2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
27 #include <sys/capability.h>
28 #include <sys/eventfd.h>
30 #include <sys/personality.h>
31 #include <sys/prctl.h>
32 #include <sys/socket.h>
39 #include <security/pam_appl.h>
43 #include <selinux/selinux.h>
51 #include <sys/apparmor.h>
54 #include "sd-messages.h"
57 #include "alloc-util.h"
59 #include "apparmor-util.h"
64 #include "capability-util.h"
67 #include "errno-list.h"
69 #include "exit-status.h"
72 #include "formats-util.h"
74 #include "glob-util.h"
81 #include "namespace.h"
82 #include "parse-util.h"
83 #include "path-util.h"
84 #include "process-util.h"
85 #include "rlimit-util.h"
88 #include "seccomp-util.h"
90 #include "securebits.h"
91 #include "selinux-util.h"
92 #include "signal-util.h"
93 #include "smack-util.h"
95 #include "string-table.h"
96 #include "string-util.h"
98 #include "syslog-util.h"
99 #include "terminal-util.h"
101 #include "user-util.h"
103 #include "utmp-wtmp.h"
105 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
106 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
108 /* This assumes there is a 'tty' group */
109 #define TTY_MODE 0620
111 #define SNDBUF_SIZE (8*1024*1024)
113 static int shift_fds(int fds
[], unsigned n_fds
) {
114 int start
, restart_from
;
119 /* Modifies the fds array! (sorts it) */
129 for (i
= start
; i
< (int) n_fds
; i
++) {
132 /* Already at right index? */
136 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
143 /* Hmm, the fd we wanted isn't free? Then
144 * let's remember that and try again from here */
145 if (nfd
!= i
+3 && restart_from
< 0)
149 if (restart_from
< 0)
152 start
= restart_from
;
158 static int flags_fds(const int fds
[], unsigned n_fds
, bool nonblock
) {
167 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
169 for (i
= 0; i
< n_fds
; i
++) {
171 r
= fd_nonblock(fds
[i
], nonblock
);
175 /* We unconditionally drop FD_CLOEXEC from the fds,
176 * since after all we want to pass these fds to our
179 r
= fd_cloexec(fds
[i
], false);
187 static const char *exec_context_tty_path(const ExecContext
*context
) {
190 if (context
->stdio_as_fds
)
193 if (context
->tty_path
)
194 return context
->tty_path
;
196 return "/dev/console";
199 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
204 path
= exec_context_tty_path(context
);
206 if (context
->tty_vhangup
) {
207 if (p
&& p
->stdin_fd
>= 0)
208 (void) terminal_vhangup_fd(p
->stdin_fd
);
210 (void) terminal_vhangup(path
);
213 if (context
->tty_reset
) {
214 if (p
&& p
->stdin_fd
>= 0)
215 (void) reset_terminal_fd(p
->stdin_fd
, true);
217 (void) reset_terminal(path
);
220 if (context
->tty_vt_disallocate
&& path
)
221 (void) vt_disallocate(path
);
224 static bool is_terminal_input(ExecInput i
) {
227 EXEC_INPUT_TTY_FORCE
,
228 EXEC_INPUT_TTY_FAIL
);
231 static bool is_terminal_output(ExecOutput o
) {
234 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
235 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
236 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
239 static bool exec_context_needs_term(const ExecContext
*c
) {
242 /* Return true if the execution context suggests we should set $TERM to something useful. */
244 if (is_terminal_input(c
->std_input
))
247 if (is_terminal_output(c
->std_output
))
250 if (is_terminal_output(c
->std_error
))
253 return !!c
->tty_path
;
256 static int open_null_as(int flags
, int nfd
) {
261 fd
= open("/dev/null", flags
|O_NOCTTY
);
266 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
274 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
275 union sockaddr_union sa
= {
276 .un
.sun_family
= AF_UNIX
,
277 .un
.sun_path
= "/run/systemd/journal/stdout",
279 uid_t olduid
= UID_INVALID
;
280 gid_t oldgid
= GID_INVALID
;
283 if (gid
!= GID_INVALID
) {
291 if (uid
!= UID_INVALID
) {
301 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
305 /* If we fail to restore the uid or gid, things will likely
306 fail later on. This should only happen if an LSM interferes. */
308 if (uid
!= UID_INVALID
)
309 (void) seteuid(olduid
);
312 if (gid
!= GID_INVALID
)
313 (void) setegid(oldgid
);
318 static int connect_logger_as(
320 const ExecContext
*context
,
330 assert(output
< _EXEC_OUTPUT_MAX
);
334 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
338 r
= connect_journal_socket(fd
, uid
, gid
);
342 if (shutdown(fd
, SHUT_RD
) < 0) {
347 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
357 context
->syslog_identifier
? context
->syslog_identifier
: ident
,
359 context
->syslog_priority
,
360 !!context
->syslog_level_prefix
,
361 output
== EXEC_OUTPUT_SYSLOG
|| output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
362 output
== EXEC_OUTPUT_KMSG
|| output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
,
363 is_terminal_output(output
));
368 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
373 static int open_terminal_as(const char *path
, mode_t mode
, int nfd
) {
379 fd
= open_terminal(path
, mode
| O_NOCTTY
);
384 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
392 static int fixup_input(ExecInput std_input
, int socket_fd
, bool apply_tty_stdin
) {
394 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
395 return EXEC_INPUT_NULL
;
397 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
398 return EXEC_INPUT_NULL
;
403 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
405 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
406 return EXEC_OUTPUT_INHERIT
;
411 static int setup_input(
412 const ExecContext
*context
,
413 const ExecParameters
*params
,
421 if (params
->stdin_fd
>= 0) {
422 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
425 /* Try to make this the controlling tty, if it is a tty, and reset it */
426 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
427 (void) reset_terminal_fd(STDIN_FILENO
, true);
432 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
436 case EXEC_INPUT_NULL
:
437 return open_null_as(O_RDONLY
, STDIN_FILENO
);
440 case EXEC_INPUT_TTY_FORCE
:
441 case EXEC_INPUT_TTY_FAIL
: {
444 fd
= acquire_terminal(exec_context_tty_path(context
),
445 i
== EXEC_INPUT_TTY_FAIL
,
446 i
== EXEC_INPUT_TTY_FORCE
,
452 if (fd
!= STDIN_FILENO
) {
453 r
= dup2(fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
461 case EXEC_INPUT_SOCKET
:
462 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
465 assert_not_reached("Unknown input type");
469 static int setup_output(
471 const ExecContext
*context
,
472 const ExecParameters
*params
,
478 dev_t
*journal_stream_dev
,
479 ino_t
*journal_stream_ino
) {
489 assert(journal_stream_dev
);
490 assert(journal_stream_ino
);
492 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
494 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
497 return STDOUT_FILENO
;
500 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
501 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
504 return STDERR_FILENO
;
507 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
508 o
= fixup_output(context
->std_output
, socket_fd
);
510 if (fileno
== STDERR_FILENO
) {
512 e
= fixup_output(context
->std_error
, socket_fd
);
514 /* This expects the input and output are already set up */
516 /* Don't change the stderr file descriptor if we inherit all
517 * the way and are not on a tty */
518 if (e
== EXEC_OUTPUT_INHERIT
&&
519 o
== EXEC_OUTPUT_INHERIT
&&
520 i
== EXEC_INPUT_NULL
&&
521 !is_terminal_input(context
->std_input
) &&
525 /* Duplicate from stdout if possible */
526 if (e
== o
|| e
== EXEC_OUTPUT_INHERIT
)
527 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
531 } else if (o
== EXEC_OUTPUT_INHERIT
) {
532 /* If input got downgraded, inherit the original value */
