input(ExecInput i
) {
240 EXEC_INPUT_TTY_FORCE
,
241 EXEC_INPUT_TTY_FAIL
);
244 static bool is_terminal_output(ExecOutput o
) {
247 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
248 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
249 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
252 static bool is_syslog_output(ExecOutput o
) {
255 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
258 static bool is_kmsg_output(ExecOutput o
) {
261 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
264 static bool exec_context_needs_term(const ExecContext
*c
) {
267 /* Return true if the execution context suggests we should set $TERM to something useful. */
269 if (is_terminal_input(c
->std_input
))
272 if (is_terminal_output(c
->std_output
))
275 if (is_terminal_output(c
->std_error
))
278 return !!c
->tty_path
;
281 static int open_null_as(int flags
, int nfd
) {
286 fd
= open("/dev/null", flags
|O_NOCTTY
);
290 return move_fd(fd
, nfd
, false);
293 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
294 static const union sockaddr_union sa
= {
295 .un
.sun_family
= AF_UNIX
,
296 .un
.sun_path
= "/run/systemd/journal/stdout",
298 uid_t olduid
= UID_INVALID
;
299 gid_t oldgid
= GID_INVALID
;
302 if (gid_is_valid(gid
)) {
305 if (setegid(gid
) < 0)
309 if (uid_is_valid(uid
)) {
312 if (seteuid(uid
) < 0) {
318 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
320 /* If we fail to restore the uid or gid, things will likely
321 fail later on. This should only happen if an LSM interferes. */
323 if (uid_is_valid(uid
))
324 (void) seteuid(olduid
);
327 if (gid_is_valid(gid
))
328 (void) setegid(oldgid
);
333 static int connect_logger_as(
335 const ExecContext
*context
,
336 const ExecParameters
*params
,
347 assert(output
< _EXEC_OUTPUT_MAX
);
351 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
355 r
= connect_journal_socket(fd
, uid
, gid
);
359 if (shutdown(fd
, SHUT_RD
) < 0) {
364 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
374 context
->syslog_identifier
?: ident
,
375 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
376 context
->syslog_priority
,
377 !!context
->syslog_level_prefix
,
378 is_syslog_output(output
),
379 is_kmsg_output(output
),
380 is_terminal_output(output
));
382 return move_fd(fd
, nfd
, false);
384 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
390 fd
= open_terminal(path
, flags
| O_NOCTTY
);
394 return move_fd(fd
, nfd
, false);
397 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
398 union sockaddr_union sa
= {
399 .sa
.sa_family
= AF_UNIX
,
405 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
408 fd
= open(path
, flags
|O_NOCTTY
, mode
);
412 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
414 if (strlen(path
) > sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
417 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
419 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
423 strncpy(sa
.un
.sun_path
, path
, sizeof(sa
.un
.sun_path
));
424 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
426 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
427 * indication that his wasn't an AF_UNIX socket after all */
430 if ((flags
& O_ACCMODE
) == O_RDONLY
)
431 r
= shutdown(fd
, SHUT_WR
);
432 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
433 r
= shutdown(fd
, SHUT_RD
);
444 static int fixup_input(
445 const ExecContext
*context
,
447 bool apply_tty_stdin
) {
453 std_input
= context
->std_input
;
455 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
456 return EXEC_INPUT_NULL
;
458 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
459 return EXEC_INPUT_NULL
;
461 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
462 return EXEC_INPUT_NULL
;
467 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
469 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
470 return EXEC_OUTPUT_INHERIT
;
475 static int setup_input(
476 const ExecContext
*context
,
477 const ExecParameters
*params
,
479 int named_iofds
[3]) {
486 if (params
->stdin_fd
>= 0) {
487 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
490 /* Try to make this the controlling tty, if it is a tty, and reset it */
491 if (isatty(STDIN_FILENO
)) {
492 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
493 (void) reset_terminal_fd(STDIN_FILENO
, true);
499 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
503 case EXEC_INPUT_NULL
:
504 return open_null_as(O_RDONLY
, STDIN_FILENO
);
507 case EXEC_INPUT_TTY_FORCE
:
508 case EXEC_INPUT_TTY_FAIL
: {
511 fd
= acquire_terminal(exec_context_tty_path(context
),
512 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
513 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
514 ACQUIRE_TERMINAL_WAIT
,
519 return move_fd(fd
, STDIN_FILENO
, false);
522 case EXEC_INPUT_SOCKET
:
523 assert(socket_fd
>= 0);
525 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
527 case EXEC_INPUT_NAMED_FD
:
528 assert(named_iofds
[STDIN_FILENO
] >= 0);
530 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
531 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
533 case EXEC_INPUT_DATA
: {
536 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
540 return move_fd(fd
, STDIN_FILENO
, false);
543 case EXEC_INPUT_FILE
: {
547 assert(context
->stdio_file
[STDIN_FILENO
]);
549 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
550 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
552 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
556 return move_fd(fd
, STDIN_FILENO
, false);
560 assert_not_reached("Unknown input type");
564 static int setup_output(
566 const ExecContext
*context
,
567 const ExecParameters
*params
,
574 dev_t
*journal_stream_dev
,
575 ino_t
*journal_stream_ino
) {
585 assert(journal_stream_dev
);
586 assert(journal_stream_ino
);
588 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
590 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
593 return STDOUT_FILENO
;
596 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
597 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
600 return STDERR_FILENO
;
603 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
604 o
= fixup_output(context
->std_output
, socket_fd
);
606 if (fileno
== STDERR_FILENO
) {
608 e
= fixup_output(context
->std_error
, socket_fd
);
610 /* This expects the input and output are already set up */
612 /* Don't change the stderr file descriptor if we inherit all
613 * the way and are not on a tty */
614 if (e
== EXEC_OUTPUT_INHERIT
&&
615 o
== EXEC_OUTPUT_INHERIT
&&
616 i
== EXEC_INPUT_NULL
&&
617 !is_terminal_input(context
->std_input
) &&
621 /* Duplicate from stdout if possible */
622 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
623 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
627 } else if (o
== EXEC_OUTPUT_INHERIT
) {
628 /* If input got downgraded, inherit the original value */
629 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
630 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
632 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
633 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
634 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
636 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
640 /* We need to open /dev/null here anew, to get the right access mode. */
641 return open_null_as(O_WRONLY
, fileno
);
646 case EXEC_OUTPUT_NULL
:
647 return open_null_as(O_WRONLY
, fileno
);
649 case EXEC_OUTPUT_TTY
:
650 if (is_terminal_input(i
))
651 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
653 /* We don't reset the terminal if this is just about output */
654 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
656 case EXEC_OUTPUT_SYSLOG
:
657 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
658 case EXEC_OUTPUT_KMSG
:
659 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
660 case EXEC_OUTPUT_JOURNAL
:
661 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
662 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
664 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
665 r
= open_null_as(O_WRONLY
, fileno
);
669 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
670 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
671 * services to detect whether they are connected to the journal or not.
673 * If both stdout and stderr are connected to a stream then let's make sure to store the data
674 * about STDERR as that's usually the best way to do logging. */
676 if (fstat(fileno
, &st
) >= 0 &&
677 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
678 *journal_stream_dev
= st
.st_dev
;
679 *journal_stream_ino
= st
.st_ino
;
684 case EXEC_OUTPUT_SOCKET
:
685 assert(socket_fd
>= 0);
687 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
689 case EXEC_OUTPUT_NAMED_FD
:
690 assert(named_iofds
[fileno
] >= 0);
692 (void) fd_nonblock(named_iofds
[fileno
], false);
693 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
695 case EXEC_OUTPUT_FILE
: {
699 assert(context
->stdio_file
[fileno
]);
701 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
702 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
705 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
707 fd
= acquire_path(context
->stdio_file
[fileno
], O_WRONLY
, 0666 & ~context
->umask
);
711 return move_fd(fd
, fileno
, false);
715 assert_not_reached("Unknown error type");
719 static int chown_terminal(int fd
, uid_t uid
) {
724 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
728 /* This might fail. What matters are the results. */
729 (void) fchown(fd
, uid
, -1);
730 (void) fchmod(fd
, TTY_MODE
);
732 if (fstat(fd
, &st
) < 0)
735 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
741 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
742 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
745 assert(_saved_stdin
);
746 assert(_saved_stdout
);
748 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
752 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
753 if (saved_stdout
< 0)
756 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
760 r
= chown_terminal(fd
, getuid());
764 r
= reset_terminal_fd(fd
, true);
768 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
773 *_saved_stdin
= saved_stdin
;
774 *_saved_stdout
= saved_stdout
;
776 saved_stdin
= saved_stdout
= -1;
781 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
784 if (err
== -ETIMEDOUT
)
785 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
788 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
792 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
793 _cleanup_close_
int fd
= -1;
797 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
801 write_confirm_error_fd(err
, fd
, u
);
804 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
808 assert(saved_stdout
);
812 if (*saved_stdin
>= 0)
813 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
816 if (*saved_stdout
>= 0)
817 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
820 *saved_stdin
= safe_close(*saved_stdin
);
821 *saved_stdout
= safe_close(*saved_stdout
);
827 CONFIRM_PRETEND_FAILURE
= -1,
828 CONFIRM_PRETEND_SUCCESS
= 0,
832 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
833 int saved_stdout
= -1, saved_stdin
= -1, r
;
834 _cleanup_free_
char *e
= NULL
;
837 /* For any internal errors, assume a positive response. */
838 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
840 write_confirm_error(r
, vc
, u
);
841 return CONFIRM_EXECUTE
;
844 /* confirm_spawn might have been disabled while we were sleeping. */
845 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
850 e
= ellipsize(cmdline
, 60, 100);
858 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
860 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
867 printf("Resuming normal execution.\n");
868 manager_disable_confirm_spawn();
872 unit_dump(u
, stdout
, " ");
873 continue; /* ask again */
875 printf("Failing execution.\n");
876 r
= CONFIRM_PRETEND_FAILURE
;
879 printf(" c - continue, proceed without asking anymore\n"
880 " D - dump, show the state of the unit\n"
881 " f - fail, don't execute the command and pretend it failed\n"
883 " i - info, show a short summary of the unit\n"
884 " j - jobs, show jobs that are in progress\n"
885 " s - skip, don't execute the command and pretend it succeeded\n"
886 " y - yes, execute the command\n");
887 continue; /* ask again */
889 printf(" Description: %s\n"
892 u
->id
, u
->description
, cmdline
);
893 continue; /* ask again */
895 manager_dump_jobs(u
->manager
, stdout
, " ");
896 continue; /* ask again */
898 /* 'n' was removed in favor of 'f'. */
899 printf("Didn't understand 'n', did you mean 'f'?\n");
900 continue; /* ask again */
902 printf("Skipping execution.\n");
903 r
= CONFIRM_PRETEND_SUCCESS
;
909 assert_not_reached("Unhandled choice");
915 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
919 static int get_fixed_user(const ExecContext
*c
, const char **user
,
920 uid_t
*uid
, gid_t
*gid
,
921 const char **home
, const char **shell
) {
930 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
931 * (i.e. are "/" or "/bin/nologin"). */
934 r
= get_user_creds_clean(&name
, uid
, gid
, home
, shell
);
942 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
952 r
= get_group_creds(&name
, gid
);
960 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
961 const char *group
, gid_t gid
,
962 gid_t
**supplementary_gids
, int *ngids
) {
966 bool keep_groups
= false;
967 gid_t
*groups
= NULL
;
968 _cleanup_free_ gid_t
*l_gids
= NULL
;
973 * If user is given, then lookup GID and supplementary groups list.