533 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
534 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
536 /* If the input is connected to anything that's not a /dev/null, inherit that... */
537 if (i
!= EXEC_INPUT_NULL
)
538 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
540 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
544 /* We need to open /dev/null here anew, to get the right access mode. */
545 return open_null_as(O_WRONLY
, fileno
);
550 case EXEC_OUTPUT_NULL
:
551 return open_null_as(O_WRONLY
, fileno
);
553 case EXEC_OUTPUT_TTY
:
554 if (is_terminal_input(i
))
555 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
557 /* We don't reset the terminal if this is just about output */
558 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
560 case EXEC_OUTPUT_SYSLOG
:
561 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
562 case EXEC_OUTPUT_KMSG
:
563 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
564 case EXEC_OUTPUT_JOURNAL
:
565 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
566 r
= connect_logger_as(unit
, context
, o
, ident
, fileno
, uid
, gid
);
568 log_unit_error_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
569 r
= open_null_as(O_WRONLY
, fileno
);
573 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
574 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
575 * services to detect whether they are connected to the journal or not. */
577 if (fstat(fileno
, &st
) >= 0) {
578 *journal_stream_dev
= st
.st_dev
;
579 *journal_stream_ino
= st
.st_ino
;
584 case EXEC_OUTPUT_SOCKET
:
585 assert(socket_fd
>= 0);
586 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
589 assert_not_reached("Unknown error type");
593 static int chown_terminal(int fd
, uid_t uid
) {
598 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
602 /* This might fail. What matters are the results. */
603 (void) fchown(fd
, uid
, -1);
604 (void) fchmod(fd
, TTY_MODE
);
606 if (fstat(fd
, &st
) < 0)
609 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
615 static int setup_confirm_stdio(int *_saved_stdin
, int *_saved_stdout
) {
616 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
619 assert(_saved_stdin
);
620 assert(_saved_stdout
);
622 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
626 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
627 if (saved_stdout
< 0)
630 fd
= acquire_terminal(
635 DEFAULT_CONFIRM_USEC
);
639 r
= chown_terminal(fd
, getuid());
643 r
= reset_terminal_fd(fd
, true);
647 if (dup2(fd
, STDIN_FILENO
) < 0)
650 if (dup2(fd
, STDOUT_FILENO
) < 0)
657 *_saved_stdin
= saved_stdin
;
658 *_saved_stdout
= saved_stdout
;
660 saved_stdin
= saved_stdout
= -1;
665 _printf_(1, 2) static int write_confirm_message(const char *format
, ...) {
666 _cleanup_close_
int fd
= -1;
671 fd
= open_terminal("/dev/console", O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
675 va_start(ap
, format
);
676 vdprintf(fd
, format
, ap
);
682 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
686 assert(saved_stdout
);
690 if (*saved_stdin
>= 0)
691 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
694 if (*saved_stdout
>= 0)
695 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
698 *saved_stdin
= safe_close(*saved_stdin
);
699 *saved_stdout
= safe_close(*saved_stdout
);
704 static int ask_for_confirmation(char *response
, char **argv
) {
705 int saved_stdout
= -1, saved_stdin
= -1, r
;
706 _cleanup_free_
char *line
= NULL
;
708 r
= setup_confirm_stdio(&saved_stdin
, &saved_stdout
);
712 line
= exec_command_line(argv
);
716 r
= ask_char(response
, "yns", "Execute %s? [Yes, No, Skip] ", line
);
718 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
723 static int enforce_groups(const ExecContext
*context
, const char *username
, gid_t gid
) {
724 bool keep_groups
= false;
729 /* Lookup and set GID and supplementary group list. Here too
730 * we avoid NSS lookups for gid=0. */
732 if (context
->group
|| username
) {
733 /* First step, initialize groups from /etc/groups */
734 if (username
&& gid
!= 0) {
735 if (initgroups(username
, gid
) < 0)
741 /* Second step, set our gids */
742 if (setresgid(gid
, gid
, gid
) < 0)
746 if (context
->supplementary_groups
) {
751 /* Final step, initialize any manually set supplementary groups */
752 assert_se((ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
)) > 0);
754 if (!(gids
= new(gid_t
, ngroups_max
)))
758 k
= getgroups(ngroups_max
, gids
);
766 STRV_FOREACH(i
, context
->supplementary_groups
) {
769 if (k
>= ngroups_max
) {
775 r
= get_group_creds(&g
, gids
+k
);
784 if (maybe_setgroups(k
, gids
) < 0) {
795 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
798 /* Sets (but doesn't look up) the uid and make sure we keep the
799 * capabilities while doing so. */
801 if (context
->capability_ambient_set
!= 0) {
803 /* First step: If we need to keep capabilities but
804 * drop privileges we need to make sure we keep our
805 * caps, while we drop privileges. */
807 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
809 if (prctl(PR_GET_SECUREBITS
) != sb
)
810 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
815 /* Second step: actually set the uids */
816 if (setresuid(uid
, uid
, uid
) < 0)
819 /* At this point we should have all necessary capabilities but
820 are otherwise a normal user. However, the caps might got
821 corrupted due to the setresuid() so we need clean them up
822 later. This is done outside of this call. */
829 static int null_conv(
831 const struct pam_message
**msg
,
832 struct pam_response
**resp
,
835 /* We don't support conversations */
842 static int setup_pam(
848 int fds
[], unsigned n_fds
) {
852 static const struct pam_conv conv
= {
857 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
858 pam_handle_t
*handle
= NULL
;
860 int pam_code
= PAM_SUCCESS
, r
;
861 char **nv
, **e
= NULL
;
862 bool close_session
= false;
863 pid_t pam_pid
= 0, parent_pid
;
870 /* We set up PAM in the parent process, then fork. The child
871 * will then stay around until killed via PR_GET_PDEATHSIG or
872 * systemd via the cgroup logic. It will then remove the PAM
873 * session again. The parent process will exec() the actual
874 * daemon. We do things this way to ensure that the main PID
875 * of the daemon is the one we initially fork()ed. */
877 r
= barrier_create(&barrier
);
881 if (log_get_max_level() < LOG_DEBUG
)
884 pam_code
= pam_start(name
, user
, &conv
, &handle
);
885 if (pam_code
!= PAM_SUCCESS
) {
891 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
892 if (pam_code
!= PAM_SUCCESS
)
896 STRV_FOREACH(nv
, *env
) {
897 pam_code
= pam_putenv(handle
, *nv
);
898 if (pam_code
!= PAM_SUCCESS
)
902 pam_code
= pam_acct_mgmt(handle
, flags
);
903 if (pam_code
!= PAM_SUCCESS
)
906 pam_code
= pam_open_session(handle
, flags
);
907 if (pam_code
!= PAM_SUCCESS
)
910 close_session
= true;
912 e
= pam_getenvlist(handle
);
914 pam_code
= PAM_BUF_ERR
;
918 /* Block SIGTERM, so that we know that it won't get lost in
921 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
923 parent_pid
= getpid();
932 int sig
, ret
= EXIT_PAM
;
934 /* The child's job is to reset the PAM session on
936 barrier_set_role(&barrier
, BARRIER_CHILD
);
938 /* This string must fit in 10 chars (i.e. the length
939 * of "/sbin/init"), to look pretty in /bin/ps */
940 rename_process("(sd-pam)");
942 /* Make sure we don't keep open the passed fds in this
943 child. We assume that otherwise only those fds are
944 open here that have been opened by PAM. */
945 close_many(fds
, n_fds
);
947 /* Drop privileges - we don't need any to pam_close_session
948 * and this will make PR_SET_PDEATHSIG work in most cases.