974 * We avoid NSS lookups for gid=0. Also we have to initialize groups
975 * here and as early as possible so we keep the list of supplementary
976 * groups of the caller.
978 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
979 /* First step, initialize groups from /etc/groups */
980 if (initgroups(user
, gid
) < 0)
986 if (strv_isempty(c
->supplementary_groups
))
990 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
991 * be positive, otherwise fail.
994 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
995 if (ngroups_max
<= 0) {
999 return -EOPNOTSUPP
; /* For all other values */
1002 l_gids
= new(gid_t
, ngroups_max
);
1008 * Lookup the list of groups that the user belongs to, we
1009 * avoid NSS lookups here too for gid=0.
1012 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1017 STRV_FOREACH(i
, c
->supplementary_groups
) {
1020 if (k
>= ngroups_max
)
1024 r
= get_group_creds(&g
, l_gids
+k
);
1032 * Sets ngids to zero to drop all supplementary groups, happens
1033 * when we are under root and SupplementaryGroups= is empty.
1040 /* Otherwise get the final list of supplementary groups */
1041 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1045 *supplementary_gids
= groups
;
1053 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1056 /* Handle SupplementaryGroups= if it is not empty */
1058 r
= maybe_setgroups(ngids
, supplementary_gids
);
1063 if (gid_is_valid(gid
)) {
1064 /* Then set our gids */
1065 if (setresgid(gid
, gid
, gid
) < 0)
1072 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1075 if (!uid_is_valid(uid
))
1078 /* Sets (but doesn't look up) the uid and make sure we keep the
1079 * capabilities while doing so. */
1081 if (context
->capability_ambient_set
!= 0) {
1083 /* First step: If we need to keep capabilities but
1084 * drop privileges we need to make sure we keep our
1085 * caps, while we drop privileges. */
1087 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1089 if (prctl(PR_GET_SECUREBITS
) != sb
)
1090 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1095 /* Second step: actually set the uids */
1096 if (setresuid(uid
, uid
, uid
) < 0)
1099 /* At this point we should have all necessary capabilities but
1100 are otherwise a normal user. However, the caps might got
1101 corrupted due to the setresuid() so we need clean them up
1102 later. This is done outside of this call. */
1109 static int null_conv(
1111 const struct pam_message
**msg
,
1112 struct pam_response
**resp
,
1113 void *appdata_ptr
) {
1115 /* We don't support conversations */
1117 return PAM_CONV_ERR
;
1122 static int setup_pam(
1129 int fds
[], unsigned n_fds
) {
1133 static const struct pam_conv conv
= {
1138 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1139 pam_handle_t
*handle
= NULL
;
1141 int pam_code
= PAM_SUCCESS
, r
;
1142 char **nv
, **e
= NULL
;
1143 bool close_session
= false;
1144 pid_t pam_pid
= 0, parent_pid
;
1151 /* We set up PAM in the parent process, then fork. The child
1152 * will then stay around until killed via PR_GET_PDEATHSIG or
1153 * systemd via the cgroup logic. It will then remove the PAM
1154 * session again. The parent process will exec() the actual
1155 * daemon. We do things this way to ensure that the main PID
1156 * of the daemon is the one we initially fork()ed. */
1158 r
= barrier_create(&barrier
);
1162 if (log_get_max_level() < LOG_DEBUG
)
1163 flags
|= PAM_SILENT
;
1165 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1166 if (pam_code
!= PAM_SUCCESS
) {
1172 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1173 if (pam_code
!= PAM_SUCCESS
)
1177 STRV_FOREACH(nv
, *env
) {
1178 pam_code
= pam_putenv(handle
, *nv
);
1179 if (pam_code
!= PAM_SUCCESS
)
1183 pam_code
= pam_acct_mgmt(handle
, flags
);
1184 if (pam_code
!= PAM_SUCCESS
)
1187 pam_code
= pam_open_session(handle
, flags
);
1188 if (pam_code
!= PAM_SUCCESS
)
1191 close_session
= true;
1193 e
= pam_getenvlist(handle
);
1195 pam_code
= PAM_BUF_ERR
;
1199 /* Block SIGTERM, so that we know that it won't get lost in
1202 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1204 parent_pid
= getpid_cached();
1206 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1210 int sig
, ret
= EXIT_PAM
;
1212 /* The child's job is to reset the PAM session on
1214 barrier_set_role(&barrier
, BARRIER_CHILD
);
1216 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1217 * are open here that have been opened by PAM. */
1218 (void) close_many(fds
, n_fds
);
1220 /* Drop privileges - we don't need any to pam_close_session
1221 * and this will make PR_SET_PDEATHSIG work in most cases.
1222 * If this fails, ignore the error - but expect sd-pam threads
1223 * to fail to exit normally */
1225 r
= maybe_setgroups(0, NULL
);
1227 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1228 if (setresgid(gid
, gid
, gid
) < 0)
1229 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1230 if (setresuid(uid
, uid
, uid
) < 0)
1231 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1233 (void) ignore_signals(SIGPIPE
, -1);
1235 /* Wait until our parent died. This will only work if
1236 * the above setresuid() succeeds, otherwise the kernel
1237 * will not allow unprivileged parents kill their privileged
1238 * children this way. We rely on the control groups kill logic
1239 * to do the rest for us. */
1240 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1243 /* Tell the parent that our setup is done. This is especially
1244 * important regarding dropping privileges. Otherwise, unit
1245 * setup might race against our setresuid(2) call.