949 * If this fails, ignore the error - but expect sd-pam threads
950 * to fail to exit normally */
951 if (setresuid(uid
, uid
, uid
) < 0)
952 log_error_errno(r
, "Error: Failed to setresuid() in sd-pam: %m");
954 (void) ignore_signals(SIGPIPE
, -1);
956 /* Wait until our parent died. This will only work if
957 * the above setresuid() succeeds, otherwise the kernel
958 * will not allow unprivileged parents kill their privileged
959 * children this way. We rely on the control groups kill logic
960 * to do the rest for us. */
961 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
964 /* Tell the parent that our setup is done. This is especially
965 * important regarding dropping privileges. Otherwise, unit
966 * setup might race against our setresuid(2) call. */
967 barrier_place(&barrier
);
969 /* Check if our parent process might already have
971 if (getppid() == parent_pid
) {
974 assert_se(sigemptyset(&ss
) >= 0);
975 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
978 if (sigwait(&ss
, &sig
) < 0) {
985 assert(sig
== SIGTERM
);
990 /* If our parent died we'll end the session */
991 if (getppid() != parent_pid
) {
992 pam_code
= pam_close_session(handle
, flags
);
993 if (pam_code
!= PAM_SUCCESS
)
1000 pam_end(handle
, pam_code
| flags
);
1004 barrier_set_role(&barrier
, BARRIER_PARENT
);
1006 /* If the child was forked off successfully it will do all the
1007 * cleanups, so forget about the handle here. */
1010 /* Unblock SIGTERM again in the parent */
1011 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1013 /* We close the log explicitly here, since the PAM modules
1014 * might have opened it, but we don't want this fd around. */
1017 /* Synchronously wait for the child to initialize. We don't care for
1018 * errors as we cannot recover. However, warn loudly if it happens. */
1019 if (!barrier_place_and_sync(&barrier
))
1020 log_error("PAM initialization failed");
1028 if (pam_code
!= PAM_SUCCESS
) {
1029 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1030 r
= -EPERM
; /* PAM errors do not map to errno */
1032 log_error_errno(r
, "PAM failed: %m");
1036 pam_code
= pam_close_session(handle
, flags
);
1038 pam_end(handle
, pam_code
| flags
);
1050 static void rename_process_from_path(const char *path
) {
1051 char process_name
[11];
1055 /* This resulting string must fit in 10 chars (i.e. the length
1056 * of "/sbin/init") to look pretty in /bin/ps */
1060 rename_process("(...)");
1066 /* The end of the process name is usually more
1067 * interesting, since the first bit might just be
1073 process_name
[0] = '(';
1074 memcpy(process_name
+1, p
, l
);
1075 process_name
[1+l
] = ')';
1076 process_name
[1+l
+1] = 0;
1078 rename_process(process_name
);
1083 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1084 if (!is_seccomp_available()) {
1086 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1093 static int apply_seccomp(const Unit
* u
, const ExecContext
*c
) {
1094 uint32_t negative_action
, action
;
1095 scmp_filter_ctx
*seccomp
;
1102 if (skip_seccomp_unavailable(u
, "syscall filtering"))
1105 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1107 seccomp
= seccomp_init(c
->syscall_whitelist
? negative_action
: SCMP_ACT_ALLOW
);
1111 if (c
->syscall_archs
) {
1113 SET_FOREACH(id
, c
->syscall_archs
, i
) {
1114 r
= seccomp_arch_add(seccomp
, PTR_TO_UINT32(id
) - 1);
1122 r
= seccomp_add_secondary_archs(seccomp
);
1127 action
= c
->syscall_whitelist
? SCMP_ACT_ALLOW
: negative_action
;
1128 SET_FOREACH(id
, c
->syscall_filter
, i
) {
1129 r
= seccomp_rule_add(seccomp
, action
, PTR_TO_INT(id
) - 1, 0);
1134 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1138 r
= seccomp_load(seccomp
);
1141 seccomp_release(seccomp
);
1145 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1146 scmp_filter_ctx
*seccomp
;
1152 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1155 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1159 r
= seccomp_add_secondary_archs(seccomp
);
1163 if (c
->address_families_whitelist
) {
1164 int af
, first
= 0, last
= 0;
1167 /* If this is a whitelist, we first block the address
1168 * families that are out of range and then everything
1169 * that is not in the set. First, we find the lowest
1170 * and highest address family in the set. */
1172 SET_FOREACH(afp
, c
->address_families
, i
) {
1173 af
= PTR_TO_INT(afp
);
1175 if (af
<= 0 || af
>= af_max())
1178 if (first
== 0 || af
< first
)
1181 if (last
== 0 || af
> last
)
1185 assert((first
== 0) == (last
== 0));
1189 /* No entries in the valid range, block everything */
1190 r
= seccomp_rule_add(
1192 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1200 /* Block everything below the first entry */
1201 r
= seccomp_rule_add(
1203 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1206 SCMP_A0(SCMP_CMP_LT
, first
));
1210 /* Block everything above the last entry */
1211 r
= seccomp_rule_add(
1213 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1216 SCMP_A0(SCMP_CMP_GT
, last
));
1220 /* Block everything between the first and last
1222 for (af
= 1; af
< af_max(); af
++) {
1224 if (set_contains(c
->address_families
, INT_TO_PTR(af
)))
1227 r
= seccomp_rule_add(
1229 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1232 SCMP_A0(SCMP_CMP_EQ
, af
));
1241 /* If this is a blacklist, then generate one rule for
1242 * each address family that are then combined in OR
1245 SET_FOREACH(af
, c
->address_families
, i
) {
1247 r
= seccomp_rule_add(
1249 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1252 SCMP_A0(SCMP_CMP_EQ
, PTR_TO_INT(af
)));
1258 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1262 r
= seccomp_load(seccomp
);
1265 seccomp_release(seccomp
);
1269 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1270 scmp_filter_ctx
*seccomp
;
1275 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1278 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1282 r
= seccomp_add_secondary_archs(seccomp
);
1286 r
= seccomp_rule_add(
1288 SCMP_ACT_ERRNO(EPERM
),
1291 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
|PROT_WRITE
, PROT_EXEC
|PROT_WRITE
));
1295 r
= seccomp_rule_add(
1297 SCMP_ACT_ERRNO(EPERM
),
1300 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
, PROT_EXEC
));
1304 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1308 r
= seccomp_load(seccomp
);
1311 seccomp_release(seccomp
);
1315 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1316 static const int permitted_policies
[] = {
1322 scmp_filter_ctx
*seccomp
;
1324 int r
, p
, max_policy
= 0;
1328 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1331 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1335 r
= seccomp_add_secondary_archs(seccomp
);
1339 /* Determine the highest policy constant we want to allow */
1340 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1341 if (permitted_policies
[i
] > max_policy
)
1342 max_policy
= permitted_policies
[i
];
1344 /* Go through all policies with lower values than that, and block them -- unless they appear in the
1346 for (p
= 0; p
< max_policy
; p
++) {
1349 /* Check if this is in the whitelist. */
1350 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1351 if (permitted_policies
[i
] == p
) {
1359 /* Deny this policy */
1360 r
= seccomp_rule_add(
1362 SCMP_ACT_ERRNO(EPERM
),
1363 SCMP_SYS(sched_setscheduler
),
1365 SCMP_A1(SCMP_CMP_EQ
, p
));
1370 /* Blacklist all other policies, i.e. the ones with higher values. Note that all comparisons are unsigned here,
1371 * hence no need no check for < 0 values. */
1372 r
= seccomp_rule_add(
1374 SCMP_ACT_ERRNO(EPERM
),
1375 SCMP_SYS(sched_setscheduler
),
1377 SCMP_A1(SCMP_CMP_GT
, max_policy
));
1381 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1385 r
= seccomp_load(seccomp
);
1388 seccomp_release(seccomp
);
1392 static int apply_protect_sysctl(Unit
*u
, const ExecContext
*c
) {
1393 scmp_filter_ctx
*seccomp
;
1398 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1399 * let's protect even those systems where this is left on in the kernel. */
1401 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1404 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1408 r
= seccomp_add_secondary_archs(seccomp
);
1412 r
= seccomp_rule_add(
1414 SCMP_ACT_ERRNO(EPERM
),
1420 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1424 r
= seccomp_load(seccomp
);
1427 seccomp_release(seccomp
);
1431 static int apply_private_devices(Unit
*u
, const ExecContext
*c
) {
1432 const SystemCallFilterSet
*set
;
1433 scmp_filter_ctx
*seccomp
;
1435 bool syscalls_found
= false;
1440 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1442 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1445 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1449 r
= seccomp_add_secondary_archs(seccomp
);
1453 for (set
= syscall_filter_sets
; set
->set_name
; set
++)
1454 if (streq(set
->set_name
, "@raw-io")) {
1455 syscalls_found
= true;
1459 /* We should never fail here */
1460 if (!syscalls_found
) {
1465 NULSTR_FOREACH(sys
, set
->value
) {
1469 #ifndef __NR_s390_pci_mmio_read
1470 if (streq(sys
, "s390_pci_mmio_read"))
1473 #ifndef __NR_s390_pci_mmio_write
1474 if (streq(sys
, "s390_pci_mmio_write"))
1481 id
= seccomp_syscall_resolve_name(sys
);
1483 r
= seccomp_rule_add(
1485 SCMP_ACT_ERRNO(EPERM
),
1491 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1495 r
= seccomp_load(seccomp
);
1498 seccomp_release(seccomp
);
1504 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1507 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1508 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1510 if (idle_pipe
[0] >= 0) {
1513 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1515 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1518 /* Signal systemd that we are bored and want to continue. */
1519 n
= write(idle_pipe
[3], "x", 1);
1521 /* Wait for systemd to react to the signal above. */
1522 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1525 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1529 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1532 static int build_environment(
1534 const ExecContext
*c
,
1535 const ExecParameters
*p
,
1538 const char *username
,
1540 dev_t journal_stream_dev
,
1541 ino_t journal_stream_ino
,
1544 _cleanup_strv_free_
char **our_env
= NULL
;
1551 our_env
= new0(char*, 13);
1556 _cleanup_free_
char *joined
= NULL
;
1558 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid()) < 0)
1560 our_env
[n_env
++] = x
;
1562 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1564 our_env
[n_env
++] = x
;
1566 joined
= strv_join(p
->fd_names
, ":");
1570 x
= strjoin("LISTEN_FDNAMES=", joined
, NULL
);
1573 our_env
[n_env
++] = x
;
1576 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1577 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid()) < 0)
1579 our_env
[n_env
++] = x
;
1581 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1583 our_env
[n_env
++] = x
;
1586 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1587 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1588 * check the database directly. */
1589 if (unit_has_name(u
, SPECIAL_DBUS_SERVICE
)) {
1590 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1593 our_env
[n_env
++] = x
;
1597 x
= strappend("HOME=", home
);
1600 our_env
[n_env
++] = x
;
1604 x
= strappend("LOGNAME=", username
);
1607 our_env
[n_env
++] = x
;
1609 x
= strappend("USER=", username
);
1612 our_env
[n_env
++] = x
;
1616 x
= strappend("SHELL=", shell
);
1619 our_env
[n_env
++] = x
;
1622 if (exec_context_needs_term(c
)) {
1623 const char *tty_path
, *term
= NULL
;
1625 tty_path
= exec_context_tty_path(c
);
1627 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1628 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1629 * passes to PID 1 ends up all the way in the console login shown. */
1631 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1632 term
= getenv("TERM");
1634 term
= default_term_for_tty(tty_path
);
1636 x
= strappend("TERM=", term
);
1639 our_env
[n_env
++] = x
;
1642 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1643 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1646 our_env
[n_env
++] = x
;
1649 our_env
[n_env
++] = NULL
;
1650 assert(n_env
<= 12);
1658 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1659 _cleanup_strv_free_
char **pass_env
= NULL
;
1660 size_t n_env
= 0, n_bufsize
= 0;
1663 STRV_FOREACH(i
, c
->pass_environment
) {
1664 _cleanup_free_
char *x
= NULL
;
1670 x
= strjoin(*i
, "=", v
, NULL
);
1673 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1675 pass_env
[n_env
++] = x
;
1676 pass_env
[n_env
] = NULL
;
1686 static bool exec_needs_mount_namespace(
1687 const ExecContext
*context
,
1688 const ExecParameters
*params
,
1689 ExecRuntime
*runtime
) {
1694 if (!strv_isempty(context
->read_write_paths
) ||
1695 !strv_isempty(context
->read_only_paths
) ||
1696 !strv_isempty(context
->inaccessible_paths
))
1699 if (context
->mount_flags
!= 0)
1702 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1705 if (context
->private_devices
||
1706 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1707 context
->protect_home
!= PROTECT_HOME_NO
||
1708 context
->protect_kernel_tunables
||
1709 context
->protect_control_groups
)
1715 static int setup_private_users(uid_t uid
, gid_t gid
) {
1716 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1717 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1718 _cleanup_close_
int unshare_ready_fd
= -1;
1719 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1725 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1726 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1727 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1728 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1729 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1730 * continues execution normally. */
1732 if (uid
!= 0 && uid_is_valid(uid
))
1734 "0 0 1\n" /* Map root → root */
1735 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1736 uid
, uid
); /* The case where the above is the same */
1738 uid_map
= strdup("0 0 1\n");
1742 if (gid
!= 0 && gid_is_valid(gid
))
1744 "0 0 1\n" /* Map root → root */
1745 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1748 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1752 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1754 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1755 if (unshare_ready_fd
< 0)
1758 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1760 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1768 _cleanup_close_
int fd
= -1;
1772 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1773 * here, after the parent opened its own user namespace. */
1776 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1778 /* Wait until the parent unshared the user namespace */
1779 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1784 /* Disable the setgroups() system call in the child user namespace, for good. */
1785 a
= procfs_file_alloca(ppid
, "setgroups");
1786 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1788 if (errno
!= ENOENT
) {
1793 /* If the file is missing the kernel is too old, let's continue anyway. */
1795 if (write(fd
, "deny\n", 5) < 0) {
1800 fd
= safe_close(fd
);
1803 /* First write the GID map */
1804 a
= procfs_file_alloca(ppid
, "gid_map");
1805 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1810 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1814 fd
= safe_close(fd
);
1816 /* The write the UID map */
1817 a
= procfs_file_alloca(ppid
, "uid_map");
1818 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1823 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1828 _exit(EXIT_SUCCESS
);
1831 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1832 _exit(EXIT_FAILURE
);
1835 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1837 if (unshare(CLONE_NEWUSER
) < 0)
1840 /* Let the child know that the namespace is ready now */
1841 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1844 /* Try to read an error code from the child */
1845 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1848 if (n
== sizeof(r
)) { /* an error code was sent to us */
1853 if (n
!= 0) /* on success we should have read 0 bytes */
1856 r
= wait_for_terminate(pid
, &si
);
1861 /* If something strange happened with the child, let's consider this fatal, too */
1862 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1868 static int setup_runtime_directory(