1247 * If the parent aborted, we'll detect this below, hence ignore
1248 * return failure here. */
1249 (void) barrier_place(&barrier
);
1251 /* Check if our parent process might already have died? */
1252 if (getppid() == parent_pid
) {
1255 assert_se(sigemptyset(&ss
) >= 0);
1256 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1259 if (sigwait(&ss
, &sig
) < 0) {
1266 assert(sig
== SIGTERM
);
1271 /* If our parent died we'll end the session */
1272 if (getppid() != parent_pid
) {
1273 pam_code
= pam_close_session(handle
, flags
);
1274 if (pam_code
!= PAM_SUCCESS
)
1281 pam_end(handle
, pam_code
| flags
);
1285 barrier_set_role(&barrier
, BARRIER_PARENT
);
1287 /* If the child was forked off successfully it will do all the
1288 * cleanups, so forget about the handle here. */
1291 /* Unblock SIGTERM again in the parent */
1292 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1294 /* We close the log explicitly here, since the PAM modules
1295 * might have opened it, but we don't want this fd around. */
1298 /* Synchronously wait for the child to initialize. We don't care for
1299 * errors as we cannot recover. However, warn loudly if it happens. */
1300 if (!barrier_place_and_sync(&barrier
))
1301 log_error("PAM initialization failed");
1309 if (pam_code
!= PAM_SUCCESS
) {
1310 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1311 r
= -EPERM
; /* PAM errors do not map to errno */
1313 log_error_errno(r
, "PAM failed: %m");
1317 pam_code
= pam_close_session(handle
, flags
);
1319 pam_end(handle
, pam_code
| flags
);
1331 static void rename_process_from_path(const char *path
) {
1332 char process_name
[11];
1336 /* This resulting string must fit in 10 chars (i.e. the length
1337 * of "/sbin/init") to look pretty in /bin/ps */
1341 rename_process("(...)");
1347 /* The end of the process name is usually more
1348 * interesting, since the first bit might just be
1354 process_name
[0] = '(';
1355 memcpy(process_name
+1, p
, l
);
1356 process_name
[1+l
] = ')';
1357 process_name
[1+l
+1] = 0;
1359 rename_process(process_name
);
1362 static bool context_has_address_families(const ExecContext
*c
) {
1365 return c
->address_families_whitelist
||
1366 !set_isempty(c
->address_families
);
1369 static bool context_has_syscall_filters(const ExecContext
*c
) {
1372 return c
->syscall_whitelist
||
1373 !hashmap_isempty(c
->syscall_filter
);
1376 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1379 if (c
->no_new_privileges
)
1382 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1385 /* We need NNP if we have any form of seccomp and are unprivileged */
1386 return context_has_address_families(c
) ||
1387 c
->memory_deny_write_execute
||
1388 c
->restrict_realtime
||
1389 exec_context_restrict_namespaces_set(c
) ||
1390 c
->protect_kernel_tunables
||
1391 c
->protect_kernel_modules
||
1392 c
->private_devices
||
1393 context_has_syscall_filters(c
) ||
1394 !set_isempty(c
->syscall_archs
) ||
1395 c
->lock_personality
;
1400 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1402 if (is_seccomp_available())
1405 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1409 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1410 uint32_t negative_action
, default_action
, action
;
1416 if (!context_has_syscall_filters(c
))
1419 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1422 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1424 if (c
->syscall_whitelist
) {
1425 default_action
= negative_action
;
1426 action
= SCMP_ACT_ALLOW
;
1428 default_action
= SCMP_ACT_ALLOW
;
1429 action
= negative_action
;
1432 if (needs_ambient_hack
) {
1433 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1438 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
);
1441 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1445 if (set_isempty(c
->syscall_archs
))
1448 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1451 return seccomp_restrict_archs(c
->syscall_archs
);
1454 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1458 if (!context_has_address_families(c
))
1461 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1464 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1467 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1471 if (!c
->memory_deny_write_execute
)
1474 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1477 return seccomp_memory_deny_write_execute();
1480 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1484 if (!c
->restrict_realtime
)
1487 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1490 return seccomp_restrict_realtime();
1493 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1497 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1498 * let's protect even those systems where this is left on in the kernel. */
1500 if (!c
->protect_kernel_tunables
)
1503 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1506 return seccomp_protect_sysctl();
1509 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1513 /* Turn off module syscalls on ProtectKernelModules=yes */
1515 if (!c
->protect_kernel_modules
)
1518 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1521 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
));
1524 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1528 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1530 if (!c
->private_devices
)
1533 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1536 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
));
1539 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1543 if (!exec_context_restrict_namespaces_set(c
))
1546 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1549 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1552 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1553 unsigned long personality
;
1559 if (!c
->lock_personality
)
1562 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1565 personality
= c
->personality
;
1567 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1568 if (personality
== PERSONALITY_INVALID
) {
1570 r
= opinionated_personality(&personality
);
1575 return seccomp_lock_personality(personality
);
1580 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1583 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1584 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1586 if (idle_pipe
[0] >= 0) {
1589 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1591 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1594 /* Signal systemd that we are bored and want to continue. */
1595 n
= write(idle_pipe
[3], "x", 1);
1597 /* Wait for systemd to react to the signal above. */
1598 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1601 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1605 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1608 static int build_environment(
1610 const ExecContext
*c
,
1611 const ExecParameters
*p
,
1614 const char *username
,
1616 dev_t journal_stream_dev
,
1617 ino_t journal_stream_ino
,
1620 _cleanup_strv_free_
char **our_env
= NULL
;
1628 our_env
= new0(char*, 14);
1633 _cleanup_free_
char *joined
= NULL
;
1635 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1637 our_env
[n_env
++] = x
;
1639 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1641 our_env
[n_env
++] = x
;
1643 joined
= strv_join(p
->fd_names
, ":");
1647 x
= strjoin("LISTEN_FDNAMES=", joined
);
1650 our_env
[n_env
++] = x
;
1653 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1654 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1656 our_env
[n_env
++] = x
;
1658 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1660 our_env
[n_env
++] = x
;
1663 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1664 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1665 * check the database directly. */
1666 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1667 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1670 our_env
[n_env
++] = x
;
1674 x
= strappend("HOME=", home
);
1677 our_env
[n_env
++] = x
;
1681 x
= strappend("LOGNAME=", username
);
1684 our_env
[n_env
++] = x
;
1686 x
= strappend("USER=", username
);
1689 our_env
[n_env
++] = x
;
1693 x
= strappend("SHELL=", shell
);
1696 our_env
[n_env
++] = x
;
1699 if (!sd_id128_is_null(u
->invocation_id
)) {
1700 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1703 our_env
[n_env
++] = x
;
1706 if (exec_context_needs_term(c
)) {
1707 const char *tty_path
, *term
= NULL
;
1709 tty_path
= exec_context_tty_path(c
);
1711 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1712 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1713 * passes to PID 1 ends up all the way in the console login shown. */
1715 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1716 term
= getenv("TERM");
1718 term
= default_term_for_tty(tty_path
);
1720 x
= strappend("TERM=", term
);
1723 our_env
[n_env
++] = x
;
1726 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1727 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1730 our_env
[n_env
++] = x
;
1733 our_env
[n_env
++] = NULL
;
1734 assert(n_env
<= 12);
1742 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1743 _cleanup_strv_free_
char **pass_env
= NULL
;
1744 size_t n_env
= 0, n_bufsize
= 0;
1747 STRV_FOREACH(i
, c
->pass_environment
) {
1748 _cleanup_free_
char *x
= NULL
;
1754 x
= strjoin(*i
, "=", v
);
1758 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1761 pass_env
[n_env
++] = x
;
1762 pass_env
[n_env
] = NULL
;
1772 static bool exec_needs_mount_namespace(
1773 const ExecContext
*context
,
1774 const ExecParameters
*params
,
1775 const ExecRuntime
*runtime
) {
1780 if (context
->root_image
)
1783 if (!strv_isempty(context
->read_write_paths
) ||
1784 !strv_isempty(context
->read_only_paths
) ||
1785 !strv_isempty(context
->inaccessible_paths
))
1788 if (context
->n_bind_mounts
> 0)
1791 if (context
->n_temporary_filesystems
> 0)
1794 if (context
->mount_flags
!= 0)
1797 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1800 if (context
->private_devices
||
1801 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1802 context
->protect_home
!= PROTECT_HOME_NO
||
1803 context
->protect_kernel_tunables
||
1804 context
->protect_kernel_modules
||
1805 context
->protect_control_groups
)
1808 if (context
->mount_apivfs
&& (context
->root_image
|| context
->root_directory
))
1811 if (context
->dynamic_user
&&
1812 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1813 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1814 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1820 static int setup_private_users(uid_t uid
, gid_t gid
) {
1821 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1822 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1823 _cleanup_close_
int unshare_ready_fd
= -1;
1824 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1829 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1830 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1831 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1832 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1833 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1834 * continues execution normally. */
1836 if (uid
!= 0 && uid_is_valid(uid
)) {
1837 r
= asprintf(&uid_map
,
1838 "0 0 1\n" /* Map root → root */
1839 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1844 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1849 if (gid
!= 0 && gid_is_valid(gid
)) {
1850 r
= asprintf(&gid_map
,
1851 "0 0 1\n" /* Map root → root */
1852 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1857 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1862 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1864 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1865 if (unshare_ready_fd
< 0)
1868 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1870 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1873 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1877 _cleanup_close_
int fd
= -1;
1881 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1882 * here, after the parent opened its own user namespace. */
1885 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1887 /* Wait until the parent unshared the user namespace */
1888 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1893 /* Disable the setgroups() system call in the child user namespace, for good. */
1894 a
= procfs_file_alloca(ppid
, "setgroups");
1895 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1897 if (errno
!= ENOENT
) {
1902 /* If the file is missing the kernel is too old, let's continue anyway. */
1904 if (write(fd
, "deny\n", 5) < 0) {
1909 fd
= safe_close(fd
);
1912 /* First write the GID map */
1913 a
= procfs_file_alloca(ppid
, "gid_map");
1914 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1919 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1923 fd
= safe_close(fd
);
1925 /* The write the UID map */
1926 a
= procfs_file_alloca(ppid
, "uid_map");
1927 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1932 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1937 _exit(EXIT_SUCCESS
);
1940 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1941 _exit(EXIT_FAILURE
);
1944 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1946 if (unshare(CLONE_NEWUSER
) < 0)
1949 /* Let the child know that the namespace is ready now */
1950 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1953 /* Try to read an error code from the child */
1954 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1957 if (n
== sizeof(r
)) { /* an error code was sent to us */
1962 if (n
!= 0) /* on success we should have read 0 bytes */
1965 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
1969 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
1975 static int setup_exec_directory(
1976 const ExecContext
*context
,
1977 const ExecParameters
*params
,
1980 ExecDirectoryType type
,
1983 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
1984 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
1985 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
1986 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
1987 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
1988 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
1995 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
1996 assert(exit_status
);
1998 if (!params
->prefix
[type
])
2001 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2002 if (!uid_is_valid(uid
))
2004 if (!gid_is_valid(gid
))
2008 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2009 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2011 p
= strjoin(params
->prefix
[type
], "/", *rt
);
2017 r
= mkdir_parents_label(p
, 0755);
2021 if (context
->dynamic_user
&&
2022 !IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2023 _cleanup_free_
char *private_root
= NULL
, *relative
= NULL
, *parent
= NULL
;
2025 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2026 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2027 * whose UID is later on reused. To lock this down we use the same trick used by container
2028 * managers to prohibit host users to get access to files of the same UID in containers: we
2029 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2030 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2031 * to make this directory permeable for the service itself.
2033 * Specifically: for a service which wants a special directory "foo/" we first create a
2034 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2035 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2036 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2037 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2038 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2039 * disabling the access boundary for the service and making sure it only gets access to the
2040 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2042 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2043 * owned by the service itself.