1869 const ExecContext
*context
,
1870 const ExecParameters
*params
,
1880 STRV_FOREACH(rt
, context
->runtime_directory
) {
1881 _cleanup_free_
char *p
;
1883 p
= strjoin(params
->runtime_prefix
, "/", *rt
, NULL
);
1887 r
= mkdir_p_label(p
, context
->runtime_directory_mode
);
1891 r
= chmod_and_chown(p
, context
->runtime_directory_mode
, uid
, gid
);
1899 static int setup_smack(
1900 const ExecContext
*context
,
1901 const ExecCommand
*command
) {
1909 if (!mac_smack_use())
1912 if (context
->smack_process_label
) {
1913 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1917 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1919 _cleanup_free_
char *exec_label
= NULL
;
1921 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1922 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1925 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
1935 static int compile_read_write_paths(
1936 const ExecContext
*context
,
1937 const ExecParameters
*params
,
1940 _cleanup_strv_free_
char **l
= NULL
;
1943 /* Compile the list of writable paths. This is the combination of the explicitly configured paths, plus all
1944 * runtime directories. */
1946 if (strv_isempty(context
->read_write_paths
) &&
1947 strv_isempty(context
->runtime_directory
)) {
1948 *ret
= NULL
; /* NOP if neither is set */
1952 l
= strv_copy(context
->read_write_paths
);
1956 STRV_FOREACH(rt
, context
->runtime_directory
) {
1959 s
= strjoin(params
->runtime_prefix
, "/", *rt
, NULL
);
1963 if (strv_consume(&l
, s
) < 0)
1973 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
1981 array
[(*n
)++] = pair
[0];
1983 array
[(*n
)++] = pair
[1];
1986 static int close_remaining_fds(
1987 const ExecParameters
*params
,
1988 ExecRuntime
*runtime
,
1989 DynamicCreds
*dcreds
,
1992 int *fds
, unsigned n_fds
) {
1994 unsigned n_dont_close
= 0;
1995 int dont_close
[n_fds
+ 12];
1999 if (params
->stdin_fd
>= 0)
2000 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2001 if (params
->stdout_fd
>= 0)
2002 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2003 if (params
->stderr_fd
>= 0)
2004 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2007 dont_close
[n_dont_close
++] = socket_fd
;
2009 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2010 n_dont_close
+= n_fds
;
2014 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2018 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2020 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2023 if (user_lookup_fd
>= 0)
2024 dont_close
[n_dont_close
++] = user_lookup_fd
;
2026 return close_all_fds(dont_close
, n_dont_close
);
2029 static bool context_has_address_families(const ExecContext
*c
) {
2032 return c
->address_families_whitelist
||
2033 !set_isempty(c
->address_families
);
2036 static bool context_has_syscall_filters(const ExecContext
*c
) {
2039 return c
->syscall_whitelist
||
2040 !set_isempty(c
->syscall_filter
) ||
2041 !set_isempty(c
->syscall_archs
);
2044 static bool context_has_no_new_privileges(const ExecContext
*c
) {
2047 if (c
->no_new_privileges
)
2050 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
2053 return context_has_address_families(c
) || /* we need NNP if we have any form of seccomp and are unprivileged */
2054 c
->memory_deny_write_execute
||
2055 c
->restrict_realtime
||
2056 c
->protect_kernel_tunables
||
2057 context_has_syscall_filters(c
);
2060 static int send_user_lookup(
2068 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2069 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2072 if (user_lookup_fd
< 0)
2075 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2078 if (writev(user_lookup_fd
,
2080 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2081 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2082 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2088 static int exec_child(
2090 ExecCommand
*command
,
2091 const ExecContext
*context
,
2092 const ExecParameters
*params
,
2093 ExecRuntime
*runtime
,
2094 DynamicCreds
*dcreds
,
2097 int *fds
, unsigned n_fds
,
2102 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2103 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2104 const char *username
= NULL
, *home
= NULL
, *shell
= NULL
, *wd
;
2105 dev_t journal_stream_dev
= 0;
2106 ino_t journal_stream_ino
= 0;
2107 bool needs_mount_namespace
;
2108 uid_t uid
= UID_INVALID
;
2109 gid_t gid
= GID_INVALID
;
2116 assert(exit_status
);
2118 rename_process_from_path(command
->path
);
2120 /* We reset exactly these signals, since they are the
2121 * only ones we set to SIG_IGN in the main daemon. All
2122 * others we leave untouched because we set them to
2123 * SIG_DFL or a valid handler initially, both of which
2124 * will be demoted to SIG_DFL. */
2125 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2126 SIGNALS_IGNORE
, -1);
2128 if (context
->ignore_sigpipe
)
2129 (void) ignore_signals(SIGPIPE
, -1);
2131 r
= reset_signal_mask();
2133 *exit_status
= EXIT_SIGNAL_MASK
;
2137 if (params
->idle_pipe
)
2138 do_idle_pipe_dance(params
->idle_pipe
);
2140 /* Close sockets very early to make sure we don't
2141 * block init reexecution because it cannot bind its
2146 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2148 *exit_status
= EXIT_FDS
;
2152 if (!context
->same_pgrp
)
2154 *exit_status
= EXIT_SETSID
;
2158 exec_context_tty_reset(context
, params
);
2160 if (params
->flags
& EXEC_CONFIRM_SPAWN
) {
2163 r
= ask_for_confirmation(&response
, argv
);
2164 if (r
== -ETIMEDOUT
)
2165 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
2167 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-r
));
2168 else if (response
== 's') {
2169 write_confirm_message("Skipping execution.\n");
2170 *exit_status
= EXIT_CONFIRM
;
2172 } else if (response
== 'n') {
2173 write_confirm_message("Failing execution.\n");
2179 if (context
->dynamic_user
&& dcreds
) {
2181 /* Make sure we bypass our own NSS module for any NSS checks */
2182 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2183 *exit_status
= EXIT_USER
;
2187 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2189 *exit_status
= EXIT_USER
;
2193 if (!uid_is_valid(uid
) || !gid_is_valid(gid
)) {
2194 *exit_status
= EXIT_USER
;
2199 username
= dcreds
->user
->name
;
2202 if (context
->user
) {
2203 username
= context
->user
;
2204 r
= get_user_creds_clean(&username
, &uid
, &gid
, &home
, &shell
);
2206 *exit_status
= EXIT_USER
;
2210 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
2211 * (i.e. are "/" or "/bin/nologin"). */
2214 if (context
->group
) {
2215 const char *g
= context
->group
;
2217 r
= get_group_creds(&g
, &gid
);
2219 *exit_status
= EXIT_GROUP
;
2225 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2227 *exit_status
= EXIT_USER
;
2231 user_lookup_fd
= safe_close(user_lookup_fd
);
2233 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2234 * must sure to drop O_NONBLOCK */
2236 (void) fd_nonblock(socket_fd
, false);
2238 r
= setup_input(context
, params
, socket_fd
);
2240 *exit_status
= EXIT_STDIN
;
2244 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2246 *exit_status
= EXIT_STDOUT
;
2250 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2252 *exit_status
= EXIT_STDERR
;
2256 if (params
->cgroup_path
) {
2257 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2259 *exit_status
= EXIT_CGROUP
;
2264 if (context
->oom_score_adjust_set
) {
2265 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2267 /* When we can't make this change due to EPERM, then
2268 * let's silently skip over it. User namespaces
2269 * prohibit write access to this file, and we
2270 * shouldn't trip up over that. */
2272 sprintf(t
, "%i", context
->oom_score_adjust
);
2273 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2274 if (r
== -EPERM
|| r
== -EACCES
) {
2276 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2279 *exit_status
= EXIT_OOM_ADJUST
;
2284 if (context
->nice_set
)
2285 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2286 *exit_status
= EXIT_NICE
;
2290 if (context
->cpu_sched_set
) {
2291 struct sched_param param
= {
2292 .sched_priority
= context
->cpu_sched_priority
,
2295 r
= sched_setscheduler(0,
2296 context
->cpu_sched_policy
|
2297 (context
->cpu_sched_reset_on_fork
?