2044 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2045 * files or sockets with other services. */
2047 private_root
= strjoin(params
->prefix
[type
], "/private");
2048 if (!private_root
) {
2053 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2054 r
= mkdir_safe_label(private_root
, 0700, 0, 0, 0);
2058 pp
= strjoin(private_root
, "/", *rt
);
2064 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2065 r
= mkdir_parents_label(pp
, 0755);
2069 if (is_dir(p
, false) > 0 &&
2070 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2072 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2073 * it over. Most likely the service has been upgraded from one that didn't use
2074 * DynamicUser=1, to one that does. */
2076 if (rename(p
, pp
) < 0) {
2081 /* Otherwise, create the actual directory for the service */
2083 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2084 if (r
< 0 && r
!= -EEXIST
)
2088 parent
= dirname_malloc(p
);
2094 r
= path_make_relative(parent
, pp
, &relative
);
2098 /* And link it up from the original place */
2099 r
= symlink_idempotent(relative
, p
);
2103 /* Lock down the access mode */
2104 if (chmod(pp
, context
->directories
[type
].mode
) < 0) {
2109 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2116 /* Don't change the owner of the configuration directory, as in the common case it is not written to by
2117 * a service, and shall not be writable. */
2118 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2121 /* Then, change the ownership of the whole tree, if necessary */
2122 r
= path_chown_recursive(pp
?: p
, uid
, gid
);
2130 *exit_status
= exit_status_table
[type
];
2135 static int setup_smack(
2136 const ExecContext
*context
,
2137 const ExecCommand
*command
) {
2144 if (context
->smack_process_label
) {
2145 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2149 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2151 _cleanup_free_
char *exec_label
= NULL
;
2153 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2154 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2157 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2167 static int compile_bind_mounts(
2168 const ExecContext
*context
,
2169 const ExecParameters
*params
,
2170 BindMount
**ret_bind_mounts
,
2171 unsigned *ret_n_bind_mounts
,
2172 char ***ret_empty_directories
) {
2174 _cleanup_strv_free_
char **empty_directories
= NULL
;
2175 BindMount
*bind_mounts
;
2176 unsigned n
, h
= 0, i
;
2177 ExecDirectoryType t
;
2182 assert(ret_bind_mounts
);
2183 assert(ret_n_bind_mounts
);
2184 assert(ret_empty_directories
);
2186 n
= context
->n_bind_mounts
;
2187 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2188 if (!params
->prefix
[t
])
2191 n
+= strv_length(context
->directories
[t
].paths
);
2195 *ret_bind_mounts
= NULL
;
2196 *ret_n_bind_mounts
= 0;
2197 *ret_empty_directories
= NULL
;
2201 bind_mounts
= new(BindMount
, n
);
2205 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2206 BindMount
*item
= context
->bind_mounts
+ i
;
2209 s
= strdup(item
->source
);
2215 d
= strdup(item
->destination
);
2222 bind_mounts
[h
++] = (BindMount
) {
2225 .read_only
= item
->read_only
,
2226 .recursive
= item
->recursive
,
2227 .ignore_enoent
= item
->ignore_enoent
,
2231 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2234 if (!params
->prefix
[t
])
2237 if (strv_isempty(context
->directories
[t
].paths
))
2240 if (context
->dynamic_user
&&
2241 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2244 /* So this is for a dynamic user, and we need to make sure the process can access its own
2245 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2246 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2248 private_root
= strjoin(params
->prefix
[t
], "/private");
2249 if (!private_root
) {
2254 r
= strv_consume(&empty_directories
, private_root
);
2259 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2262 if (context
->dynamic_user
&&
2263 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2264 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2266 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2279 bind_mounts
[h
++] = (BindMount
) {
2284 .ignore_enoent
= false,
2291 *ret_bind_mounts
= bind_mounts
;
2292 *ret_n_bind_mounts
= n
;
2293 *ret_empty_directories
= empty_directories
;
2295 empty_directories
= NULL
;
2300 bind_mount_free_many(bind_mounts
, h
);
2304 static int apply_mount_namespace(
2306 const ExecCommand
*command
,
2307 const ExecContext
*context
,
2308 const ExecParameters
*params
,
2309 const ExecRuntime
*runtime
) {
2311 _cleanup_strv_free_
char **empty_directories
= NULL
;
2312 char *tmp
= NULL
, *var
= NULL
;
2313 const char *root_dir
= NULL
, *root_image
= NULL
;
2314 NamespaceInfo ns_info
= {
2315 .ignore_protect_paths
= false,
2316 .private_dev
= context
->private_devices
,
2317 .protect_control_groups
= context
->protect_control_groups
,
2318 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2319 .protect_kernel_modules
= context
->protect_kernel_modules
,
2320 .mount_apivfs
= context
->mount_apivfs
,
2322 bool needs_sandboxing
;
2323 BindMount
*bind_mounts
= NULL
;
2324 unsigned n_bind_mounts
= 0;
2329 /* The runtime struct only contains the parent of the private /tmp,
2330 * which is non-accessible to world users. Inside of it there's a /tmp
2331 * that is sticky, and that's the one we want to use here. */
2333 if (context
->private_tmp
&& runtime
) {
2334 if (runtime
->tmp_dir
)
2335 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2336 if (runtime
->var_tmp_dir
)
2337 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2340 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2341 root_image
= context
->root_image
;
2344 root_dir
= context
->root_directory
;
2347 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2352 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2353 * sandbox info, otherwise enforce it, don't ignore protected paths and
2354 * fail if we are enable to apply the sandbox inside the mount namespace.
2356 if (!context
->dynamic_user
&& root_dir
)
2357 ns_info
.ignore_protect_paths
= true;
2359 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2361 r
= setup_namespace(root_dir
, root_image
,
2362 &ns_info
, context
->read_write_paths
,
2363 needs_sandboxing
? context
->read_only_paths
: NULL
,
2364 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2368 context
->temporary_filesystems
,
2369 context
->n_temporary_filesystems
,
2372 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2373 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2374 context
->mount_flags
,
2375 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2377 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2379 /* If we couldn't set up the namespace this is probably due to a
2380 * missing capability. In this case, silently proceeed. */
2381 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2382 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2389 static int apply_working_directory(
2390 const ExecContext
*context
,
2391 const ExecParameters
*params
,
2393 const bool needs_mount_ns
,
2399 assert(exit_status
);
2401 if (context
->working_directory_home
) {
2404 *exit_status
= EXIT_CHDIR
;
2410 } else if (context
->working_directory
)
2411 wd
= context
->working_directory
;
2415 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2416 if (!needs_mount_ns
&& context
->root_directory
)
2417 if (chroot(context
->root_directory
) < 0) {
2418 *exit_status
= EXIT_CHROOT
;
2424 d
= prefix_roota(context
->root_directory
, wd
);
2426 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2427 *exit_status
= EXIT_CHDIR
;
2434 static int setup_keyring(
2436 const ExecContext
*context
,
2437 const ExecParameters
*p
,
2438 uid_t uid
, gid_t gid
) {
2440 key_serial_t keyring
;
2447 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2448 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2449 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2450 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2451 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2452 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2454 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2457 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2460 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2461 if (keyring
== -1) {
2462 if (errno
== ENOSYS
)
2463 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2464 else if (IN_SET(errno
, EACCES
, EPERM
))
2465 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2466 else if (errno
== EDQUOT
)
2467 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2469 return log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2474 /* Populate they keyring with the invocation ID by default. */
2475 if (!sd_id128_is_null(u
->invocation_id
)) {
2478 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2480 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2482 if (keyctl(KEYCTL_SETPERM
, key
,
2483 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2484 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2485 return log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2489 /* And now, make the keyring owned by the service's user */
2490 if (uid_is_valid(uid
) || gid_is_valid(gid
))
2491 if (keyctl(KEYCTL_CHOWN
, keyring
, uid
, gid
, 0) < 0)
2492 return log_unit_error_errno(u
, errno
, "Failed to change ownership of session keyring: %m");
2494 /* When requested link the user keyring into the session keyring. */
2495 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2499 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things
2500 * set up properly by the kernel. If we don't do that then we can't create it atomically, and that
2501 * sucks for parallel execution. This mimics what pam_keyinit does, too.*/
2503 saved_uid
= getuid();
2504 saved_gid
= getgid();
2506 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2507 if (setregid(gid
, -1) < 0)
2508 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2511 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2512 if (setreuid(uid
, -1) < 0) {
2513 (void) setregid(saved_gid
, -1);
2514 return log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2518 if (keyctl(KEYCTL_LINK
,
2519 KEY_SPEC_USER_KEYRING
,
2520 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2524 (void) setreuid(saved_uid
, -1);
2525 (void) setregid(saved_gid
, -1);
2527 return log_unit_error_errno(u
, r
, "Failed to link user keyring into session keyring: %m");
2530 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2531 if (setreuid(saved_uid
, -1) < 0) {
2532 (void) setregid(saved_gid
, -1);
2533 return log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2537 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2538 if (setregid(saved_gid
, -1) < 0)
2539 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2546 static void append_socket_pair(int *array
, unsigned *n
, const int pair
[2]) {
2554 array
[(*n
)++] = pair
[0];
2556 array
[(*n
)++] = pair
[1];
2559 static int close_remaining_fds(
2560 const ExecParameters
*params
,
2561 const ExecRuntime
*runtime
,
2562 const DynamicCreds
*dcreds
,
2565 int *fds
, unsigned n_fds
) {
2567 unsigned n_dont_close
= 0;
2568 int dont_close
[n_fds
+ 12];
2572 if (params
->stdin_fd
>= 0)
2573 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2574 if (params
->stdout_fd
>= 0)
2575 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2576 if (params
->stderr_fd
>= 0)
2577 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2580 dont_close
[n_dont_close
++] = socket_fd
;
2582 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2583 n_dont_close
+= n_fds
;
2587 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2591 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2593 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2596 if (user_lookup_fd
>= 0)
2597 dont_close
[n_dont_close
++] = user_lookup_fd
;
2599 return close_all_fds(dont_close
, n_dont_close
);
2602 static int send_user_lookup(
2610 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2611 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2614 if (user_lookup_fd
< 0)
2617 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2620 if (writev(user_lookup_fd
,
2622 IOVEC_INIT(&uid
, sizeof(uid
)),
2623 IOVEC_INIT(&gid
, sizeof(gid
)),
2624 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2630 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2637 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2642 if (!c
->working_directory_home
)
2646 /* Hardcode /root as home directory for UID 0 */
2651 r
= get_home_dir(buf
);