2298 SCHED_RESET_ON_FORK
: 0),
2301 *exit_status
= EXIT_SETSCHEDULER
;
2306 if (context
->cpuset
)
2307 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2308 *exit_status
= EXIT_CPUAFFINITY
;
2312 if (context
->ioprio_set
)
2313 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2314 *exit_status
= EXIT_IOPRIO
;
2318 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2319 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2320 *exit_status
= EXIT_TIMERSLACK
;
2324 if (context
->personality
!= PERSONALITY_INVALID
)
2325 if (personality(context
->personality
) < 0) {
2326 *exit_status
= EXIT_PERSONALITY
;
2330 if (context
->utmp_id
)
2331 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0), context
->tty_path
,
2332 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2333 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2335 username
? "root" : context
->user
);
2337 if (context
->user
&& is_terminal_input(context
->std_input
)) {
2338 r
= chown_terminal(STDIN_FILENO
, uid
);
2340 *exit_status
= EXIT_STDIN
;
2345 /* If delegation is enabled we'll pass ownership of the cgroup
2346 * (but only in systemd's own controller hierarchy!) to the
2347 * user of the new process. */
2348 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2349 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2351 *exit_status
= EXIT_CGROUP
;
2356 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2358 *exit_status
= EXIT_CGROUP
;
2363 if (!strv_isempty(context
->runtime_directory
) && params
->runtime_prefix
) {
2364 r
= setup_runtime_directory(context
, params
, uid
, gid
);
2366 *exit_status
= EXIT_RUNTIME_DIRECTORY
;
2371 r
= build_environment(
2383 *exit_status
= EXIT_MEMORY
;
2387 r
= build_pass_environment(context
, &pass_env
);
2389 *exit_status
= EXIT_MEMORY
;
2393 accum_env
= strv_env_merge(5,
2394 params
->environment
,
2397 context
->environment
,
2401 *exit_status
= EXIT_MEMORY
;
2404 accum_env
= strv_env_clean(accum_env
);
2406 (void) umask(context
->umask
);
2408 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2409 r
= setup_smack(context
, command
);
2411 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2415 if (context
->pam_name
&& username
) {
2416 r
= setup_pam(context
->pam_name
, username
, uid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2418 *exit_status
= EXIT_PAM
;
2424 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2425 r
= setup_netns(runtime
->netns_storage_socket
);
2427 *exit_status
= EXIT_NETWORK
;
2432 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2433 if (needs_mount_namespace
) {
2434 _cleanup_free_
char **rw
= NULL
;
2435 char *tmp
= NULL
, *var
= NULL
;
2437 /* The runtime struct only contains the parent
2438 * of the private /tmp, which is
2439 * non-accessible to world users. Inside of it
2440 * there's a /tmp that is sticky, and that's
2441 * the one we want to use here. */
2443 if (context
->private_tmp
&& runtime
) {
2444 if (runtime
->tmp_dir
)
2445 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2446 if (runtime
->var_tmp_dir
)
2447 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2450 r
= compile_read_write_paths(context
, params
, &rw
);
2452 *exit_status
= EXIT_NAMESPACE
;
2456 r
= setup_namespace(
2457 (params
->flags
& EXEC_APPLY_CHROOT
) ? context
->root_directory
: NULL
,
2459 context
->read_only_paths
,
2460 context
->inaccessible_paths
,
2463 context
->private_devices
,
2464 context
->protect_kernel_tunables
,
2465 context
->protect_control_groups
,
2466 context
->protect_home
,
2467 context
->protect_system
,
2468 context
->mount_flags
);
2470 /* If we couldn't set up the namespace this is
2471 * probably due to a missing capability. In this case,
2472 * silently proceeed. */
2473 if (r
== -EPERM
|| r
== -EACCES
) {
2475 log_unit_debug_errno(unit
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2478 *exit_status
= EXIT_NAMESPACE
;
2483 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2484 r
= enforce_groups(context
, username
, gid
);
2486 *exit_status
= EXIT_GROUP
;
2491 if (context
->working_directory_home
)
2493 else if (context
->working_directory
)
2494 wd
= context
->working_directory
;
2498 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2499 if (!needs_mount_namespace
&& context
->root_directory
)
2500 if (chroot(context
->root_directory
) < 0) {
2501 *exit_status
= EXIT_CHROOT
;
2505 if (chdir(wd
) < 0 &&
2506 !context
->working_directory_missing_ok
) {
2507 *exit_status
= EXIT_CHDIR
;
2513 d
= strjoina(strempty(context
->root_directory
), "/", strempty(wd
));
2515 !context
->working_directory_missing_ok
) {
2516 *exit_status
= EXIT_CHDIR
;
2522 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) &&
2523 mac_selinux_use() &&
2524 params
->selinux_context_net
&&
2526 !command
->privileged
) {
2528 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2530 *exit_status
= EXIT_SELINUX_CONTEXT
;
2536 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2537 r
= setup_private_users(uid
, gid
);
2539 *exit_status
= EXIT_USER
;
2544 /* We repeat the fd closing here, to make sure that
2545 * nothing is leaked from the PAM modules. Note that
2546 * we are more aggressive this time since socket_fd
2547 * and the netns fds we don't need anymore. The custom
2548 * endpoint fd was needed to upload the policy and can
2549 * now be closed as well. */
2550 r
= close_all_fds(fds
, n_fds
);
2552 r
= shift_fds(fds
, n_fds
);
2554 r
= flags_fds(fds
, n_fds
, context
->non_blocking
);
2556 *exit_status
= EXIT_FDS
;
2560 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2562 int secure_bits
= context
->secure_bits
;
2564 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2566 if (!context
->rlimit
[i
])
2569 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2571 *exit_status
= EXIT_LIMITS
;
2576 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2577 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2578 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2579 *exit_status
= EXIT_LIMITS
;
2584 if (!cap_test_all(context
->capability_bounding_set
)) {
2585 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2587 *exit_status
= EXIT_CAPABILITIES
;
2592 /* This is done before enforce_user, but ambient set
2593 * does not survive over setresuid() if keep_caps is not set. */
2594 if (context
->capability_ambient_set
!= 0) {
2595 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2597 *exit_status
= EXIT_CAPABILITIES
;
2602 if (context
->user
) {
2603 r
= enforce_user(context
, uid
);
2605 *exit_status
= EXIT_USER
;
2608 if (context
->capability_ambient_set
!= 0) {
2610 /* Fix the ambient capabilities after user change. */
2611 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2613 *exit_status
= EXIT_CAPABILITIES
;
2617 /* If we were asked to change user and ambient capabilities
2618 * were requested, we had to add keep-caps to the securebits
2619 * so that we would maintain the inherited capability set
2620 * through the setresuid(). Make sure that the bit is added
2621 * also to the context secure_bits so that we don't try to
2622 * drop the bit away next. */
2624 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2628 /* PR_GET_SECUREBITS is not privileged, while
2629 * PR_SET_SECUREBITS is. So to suppress
2630 * potential EPERMs we'll try not to call
2631 * PR_SET_SECUREBITS unless necessary. */
2632 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2633 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2634 *exit_status
= EXIT_SECUREBITS
;
2638 if (context_has_no_new_privileges(context
))
2639 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2640 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2645 if (context_has_address_families(context
)) {
2646 r
= apply_address_families(unit
, context
);
2648 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2653 if (context
->memory_deny_write_execute
) {
2654 r
= apply_memory_deny_write_execute(unit
, context
);
2656 *exit_status
= EXIT_SECCOMP
;
2661 if (context
->restrict_realtime
) {
2662 r
= apply_restrict_realtime(unit
, context
);
2664 *exit_status
= EXIT_SECCOMP
;
2669 if (context
->protect_kernel_tunables
) {
2670 r
= apply_protect_sysctl(unit
, context
);
2672 *exit_status
= EXIT_SECCOMP
;
2677 if (context
->private_devices
) {
2678 r
= apply_private_devices(unit
, context
);
2680 *exit_status
= EXIT_SECCOMP
;
2685 if (context_has_syscall_filters(context
)) {
2686 r
= apply_seccomp(unit
, context
);
2688 *exit_status
= EXIT_SECCOMP
;
2695 if (mac_selinux_use()) {
2696 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2699 r
= setexeccon(exec_context
);
2701 *exit_status
= EXIT_SELINUX_CONTEXT
;
2708 #ifdef HAVE_APPARMOR
2709 if (context
->apparmor_profile