2659 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2660 _cleanup_strv_free_
char ** list
= NULL
;
2661 ExecDirectoryType t
;
2668 assert(c
->dynamic_user
);
2670 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2671 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2674 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2677 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2683 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2686 if (t
== EXEC_DIRECTORY_RUNTIME
)
2687 e
= strjoin(p
->prefix
[t
], "/", *i
);
2689 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2693 r
= strv_consume(&list
, e
);
2705 static char *exec_command_line(char **argv
);
2707 static int exec_child(
2709 const ExecCommand
*command
,
2710 const ExecContext
*context
,
2711 const ExecParameters
*params
,
2712 ExecRuntime
*runtime
,
2713 DynamicCreds
*dcreds
,
2718 unsigned n_storage_fds
,
2719 unsigned n_socket_fds
,
2724 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2725 _cleanup_free_
char *home_buffer
= NULL
;
2726 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2727 const char *username
= NULL
, *groupname
= NULL
;
2728 const char *home
= NULL
, *shell
= NULL
;
2729 dev_t journal_stream_dev
= 0;
2730 ino_t journal_stream_ino
= 0;
2731 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2732 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2733 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2734 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2736 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2737 bool use_selinux
= false;
2740 bool use_smack
= false;
2743 bool use_apparmor
= false;
2745 uid_t uid
= UID_INVALID
;
2746 gid_t gid
= GID_INVALID
;
2747 int i
, r
, ngids
= 0;
2749 ExecDirectoryType dt
;
2756 assert(exit_status
);
2758 rename_process_from_path(command
->path
);
2760 /* We reset exactly these signals, since they are the
2761 * only ones we set to SIG_IGN in the main daemon. All
2762 * others we leave untouched because we set them to
2763 * SIG_DFL or a valid handler initially, both of which
2764 * will be demoted to SIG_DFL. */
2765 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2766 SIGNALS_IGNORE
, -1);
2768 if (context
->ignore_sigpipe
)
2769 (void) ignore_signals(SIGPIPE
, -1);
2771 r
= reset_signal_mask();
2773 *exit_status
= EXIT_SIGNAL_MASK
;
2774 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2777 if (params
->idle_pipe
)
2778 do_idle_pipe_dance(params
->idle_pipe
);
2780 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2781 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2782 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2783 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2786 log_set_open_when_needed(true);
2788 /* In case anything used libc syslog(), close this here, too */
2791 n_fds
= n_storage_fds
+ n_socket_fds
;
2792 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2794 *exit_status
= EXIT_FDS
;
2795 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2798 if (!context
->same_pgrp
)
2800 *exit_status
= EXIT_SETSID
;
2801 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2804 exec_context_tty_reset(context
, params
);
2806 if (unit_shall_confirm_spawn(unit
)) {
2807 const char *vc
= params
->confirm_spawn
;
2808 _cleanup_free_
char *cmdline
= NULL
;
2810 cmdline
= exec_command_line(argv
);
2812 *exit_status
= EXIT_MEMORY
;
2816 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2817 if (r
!= CONFIRM_EXECUTE
) {
2818 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2819 *exit_status
= EXIT_SUCCESS
;
2822 *exit_status
= EXIT_CONFIRM
;
2823 log_unit_error(unit
, "Execution cancelled by the user");
2828 if (context
->dynamic_user
&& dcreds
) {
2829 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2831 /* Make sure we bypass our own NSS module for any NSS checks */
2832 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2833 *exit_status
= EXIT_USER
;
2834 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2837 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2839 *exit_status
= EXIT_MEMORY
;
2843 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2845 *exit_status
= EXIT_USER
;
2847 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2850 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2853 if (!uid_is_valid(uid
)) {
2854 *exit_status
= EXIT_USER
;
2855 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
2859 if (!gid_is_valid(gid
)) {
2860 *exit_status
= EXIT_USER
;
2861 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
2866 username
= dcreds
->user
->name
;
2869 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2871 *exit_status
= EXIT_USER
;
2872 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
2875 r
= get_fixed_group(context
, &groupname
, &gid
);
2877 *exit_status
= EXIT_GROUP
;
2878 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
2882 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2883 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2884 &supplementary_gids
, &ngids
);
2886 *exit_status
= EXIT_GROUP
;
2887 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
2890 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2892 *exit_status
= EXIT_USER
;
2893 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
2896 user_lookup_fd
= safe_close(user_lookup_fd
);
2898 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2900 *exit_status
= EXIT_CHDIR
;
2901 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
2904 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2905 * must sure to drop O_NONBLOCK */
2907 (void) fd_nonblock(socket_fd
, false);
2909 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2911 *exit_status
= EXIT_STDIN
;
2912 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
2915 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2917 *exit_status
= EXIT_STDOUT
;
2918 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
2921 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2923 *exit_status
= EXIT_STDERR
;
2924 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
2927 if (params
->cgroup_path
) {
2928 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2930 *exit_status
= EXIT_CGROUP
;
2931 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", params
->cgroup_path
);
2935 if (context
->oom_score_adjust_set
) {
2936 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2938 /* When we can't make this change due to EPERM, then
2939 * let's silently skip over it. User namespaces
2940 * prohibit write access to this file, and we
2941 * shouldn't trip up over that. */
2943 sprintf(t
, "%i", context
->oom_score_adjust
);
2944 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2945 if (IN_SET(r
, -EPERM
, -EACCES
))
2946 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2948 *exit_status
= EXIT_OOM_ADJUST
;
2949 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
2953 if (context
->nice_set
)
2954 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2955 *exit_status
= EXIT_NICE
;
2956 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
2959 if (context
->cpu_sched_set
) {
2960 struct sched_param param
= {
2961 .sched_priority
= context
->cpu_sched_priority
,
2964 r
= sched_setscheduler(0,
2965 context
->cpu_sched_policy
|
2966 (context
->cpu_sched_reset_on_fork
?
2967 SCHED_RESET_ON_FORK
: 0),
2970 *exit_status
= EXIT_SETSCHEDULER
;
2971 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
2975 if (context
->cpuset
)
2976 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2977 *exit_status
= EXIT_CPUAFFINITY
;
2978 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
2981 if (context
->ioprio_set
)
2982 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2983 *exit_status
= EXIT_IOPRIO
;
2984 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
2987 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2988 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2989 *exit_status
= EXIT_TIMERSLACK
;
2990 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
2993 if (context
->personality
!= PERSONALITY_INVALID
) {
2994 r
= safe_personality(context
->personality
);
2996 *exit_status
= EXIT_PERSONALITY
;
2997 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3001 if (context
->utmp_id
)
3002 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3004 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3005 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3009 if (context
->user
) {
3010 r
= chown_terminal(STDIN_FILENO
, uid
);
3012 *exit_status
= EXIT_STDIN
;
3013 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3017 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroupsv1
3018 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3019 * safe. On cgroupsv2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3020 * touch a single hierarchy too. */
3021 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3022 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3024 *exit_status
= EXIT_CGROUP
;
3025 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3029 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3030 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3032 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3035 r
= build_environment(
3047 *exit_status
= EXIT_MEMORY
;
3051 r
= build_pass_environment(context
, &pass_env
);
3053 *exit_status
= EXIT_MEMORY
;
3057 accum_env
= strv_env_merge(5,
3058 params
->environment
,
3061 context
->environment
,
3065 *exit_status
= EXIT_MEMORY
;
3068 accum_env
= strv_env_clean(accum_env
);
3070 (void) umask(context
->umask
);
3072 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3074 *exit_status
= EXIT_KEYRING
;
3075 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3078 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3079 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3081 /* 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 */
3082 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3084 /* 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 */
3085 if (needs_ambient_hack
)
3086 needs_setuid
= false;
3088 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3090 if (needs_sandboxing
) {
3091 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3092 * present. The actual MAC context application will happen later, as late as possible, to avoid
3093 * impacting our own code paths. */
3096 use_selinux
= mac_selinux_use();
3099 use_smack
= mac_smack_use();
3102 use_apparmor
= mac_apparmor_use();
3107 if (context
->pam_name
&& username
) {
3108 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3110 *exit_status
= EXIT_PAM
;
3111 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3116 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3117 if (ns_type_supported(NAMESPACE_NET
)) {
3118 r
= setup_netns(runtime
->netns_storage_socket
);
3120 *exit_status
= EXIT_NETWORK
;
3121 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3124 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3127 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3128 if (needs_mount_namespace
) {
3129 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3131 *exit_status
= EXIT_NAMESPACE
;
3132 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3136 /* Apply just after mount namespace setup */
3137 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3139 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3141 /* Drop groups as early as possbile */
3143 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3145 *exit_status
= EXIT_GROUP
;
3146 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3150 if (needs_sandboxing
) {
3152 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3153 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3155 *exit_status
= EXIT_SELINUX_CONTEXT
;
3156 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3161 if (context
->private_users
) {
3162 r
= setup_private_users(uid
, gid
);
3164 *exit_status
= EXIT_USER
;
3165 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3170 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3171 * more aggressive this time since socket_fd and the netns fds we don't need anymore. The custom endpoint fd
3172 * was needed to upload the policy and can now be closed as well. */
3173 r
= close_all_fds(fds
, n_fds
);
3175 r
= shift_fds(fds
, n_fds
);
3177 r
= flags_fds(fds
, n_storage_fds
, n_socket_fds
, context
->non_blocking
);
3179 *exit_status
= EXIT_FDS
;
3180 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3183 secure_bits
= context
->secure_bits
;
3185 if (needs_sandboxing
) {
3188 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3190 if (!context
->rlimit
[i
])
3193 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