&& mac_apparmor_use()) {
2710 r
= aa_change_onexec(context
->apparmor_profile
);
2711 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2712 *exit_status
= EXIT_APPARMOR_PROFILE
;
2719 final_argv
= replace_env_argv(argv
, accum_env
);
2721 *exit_status
= EXIT_MEMORY
;
2725 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2726 _cleanup_free_
char *line
;
2728 line
= exec_command_line(final_argv
);
2731 log_struct(LOG_DEBUG
,
2733 "EXECUTABLE=%s", command
->path
,
2734 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2740 execve(command
->path
, final_argv
, accum_env
);
2741 *exit_status
= EXIT_EXEC
;
2745 int exec_spawn(Unit
*unit
,
2746 ExecCommand
*command
,
2747 const ExecContext
*context
,
2748 const ExecParameters
*params
,
2749 ExecRuntime
*runtime
,
2750 DynamicCreds
*dcreds
,
2753 _cleanup_strv_free_
char **files_env
= NULL
;
2754 int *fds
= NULL
; unsigned n_fds
= 0;
2755 _cleanup_free_
char *line
= NULL
;
2765 assert(params
->fds
|| params
->n_fds
<= 0);
2767 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2768 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2769 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
2771 if (params
->n_fds
!= 1) {
2772 log_unit_error(unit
, "Got more than one socket.");
2776 socket_fd
= params
->fds
[0];
2780 n_fds
= params
->n_fds
;
2783 r
= exec_context_load_environment(unit
, context
, &files_env
);
2785 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
2787 argv
= params
->argv
?: command
->argv
;
2788 line
= exec_command_line(argv
);
2792 log_struct(LOG_DEBUG
,
2794 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
2795 "EXECUTABLE=%s", command
->path
,
2799 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
2804 r
= exec_child(unit
,
2814 unit
->manager
->user_lookup_fds
[1],
2818 log_struct_errno(LOG_ERR
, r
,
2819 LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED
),
2821 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
2822 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
2824 "EXECUTABLE=%s", command
->path
,
2831 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
2833 /* We add the new process to the cgroup both in the child (so
2834 * that we can be sure that no user code is ever executed
2835 * outside of the cgroup) and in the parent (so that we can be
2836 * sure that when we kill the cgroup the process will be
2838 if (params
->cgroup_path
)
2839 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
2841 exec_status_start(&command
->exec_status
, pid
);
2847 void exec_context_init(ExecContext
*c
) {
2851 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
2852 c
->cpu_sched_policy
= SCHED_OTHER
;
2853 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
2854 c
->syslog_level_prefix
= true;
2855 c
->ignore_sigpipe
= true;
2856 c
->timer_slack_nsec
= NSEC_INFINITY
;
2857 c
->personality
= PERSONALITY_INVALID
;
2858 c
->runtime_directory_mode
= 0755;
2859 c
->capability_bounding_set
= CAP_ALL
;
2862 void exec_context_done(ExecContext
*c
) {
2867 c
->environment
= strv_free(c
->environment
);
2868 c
->environment_files
= strv_free(c
->environment_files
);
2869 c
->pass_environment
= strv_free(c
->pass_environment
);
2871 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
2872 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
2874 c
->working_directory
= mfree(c
->working_directory
);
2875 c
->root_directory
= mfree(c
->root_directory
);
2876 c
->tty_path
= mfree(c
->tty_path
);
2877 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
2878 c
->user
= mfree(c
->user
);
2879 c
->group
= mfree(c
->group
);
2881 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
2883 c
->pam_name
= mfree(c
->pam_name
);
2885 c
->read_only_paths
= strv_free(c
->read_only_paths
);
2886 c
->read_write_paths
= strv_free(c
->read_write_paths
);
2887 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
2890 CPU_FREE(c
->cpuset
);
2892 c
->utmp_id
= mfree(c
->utmp_id
);
2893 c
->selinux_context
= mfree(c
->selinux_context
);
2894 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
2896 c
->syscall_filter
= set_free(c
->syscall_filter
);
2897 c
->syscall_archs
= set_free(c
->syscall_archs
);
2898 c
->address_families
= set_free(c
->address_families
);
2900 c
->runtime_directory
= strv_free(c
->runtime_directory
);
2903 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
2908 if (!runtime_prefix
)
2911 STRV_FOREACH(i
, c
->runtime_directory
) {
2912 _cleanup_free_
char *p
;
2914 p
= strjoin(runtime_prefix
, "/", *i
, NULL
);
2918 /* We execute this synchronously, since we need to be
2919 * sure this is gone when we start the service
2921 (void) rm_rf(p
, REMOVE_ROOT
);
2927 void exec_command_done(ExecCommand
*c
) {
2930 c
->path
= mfree(c
->path
);
2932 c
->argv
= strv_free(c
->argv
);
2935 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
2938 for (i
= 0; i
< n
; i
++)
2939 exec_command_done(c
+i
);
2942 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
2946 LIST_REMOVE(command
, c
, i
);
2947 exec_command_done(i
);
2954 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
2957 for (i
= 0; i
< n
; i
++)
2958 c
[i
] = exec_command_free_list(c
[i
]);
2961 typedef struct InvalidEnvInfo
{
2966 static void invalid_env(const char *p
, void *userdata
) {
2967 InvalidEnvInfo
*info
= userdata
;
2969 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
2972 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
2973 char **i
, **r
= NULL
;
2978 STRV_FOREACH(i
, c
->environment_files
) {
2981 bool ignore
= false;
2983 _cleanup_globfree_ glob_t pglob
= {};
2993 if (!path_is_absolute(fn
)) {
3001 /* Filename supports globbing, take all matching files */
3003 if (glob(fn
, 0, NULL
, &pglob
) != 0) {
3008 return errno
> 0 ? -errno
: -EINVAL
;
3010 count
= pglob
.gl_pathc
;
3018 for (n
= 0; n
< count
; n
++) {
3019 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3027 /* Log invalid environment variables with filename */
3029 InvalidEnvInfo info
= {
3031 .path
= pglob
.gl_pathv
[n
]
3034 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3042 m
= strv_env_merge(2, r
, p
);
3058 static bool tty_may_match_dev_console(const char *tty
) {
3059 _cleanup_free_
char *active
= NULL
;
3065 if (startswith(tty
, "/dev/"))
3068 /* trivial identity? */
3069 if (streq(tty
, "console"))
3072 console
= resolve_dev_console(&active
);
3073 /* if we could not resolve, assume it may */
3077 /* "tty0" means the active VC, so it may be the same sometimes */
3078 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3081 bool exec_context_may_touch_console(ExecContext
*ec
) {
3083 return (ec
->tty_reset
||
3085 ec
->tty_vt_disallocate
||
3086 is_terminal_input(ec
->std_input
) ||
3087 is_terminal_output(ec
->std_output
) ||
3088 is_terminal_output(ec
->std_error
)) &&
3089 tty_may_match_dev_console(exec_context_tty_path(ec
));
3092 static void strv_fprintf(FILE *f
, char **l
) {
3098 fprintf(f
, " %s", *g
);
3101 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3108 prefix
= strempty(prefix
);
3112 "%sWorkingDirectory: %s\n"
3113 "%sRootDirectory: %s\n"
3114 "%sNonBlocking: %s\n"
3115 "%sPrivateTmp: %s\n"
3116 "%sPrivateDevices: %s\n"
3117 "%sProtectKernelTunables: %s\n"
3118 "%sProtectControlGroups: %s\n"
3119 "%sPrivateNetwork: %s\n"
3120 "%sPrivateUsers: %s\n"
3121 "%sProtectHome: %s\n"
3122 "%sProtectSystem: %s\n"
3123 "%sIgnoreSIGPIPE: %s\n"
3124 "%sMemoryDenyWriteExecute: %s\n"
3125 "%sRestrictRealtime: %s\n",
3127 prefix
, c
->working_directory
? c
->working_directory
: "/",
3128 prefix
, c
->root_directory
? c
->root_directory
: "/",
3129 prefix
, yes_no(c
->non_blocking
),
3130 prefix
, yes_no(c
->private_tmp
),
3131 prefix
, yes_no(c
->private_devices
),
3132 prefix
, yes_no(c
->protect_kernel_tunables
),
3133 prefix
, yes_no(c
->protect_control_groups
),
3134 prefix
, yes_no(c
->private_network
),
3135 prefix
, yes_no(c
->private_users
),
3136 prefix
, protect_home_to_string(c
->protect_home
),
3137 prefix
, protect_system_to_string(c
->protect_system
),
3138 prefix
, yes_no(c
->ignore_sigpipe
),
3139 prefix
, yes_no(c
->memory_deny_write_execute
),
3140 prefix
, yes_no(c
->restrict_realtime
));
3142 STRV_FOREACH(e
, c
->environment
)
3143 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3145 STRV_FOREACH(e
, c
->environment_files
)
3146 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3148 STRV_FOREACH(e
, c
->pass_environment
)
3149 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3151 fprintf(f
, "%sRuntimeDirectoryMode: %04o\n", prefix
, c
->runtime_directory_mode
);
3153 STRV_FOREACH(d
, c
->runtime_directory
)
3154 fprintf(f
, "%sRuntimeDirectory: %s\n", prefix
, *d
);
3161 if (c
->oom_score_adjust_set
)
3163 "%sOOMScoreAdjust: %i\n",