3195 *exit_status
= EXIT_LIMITS
;
3196 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit %s: %m", rlimit_to_string(i
));
3200 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
3201 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3202 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3203 *exit_status
= EXIT_LIMITS
;
3204 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3209 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3210 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3212 r
= setup_smack(context
, command
);
3214 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3215 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3220 bset
= context
->capability_bounding_set
;
3221 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3222 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3223 * instead of us doing that */
3224 if (needs_ambient_hack
)
3225 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3226 (UINT64_C(1) << CAP_SETUID
) |
3227 (UINT64_C(1) << CAP_SETGID
);
3229 if (!cap_test_all(bset
)) {
3230 r
= capability_bounding_set_drop(bset
, false);
3232 *exit_status
= EXIT_CAPABILITIES
;
3233 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3237 /* This is done before enforce_user, but ambient set
3238 * does not survive over setresuid() if keep_caps is not set. */
3239 if (!needs_ambient_hack
&&
3240 context
->capability_ambient_set
!= 0) {
3241 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3243 *exit_status
= EXIT_CAPABILITIES
;
3244 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3250 if (context
->user
) {
3251 r
= enforce_user(context
, uid
);
3253 *exit_status
= EXIT_USER
;
3254 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3257 if (!needs_ambient_hack
&&
3258 context
->capability_ambient_set
!= 0) {
3260 /* Fix the ambient capabilities after user change. */
3261 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3263 *exit_status
= EXIT_CAPABILITIES
;
3264 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3267 /* If we were asked to change user and ambient capabilities
3268 * were requested, we had to add keep-caps to the securebits
3269 * so that we would maintain the inherited capability set
3270 * through the setresuid(). Make sure that the bit is added
3271 * also to the context secure_bits so that we don't try to
3272 * drop the bit away next. */
3274 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3279 if (needs_sandboxing
) {
3280 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3281 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3282 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3283 * are restricted. */
3287 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3290 r
= setexeccon(exec_context
);
3292 *exit_status
= EXIT_SELINUX_CONTEXT
;
3293 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3300 if (use_apparmor
&& context
->apparmor_profile
) {
3301 r
= aa_change_onexec(context
->apparmor_profile
);
3302 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3303 *exit_status
= EXIT_APPARMOR_PROFILE
;
3304 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3309 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3310 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3311 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3312 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3313 *exit_status
= EXIT_SECUREBITS
;
3314 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3317 if (context_has_no_new_privileges(context
))
3318 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3319 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3320 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3324 r
= apply_address_families(unit
, context
);
3326 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3327 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3330 r
= apply_memory_deny_write_execute(unit
, context
);
3332 *exit_status
= EXIT_SECCOMP
;
3333 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3336 r
= apply_restrict_realtime(unit
, context
);
3338 *exit_status
= EXIT_SECCOMP
;
3339 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3342 r
= apply_restrict_namespaces(unit
, context
);
3344 *exit_status
= EXIT_SECCOMP
;
3345 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3348 r
= apply_protect_sysctl(unit
, context
);
3350 *exit_status
= EXIT_SECCOMP
;
3351 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3354 r
= apply_protect_kernel_modules(unit
, context
);
3356 *exit_status
= EXIT_SECCOMP
;
3357 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3360 r
= apply_private_devices(unit
, context
);
3362 *exit_status
= EXIT_SECCOMP
;
3363 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3366 r
= apply_syscall_archs(unit
, context
);
3368 *exit_status
= EXIT_SECCOMP
;
3369 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3372 r
= apply_lock_personality(unit
, context
);
3374 *exit_status
= EXIT_SECCOMP
;
3375 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3378 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3379 * by the filter as little as possible. */
3380 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3382 *exit_status
= EXIT_SECCOMP
;
3383 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3388 if (!strv_isempty(context
->unset_environment
)) {
3391 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3393 *exit_status
= EXIT_MEMORY
;
3397 strv_free(accum_env
);
3401 final_argv
= replace_env_argv(argv
, accum_env
);
3403 *exit_status
= EXIT_MEMORY
;
3407 if (DEBUG_LOGGING
) {
3408 _cleanup_free_
char *line
;
3410 line
= exec_command_line(final_argv
);
3412 log_struct(LOG_DEBUG
,
3413 "EXECUTABLE=%s", command
->path
,
3414 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3416 LOG_UNIT_INVOCATION_ID(unit
),
3421 execve(command
->path
, final_argv
, accum_env
);
3423 if (errno
== ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3425 log_struct_errno(LOG_INFO
, errno
,
3426 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3428 LOG_UNIT_INVOCATION_ID(unit
),
3429 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3431 "EXECUTABLE=%s", command
->path
,
3437 *exit_status
= EXIT_EXEC
;
3438 return log_unit_error_errno(unit
, errno
, "Failed to execute command: %m");
3441 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3442 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3444 int exec_spawn(Unit
*unit
,
3445 ExecCommand
*command
,
3446 const ExecContext
*context
,
3447 const ExecParameters
*params
,
3448 ExecRuntime
*runtime
,
3449 DynamicCreds
*dcreds
,
3452 _cleanup_strv_free_
char **files_env
= NULL
;
3454 unsigned n_storage_fds
= 0, n_socket_fds
= 0;
3455 _cleanup_free_
char *line
= NULL
;
3457 int named_iofds
[3] = { -1, -1, -1 };
3466 assert(params
->fds
|| (params
->n_storage_fds
+ params
->n_socket_fds
<= 0));
3468 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3469 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3470 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3472 if (params
->n_socket_fds
> 1) {
3473 log_unit_error(unit
, "Got more than one socket.");
3477 if (params
->n_socket_fds
== 0) {
3478 log_unit_error(unit
, "Got no socket.");
3482 socket_fd
= params
->fds
[0];
3486 n_storage_fds
= params
->n_storage_fds
;
3487 n_socket_fds
= params
->n_socket_fds
;
3490 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3492 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3494 r
= exec_context_load_environment(unit
, context
, &files_env
);
3496 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3498 argv
= params
->argv
?: command
->argv
;
3499 line
= exec_command_line(argv
);
3503 log_struct(LOG_DEBUG
,
3504 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3505 "EXECUTABLE=%s", command
->path
,
3507 LOG_UNIT_INVOCATION_ID(unit
),
3512 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3515 int exit_status
= EXIT_SUCCESS
;
3517 r
= exec_child(unit
,
3530 unit
->manager
->user_lookup_fds
[1],
3534 log_struct_errno(LOG_ERR
, r
,
3535 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3537 LOG_UNIT_INVOCATION_ID(unit
),
3538 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3539 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3541 "EXECUTABLE=%s", command
->path
,
3548 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3550 /* We add the new process to the cgroup both in the child (so
3551 * that we can be sure that no user code is ever executed
3552 * outside of the cgroup) and in the parent (so that we can be
3553 * sure that when we kill the cgroup the process will be
3555 if (params
->cgroup_path
)
3556 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3558 exec_status_start(&command
->exec_status
, pid
);
3564 void exec_context_init(ExecContext
*c
) {
3565 ExecDirectoryType i
;
3570 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3571 c
->cpu_sched_policy
= SCHED_OTHER
;
3572 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3573 c
->syslog_level_prefix
= true;
3574 c
->ignore_sigpipe
= true;
3575 c
->timer_slack_nsec
= NSEC_INFINITY
;
3576 c
->personality
= PERSONALITY_INVALID
;
3577 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3578 c
->directories
[i
].mode
= 0755;
3579 c
->capability_bounding_set
= CAP_ALL
;
3580 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
3581 c
->log_level_max
= -1;
3584 void exec_context_done(ExecContext
*c
) {
3585 ExecDirectoryType i
;
3590 c
->environment
= strv_free(c
->environment
);
3591 c
->environment_files
= strv_free(c
->environment_files
);
3592 c
->pass_environment
= strv_free(c
->pass_environment
);
3593 c
->unset_environment
= strv_free(c
->unset_environment
);
3595 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
3596 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
3598 for (l
= 0; l
< 3; l
++) {
3599 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3600 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3603 c
->working_directory
= mfree(c
->working_directory
);
3604 c
->root_directory
= mfree(c
->root_directory
);
3605 c
->root_image
= mfree(c
->root_image
);
3606 c
->tty_path
= mfree(c
->tty_path
);
3607 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3608 c
->user
= mfree(c
->user
);
3609 c
->group
= mfree(c
->group
);
3611 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3613 c
->pam_name
= mfree(c
->pam_name
);
3615 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3616 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3617 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3619 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3620 c
->bind_mounts
= NULL
;
3621 c
->n_bind_mounts
= 0;
3622 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3623 c
->temporary_filesystems
= NULL
;
3624 c
->n_temporary_filesystems
= 0;
3626 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3628 c
->utmp_id
= mfree(c
->utmp_id
);
3629 c
->selinux_context
= mfree(c
->selinux_context
);
3630 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3631 c
->smack_process_label
= mfree(c
->smack_process_label
);
3633 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3634 c
->syscall_archs
= set_free(c
->syscall_archs
);
3635 c
->address_families
= set_free(c
->address_families
);
3637 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3638 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3640 c
->log_level_max
= -1;
3642 exec_context_free_log_extra_fields(c
);
3644 c
->stdin_data
= mfree(c
->stdin_data
);
3645 c
->stdin_data_size
= 0;
3648 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3653 if (!runtime_prefix
)
3656 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3657 _cleanup_free_
char *p
;
3659 p
= strjoin(runtime_prefix
, "/", *i
);
3663 /* We execute this synchronously, since we need to be sure this is gone when we start the service
3665 (void) rm_rf(p
, REMOVE_ROOT
);
3671 static void exec_command_done(ExecCommand
*c
) {
3674 c
->path
= mfree(c
->path
);
3676 c
->argv
= strv_free(c
->argv
);
3679 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3682 for (i
= 0; i
< n
; i
++)
3683 exec_command_done(c
+i
);
3686 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3690 LIST_REMOVE(command
, c
, i
);
3691 exec_command_done(i
);
3698 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3701 for (i
= 0; i
< n
; i
++)
3702 c
[i
] = exec_command_free_list(c
[i
]);
3705 typedef struct InvalidEnvInfo
{
3710 static void invalid_env(const char *p
, void *userdata
) {
3711 InvalidEnvInfo
*info
= userdata
;
3713 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3716 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3722 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3725 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3728 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3731 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3734 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3737 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3744 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3745 unsigned i
, targets
;
3746 const char* stdio_fdname
[3];
3752 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3753 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3754 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3756 for (i
= 0; i
< 3; i
++)
3757 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3759 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3761 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3762 if (named_iofds
[STDIN_FILENO
] < 0 &&
3763 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3764 stdio_fdname
[STDIN_FILENO
] &&
3765 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3767 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3770 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3771 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3772 stdio_fdname
[STDOUT_FILENO
] &&
3773 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3775 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3778 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3779 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3780 stdio_fdname
[STDERR_FILENO
] &&
3781 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3783 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3787 return targets
== 0 ? 0 : -ENOENT
;
3790 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
3791 char **i
, **r
= NULL
;
3796 STRV_FOREACH(i
, c
->environment_files
) {
3800 bool ignore
= false;
3802 _cleanup_globfree_ glob_t pglob
= {};
3811 if (!path_is_absolute(fn
)) {
3819 /* Filename supports globbing, take all matching files */
3820 k
= safe_glob(fn
, 0, &pglob
);
3829 /* When we don't match anything, -ENOENT should be returned */
3830 assert(pglob
.gl_pathc
> 0);
3832 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
3833 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3841 /* Log invalid environment variables with filename */
3843 InvalidEnvInfo info
= {
3845 .path
= pglob
.gl_pathv
[n
]
3848 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3856 m
= strv_env_merge(2, r
, p
);
3872 static bool tty_may_match_dev_console(const char *tty
) {
3873 _cleanup_free_
char *resolved
= NULL
;
3878 tty
= skip_dev_prefix(tty
);
3880 /* trivial identity? */
3881 if (streq(tty
, "console"))
3884 if (resolve_dev_console(&resolved
) < 0)
3885 return true; /* if we could not resolve, assume it may */
3887 /* "tty0" means the active VC, so it may be the same sometimes */
3888 return streq(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
3891 bool exec_context_may_touch_console(const ExecContext
*ec
) {
3893 return (ec
->tty_reset
||
3895 ec
->tty_vt_disallocate
||
3896 is_terminal_input(ec
->std_input
) ||
3897 is_terminal_output(ec
->std_output
) ||
3898 is_terminal_output(ec
->std_error
)) &&
3899 tty_may_match_dev_console(exec_context_tty_path(ec
));
3902 static void strv_fprintf(FILE *f
, char **l
) {
3908 fprintf(f
, " %s", *g
);
3911 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
3912 ExecDirectoryType dt
;
3920 prefix
= strempty(prefix
);
3924 "%sWorkingDirectory: %s\n"
3925 "%sRootDirectory: %s\n"
3926 "%sNonBlocking: %s\n"
3927 "%sPrivateTmp: %s\n"
3928 "%sPrivateDevices: %s\n"
3929 "%sProtectKernelTunables: %s\n"
3930 "%sProtectKernelModules: %s\n"
3931 "%sProtectControlGroups: %s\n"
3932 "%sPrivateNetwork: %s\n"
3933 "%sPrivateUsers: %s\n"
3934 "%sProtectHome: %s\n"
3935 "%sProtectSystem: %s\n"
3936 "%sMountAPIVFS: %s\n"
3937 "%sIgnoreSIGPIPE: %s\n"
3938 "%sMemoryDenyWriteExecute: %s\n"
3939 "%sRestrictRealtime: %s\n"
3940 "%sKeyringMode: %s\n",
3942 prefix
, c
->working_directory
? c
->working_directory
: "/",
3943 prefix
, c
->root_directory
? c
->root_directory
: "/",
3944 prefix
, yes_no(c
->non_blocking
),
3945 prefix
, yes_no(c
->private_tmp
),
3946 prefix
, yes_no(c
->private_devices
),
3947 prefix
, yes_no(c
->protect_kernel_tunables
),
3948 prefix
, yes_no(c
->protect_kernel_modules
),
3949 prefix
, yes_no(c
->protect_control_groups
),
3950 prefix
, yes_no(c
->private_network
),
3951 prefix
, yes_no(c
->private_users
),
3952 prefix
, protect_home_to_string(c
->protect_home
),
3953 prefix
, protect_system_to_string(c
->protect_system
),
3954 prefix
, yes_no(c
->mount_apivfs
),
3955 prefix
, yes_no(c
->ignore_sigpipe
),
3956 prefix
, yes_no(c
->memory_deny_write_execute
),
3957 prefix
, yes_no(c
->restrict_realtime
),
3958 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
));
3961 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3963 STRV_FOREACH(e
, c
->environment
)
3964 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3966 STRV_FOREACH(e
, c
->environment_files
)
3967 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3969 STRV_FOREACH(e
, c
->pass_environment
)
3970 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3972 STRV_FOREACH(e
, c
->unset_environment
)
3973 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
3975 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
3977 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3978 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
3980 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
3981 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
3989 if (c
->oom_score_adjust_set
)
3991 "%sOOMScoreAdjust: %i\n",
3992 prefix
, c
->oom_score_adjust
);
3994 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3996 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3997 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3998 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3999 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4002 if (c
->ioprio_set
) {
4003 _cleanup_free_
char *class_str
= NULL
;
4005 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4007 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4009 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4012 if (c
->cpu_sched_set
) {
4013 _cleanup_free_
char *policy_str
= NULL
;
4015 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4017 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4020 "%sCPUSchedulingPriority: %i\n"
4021 "%sCPUSchedulingResetOnFork: %s\n",
4022 prefix
, c
->cpu_sched_priority
,
4023 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4027 fprintf(f
, "%sCPUAffinity:", prefix
);
4028 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4029 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4030 fprintf(f
, " %u", i
);
4034 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4035 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4038 "%sStandardInput: %s\n"
4039 "%sStandardOutput: %s\n"
4040 "%sStandardError: %s\n",
4041 prefix
, exec_input_to_string(c
->std_input
),
4042 prefix
, exec_output_to_string(c
->std_output
),
4043 prefix
, exec_output_to_string(c
->std_error
));
4045 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4046 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4047 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4048 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4049 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4050 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4052 if (c
->std_input
== EXEC_INPUT_FILE
)
4053 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4054 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4055 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4056 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4057 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4063 "%sTTYVHangup: %s\n"
4064 "%sTTYVTDisallocate: %s\n",
4065 prefix
, c
->tty_path
,
4066 prefix
, yes_no(c
->tty_reset
),
4067 prefix
, yes_no(c
->tty_vhangup
),
4068 prefix
, yes_no(c
->tty_vt_disallocate
));
4070 if (IN_SET(c
->std_output
,
4073 EXEC_OUTPUT_JOURNAL
,
4074 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4075 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4076 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4077 IN_SET(c
->std_error
,
4080 EXEC_OUTPUT_JOURNAL
,
4081 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4082 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4083 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4085 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4087 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4089 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4091 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4093 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4096 if (c
->log_level_max
>= 0) {
4097 _cleanup_free_
char *t
= NULL
;
4099 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4101 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4104 if (c
->n_log_extra_fields
> 0) {
4107 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4108 fprintf(f
, "%sLogExtraFields: ", prefix
);
4109 fwrite(c
->log_extra_fields
[j
].iov_base
,
4110 1, c
->log_extra_fields
[j
].iov_len
,
4116 if (c
->secure_bits
) {
4117 _cleanup_free_
char *str
= NULL
;
4119 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4121 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4124 if (c
->capability_bounding_set
!= CAP_ALL
) {
4125 _cleanup_free_
char *str
= NULL
;
4127 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4129 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4132 if (c
->capability_ambient_set
!= 0) {
4133 _cleanup_free_
char *str
= NULL
;
4135 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4137 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4141 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4143 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4145 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4147 if (!strv_isempty(c
->supplementary_groups
)) {
4148 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4149 strv_fprintf(f
, c
->supplementary_groups
);
4154 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4156 if (!strv_isempty(c
->read_write_paths
)) {
4157 fprintf(f
, "%sReadWritePaths:", prefix
);
4158 strv_fprintf(f
, c
->read_write_paths
);
4162 if (!strv_isempty(c
->read_only_paths
)) {
4163 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4164 strv_fprintf(f
, c
->read_only_paths
);
4168 if (!strv_isempty(c
->inaccessible_paths
)) {
4169 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4170 strv_fprintf(f
, c
->inaccessible_paths
);
4174 if (c
->n_bind_mounts
> 0)
4175 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4176 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4177 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4178 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4179 c
->bind_mounts
[i
].source
,
4180 c
->bind_mounts
[i
].destination
,
4181 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4183 if (c
->n_temporary_filesystems
> 0)
4184 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4185 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4187 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4189 isempty(t
->options
) ? "" : ":",
4190 strempty(t
->options
));
4195 "%sUtmpIdentifier: %s\n",
4196 prefix
, c
->utmp_id
);
4198 if (c
->selinux_context
)
4200 "%sSELinuxContext: %s%s\n",
4201 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4203 if (c
->apparmor_profile
)
4205 "%sAppArmorProfile: %s%s\n",
4206 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4208 if (c
->smack_process_label
)
4210 "%sSmackProcessLabel: %s%s\n",
4211 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4213 if (c
->personality
!= PERSONALITY_INVALID
)
4215 "%sPersonality: %s\n",
4216 prefix
, strna(personality_to_string(c
->personality
)));
4219 "%sLockPersonality: %s\n",
4220 prefix
, yes_no(c
->lock_personality
));
4222 if (c
->syscall_filter
) {
4230 "%sSystemCallFilter: ",
4233 if (!c
->syscall_whitelist
)
4237 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4238 _cleanup_free_
char *name
= NULL
;
4239 const char *errno_name
= NULL
;
4240 int num
= PTR_TO_INT(val
);
4247 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4248 fputs(strna(name
), f
);
4251 errno_name
= errno_to_name(num
);
4253 fprintf(f
, ":%s", errno_name
);
4255 fprintf(f
, ":%d", num
);
4263 if (c
->syscall_archs
) {
4270 "%sSystemCallArchitectures:",
4274 SET_FOREACH(id
, c
->syscall_archs
, j
)
4275 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4280 if (exec_context_restrict_namespaces_set(c
)) {
4281 _cleanup_free_
char *s
= NULL
;
4283 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
4285 fprintf(f
, "%sRestrictNamespaces: %s\n",
4289 if (c
->syscall_errno
> 0) {
4290 const char *errno_name
;
4292 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4294 errno_name
= errno_to_name(c
->syscall_errno
);
4296 fprintf(f
, "%s\n", errno_name
);
4298 fprintf(f
, "%d\n", c
->syscall_errno
);
4301 if (c
->apparmor_profile
)
4303 "%sAppArmorProfile: %s%s\n",
4304 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4307 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4310 /* Returns true if the process forked off would run under
4311 * an unchanged UID or as root. */
4316 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4322 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4330 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4332 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4337 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4342 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4343 free(c
->log_extra_fields
[l
].iov_base
);
4344 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4345 c
->n_log_extra_fields
= 0;
4348 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4353 dual_timestamp_get(&s
->start_timestamp
);
4356 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4359 if (s
->pid
&& s
->pid
!= pid
)
4363 dual_timestamp_get(&s
->exit_timestamp
);
4369 if (context
->utmp_id
)
4370 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4372 exec_context_tty_reset(context
, NULL
);
4376 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4377 char buf
[FORMAT_TIMESTAMP_MAX
];
4385 prefix
= strempty(prefix
);
4388 "%sPID: "PID_FMT
"\n",
4391 if (dual_timestamp_is_set(&s
->start_timestamp
))
4393 "%sStart Timestamp: %s\n",
4394 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4396 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4398 "%sExit Timestamp: %s\n"
4400 "%sExit Status: %i\n",
4401 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4402 prefix
, sigchld_code_to_string(s
->code
),
4406 static char *exec_command_line(char **argv
) {
4414 STRV_FOREACH(a
, argv
)
4422 STRV_FOREACH(a
, argv
) {
4429 if (strpbrk(*a
, WHITESPACE
)) {
4440 /* FIXME: this doesn't really handle arguments that have
4441 * spaces and ticks in them */
4446 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4447 _cleanup_free_
char *cmd
= NULL
;
4448 const char *prefix2
;
4453 prefix
= strempty(prefix
);
4454 prefix2
= strjoina(prefix
, "\t");
4456 cmd
= exec_command_line(c
->argv
);
4458 "%sCommand Line: %s\n",
4459 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4461 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4464 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4467 prefix
= strempty(prefix
);
4469 LIST_FOREACH(command
, c
, c
)
4470 exec_command_dump(c
, f
, prefix
);
4473 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4480 /* It's kind of important, that we keep the order here */
4481 LIST_FIND_TAIL(command
, *l
, end
);
4482 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4487 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4495 l
= strv_new_ap(path
, ap
);
4516 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4517 _cleanup_strv_free_
char **l
= NULL
;
4525 l
= strv_new_ap(path
, ap
);
4531 r
= strv_extend_strv(&c
->argv
, l
, false);
4538 static void *remove_tmpdir_thread(void *p
) {
4539 _cleanup_free_
char *path
= p
;
4541 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4545 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4552 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4554 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4555 if (destroy
&& rt
->tmp_dir
) {
4556 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4558 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4560 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4567 if (destroy
&& rt
->var_tmp_dir
) {
4568 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4570 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4572 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4573 free(rt
->var_tmp_dir
);
4576 rt
->var_tmp_dir
= NULL
;
4579 rt
->id
= mfree(rt
->id
);
4580 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4581 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4582 safe_close_pair(rt
->netns_storage_socket
);
4586 static void exec_runtime_freep(ExecRuntime
**rt
) {
4588 (void) exec_runtime_free(*rt
, false);
4591 static int exec_runtime_allocate(ExecRuntime
**rt
) {
4594 *rt
= new0(ExecRuntime
, 1);
4598 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
4602 static int exec_runtime_add(
4605 const char *tmp_dir
,
4606 const char *var_tmp_dir
,
4607 const int netns_storage_socket
[2],
4608 ExecRuntime
**ret
) {
4610 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4616 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4620 r
= exec_runtime_allocate(&rt
);
4624 rt
->id
= strdup(id
);
4629 rt
->tmp_dir
= strdup(tmp_dir
);
4633 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4634 assert(var_tmp_dir
);
4635 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4636 if (!rt
->var_tmp_dir
)
4640 if (netns_storage_socket
) {
4641 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4642 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4645 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4654 /* do not remove created ExecRuntime object when the operation succeeds. */
4659 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4660 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4661 _cleanup_close_pair_
int netns_storage_socket
[2] = {-1, -1};
4668 /* It is not necessary to create ExecRuntime object. */
4669 if (!c
->private_network
&& !c
->private_tmp
)
4672 if (c
->private_tmp
) {
4673 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
4678 if (c
->private_network
) {
4679 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
4683 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
4688 netns_storage_socket
[0] = -1;
4689 netns_storage_socket
[1] = -1;
4693 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
4701 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
4703 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
4709 /* If not found, then create a new object. */
4710 r
= exec_runtime_make(m
, c
, id
, &rt
);
4712 /* When r == 0, it is not necessary to create ExecRuntime object. */
4716 /* increment reference counter. */
4722 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
4726 assert(rt
->n_ref
> 0);
4732 return exec_runtime_free(rt
, destroy
);
4735 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
4743 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
4744 fprintf(f
, "exec-runtime=%s", rt
->id
);
4747 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
4749 if (rt
->var_tmp_dir
)
4750 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
4752 if (rt
->netns_storage_socket
[0] >= 0) {
4755 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4759 fprintf(f
, " netns-socket-0=%i", copy
);
4762 if (rt
->netns_storage_socket
[1] >= 0) {
4765 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
4769 fprintf(f
, " netns-socket-1=%i", copy
);
4778 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
4779 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
4783 /* This is for the migration from old (v237 or earlier) deserialization text.
4784 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
4785 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
4786 * so or not from the serialized text, then we always creates a new object owned by this. */
4792 /* Manager manages ExecRuntime objects by the unit id.
4793 * So, we omit the serialized text when the unit does not have id (yet?)... */
4794 if (isempty(u
->id
)) {
4795 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
4799 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
4801 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
4805 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
4807 r
= exec_runtime_allocate(&rt_create
);
4811 rt_create
->id
= strdup(u
->id
);
4818 if (streq(key
, "tmp-dir")) {
4821 copy
= strdup(value
);
4825 free_and_replace(rt
->tmp_dir
, copy
);
4827 } else if (streq(key
, "var-tmp-dir")) {
4830 copy
= strdup(value
);
4834 free_and_replace(rt
->var_tmp_dir
, copy
);
4836 } else if (streq(key
, "netns-socket-0")) {
4839 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
4840 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4844 safe_close(rt
->netns_storage_socket
[0]);
4845 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4847 } else if (streq(key
, "netns-socket-1")) {
4850 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
4851 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4855 safe_close(rt
->netns_storage_socket
[1]);
4856 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
4861 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
4863 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
4865 log_unit_debug_errno(u
, r
, "Failed to put runtime paramter to manager's storage: %m");
4869 rt_create
->manager
= u
->manager
;
4878 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
4879 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4880 int r
, fd0
= -1, fd1
= -1;
4881 const char *p
, *v
= value
;
4888 n
= strcspn(v
, " ");
4889 id
= strndupa(v
, n
);
4894 v
= startswith(p
, "tmp-dir=");
4896 n
= strcspn(v
, " ");
4897 tmp_dir
= strndupa(v
, n
);
4903 v
= startswith(p
, "var-tmp-dir=");
4905 n
= strcspn(v
, " ");
4906 var_tmp_dir
= strndupa(v
, n
);
4912 v
= startswith(p
, "netns-socket-0=");
4916 n
= strcspn(v
, " ");
4917 buf
= strndupa(v
, n
);
4918 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
4919 log_debug("Unable to process exec-runtime netns fd specification.");
4922 fd0
= fdset_remove(fds
, fd0
);
4928 v
= startswith(p
, "netns-socket-1=");
4932 n
= strcspn(v
, " ");
4933 buf
= strndupa(v
, n
);
4934 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
4935 log_debug("Unable to process exec-runtime netns fd specification.");
4938 fd1
= fdset_remove(fds
, fd1
);
4943 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
4945 log_debug_errno(r
, "Failed to add exec-runtime: %m");
4950 void exec_runtime_vacuum(Manager
*m
) {
4956 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
4958 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
4962 (void) exec_runtime_free(rt
, false);
4966 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4967 [EXEC_INPUT_NULL
] = "null",
4968 [EXEC_INPUT_TTY
] = "tty",
4969 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4970 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4971 [EXEC_INPUT_SOCKET
] = "socket",
4972 [EXEC_INPUT_NAMED_FD
] = "fd",
4973 [EXEC_INPUT_DATA
] = "data",
4974 [EXEC_INPUT_FILE
] = "file",
4977 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4979 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4980 [EXEC_OUTPUT_INHERIT
] = "inherit",
4981 [EXEC_OUTPUT_NULL
] = "null",
4982 [EXEC_OUTPUT_TTY
] = "tty",
4983 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4984 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4985 [EXEC_OUTPUT_KMSG
] = "kmsg",
4986 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4987 [EXEC_OUTPUT_JOURNAL
] = "journal",
4988 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4989 [EXEC_OUTPUT_SOCKET
] = "socket",
4990 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4991 [EXEC_OUTPUT_FILE
] = "file",
4994 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4996 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4997 [EXEC_UTMP_INIT
] = "init",
4998 [EXEC_UTMP_LOGIN
] = "login",
4999 [EXEC_UTMP_USER
] = "user",
5002 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5004 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5005 [EXEC_PRESERVE_NO
] = "no",
5006 [EXEC_PRESERVE_YES
] = "yes",
5007 [EXEC_PRESERVE_RESTART
] = "restart",
5010 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5012 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5013 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5014 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5015 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5016 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5017 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5020 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5022 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5023 [EXEC_KEYRING_INHERIT
] = "inherit",
5024 [EXEC_KEYRING_PRIVATE
] = "private",
5025 [EXEC_KEYRING_SHARED
] = "shared",
5028 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);