3164 prefix
, c
->oom_score_adjust
);
3166 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3168 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3169 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3170 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3171 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3174 if (c
->ioprio_set
) {
3175 _cleanup_free_
char *class_str
= NULL
;
3177 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3179 "%sIOSchedulingClass: %s\n"
3180 "%sIOPriority: %i\n",
3181 prefix
, strna(class_str
),
3182 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
3185 if (c
->cpu_sched_set
) {
3186 _cleanup_free_
char *policy_str
= NULL
;
3188 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3190 "%sCPUSchedulingPolicy: %s\n"
3191 "%sCPUSchedulingPriority: %i\n"
3192 "%sCPUSchedulingResetOnFork: %s\n",
3193 prefix
, strna(policy_str
),
3194 prefix
, c
->cpu_sched_priority
,
3195 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3199 fprintf(f
, "%sCPUAffinity:", prefix
);
3200 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3201 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3202 fprintf(f
, " %u", i
);
3206 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3207 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3210 "%sStandardInput: %s\n"
3211 "%sStandardOutput: %s\n"
3212 "%sStandardError: %s\n",
3213 prefix
, exec_input_to_string(c
->std_input
),
3214 prefix
, exec_output_to_string(c
->std_output
),
3215 prefix
, exec_output_to_string(c
->std_error
));
3221 "%sTTYVHangup: %s\n"
3222 "%sTTYVTDisallocate: %s\n",
3223 prefix
, c
->tty_path
,
3224 prefix
, yes_no(c
->tty_reset
),
3225 prefix
, yes_no(c
->tty_vhangup
),
3226 prefix
, yes_no(c
->tty_vt_disallocate
));
3228 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3229 c
->std_output
== EXEC_OUTPUT_KMSG
||
3230 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3231 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3232 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3233 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3234 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3235 c
->std_error
== EXEC_OUTPUT_KMSG
||
3236 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3237 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3238 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3239 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3241 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3243 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3244 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3247 "%sSyslogFacility: %s\n"
3248 "%sSyslogLevel: %s\n",
3249 prefix
, strna(fac_str
),
3250 prefix
, strna(lvl_str
));
3254 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
3256 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
3257 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
3258 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
3259 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
3260 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
3261 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
3263 if (c
->capability_bounding_set
!= CAP_ALL
) {
3265 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
3267 for (l
= 0; l
<= cap_last_cap(); l
++)
3268 if (c
->capability_bounding_set
& (UINT64_C(1) << l
))
3269 fprintf(f
, " %s", strna(capability_to_name(l
)));
3274 if (c
->capability_ambient_set
!= 0) {
3276 fprintf(f
, "%sAmbientCapabilities:", prefix
);
3278 for (l
= 0; l
<= cap_last_cap(); l
++)
3279 if (c
->capability_ambient_set
& (UINT64_C(1) << l
))
3280 fprintf(f
, " %s", strna(capability_to_name(l
)));
3286 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3288 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3290 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3292 if (strv_length(c
->supplementary_groups
) > 0) {
3293 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3294 strv_fprintf(f
, c
->supplementary_groups
);
3299 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3301 if (strv_length(c
->read_write_paths
) > 0) {
3302 fprintf(f
, "%sReadWritePaths:", prefix
);
3303 strv_fprintf(f
, c
->read_write_paths
);
3307 if (strv_length(c
->read_only_paths
) > 0) {
3308 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3309 strv_fprintf(f
, c
->read_only_paths
);
3313 if (strv_length(c
->inaccessible_paths
) > 0) {
3314 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3315 strv_fprintf(f
, c
->inaccessible_paths
);
3321 "%sUtmpIdentifier: %s\n",
3322 prefix
, c
->utmp_id
);
3324 if (c
->selinux_context
)
3326 "%sSELinuxContext: %s%s\n",
3327 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3329 if (c
->personality
!= PERSONALITY_INVALID
)
3331 "%sPersonality: %s\n",
3332 prefix
, strna(personality_to_string(c
->personality
)));
3334 if (c
->syscall_filter
) {
3342 "%sSystemCallFilter: ",
3345 if (!c
->syscall_whitelist
)
3349 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3350 _cleanup_free_
char *name
= NULL
;
3357 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3358 fputs(strna(name
), f
);
3365 if (c
->syscall_archs
) {
3372 "%sSystemCallArchitectures:",
3376 SET_FOREACH(id
, c
->syscall_archs
, j
)
3377 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3382 if (c
->syscall_errno
> 0)
3384 "%sSystemCallErrorNumber: %s\n",
3385 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3387 if (c
->apparmor_profile
)
3389 "%sAppArmorProfile: %s%s\n",
3390 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3393 bool exec_context_maintains_privileges(ExecContext
*c
) {
3396 /* Returns true if the process forked off would run under
3397 * an unchanged UID or as root. */
3402 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3408 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3413 dual_timestamp_get(&s
->start_timestamp
);
3416 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3419 if (s
->pid
&& s
->pid
!= pid
)
3423 dual_timestamp_get(&s
->exit_timestamp
);
3429 if (context
->utmp_id
)
3430 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3432 exec_context_tty_reset(context
, NULL
);
3436 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3437 char buf
[FORMAT_TIMESTAMP_MAX
];
3445 prefix
= strempty(prefix
);
3448 "%sPID: "PID_FMT
"\n",
3451 if (dual_timestamp_is_set(&s
->start_timestamp
))
3453 "%sStart Timestamp: %s\n",
3454 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3456 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3458 "%sExit Timestamp: %s\n"
3460 "%sExit Status: %i\n",
3461 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3462 prefix
, sigchld_code_to_string(s
->code
),
3466 char *exec_command_line(char **argv
) {
3474 STRV_FOREACH(a
, argv
)
3477 if (!(n
= new(char, k
)))
3481 STRV_FOREACH(a
, argv
) {
3488 if (strpbrk(*a
, WHITESPACE
)) {
3499 /* FIXME: this doesn't really handle arguments that have
3500 * spaces and ticks in them */
3505 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3506 _cleanup_free_
char *cmd
= NULL
;
3507 const char *prefix2
;
3512 prefix
= strempty(prefix
);
3513 prefix2
= strjoina(prefix
, "\t");
3515 cmd
= exec_command_line(c
->argv
);
3517 "%sCommand Line: %s\n",
3518 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3520 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3523 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3526 prefix
= strempty(prefix
);
3528 LIST_FOREACH(command
, c
, c
)
3529 exec_command_dump(c
, f
, prefix
);
3532 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3539 /* It's kind of important, that we keep the order here */
3540 LIST_FIND_TAIL(command
, *l
, end
);
3541 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3546 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3554 l
= strv_new_ap(path
, ap
);
3575 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3576 _cleanup_strv_free_
char **l
= NULL
;
3584 l
= strv_new_ap(path
, ap
);
3590 r
= strv_extend_strv(&c
->argv
, l
, false);
3598 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3603 *rt
= new0(ExecRuntime
, 1);
3608 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3613 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
3623 if (!c
->private_network
&& !c
->private_tmp
)
3626 r
= exec_runtime_allocate(rt
);
3630 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
3631 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
3635 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
3636 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
3644 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
3646 assert(r
->n_ref
> 0);
3652 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
3657 assert(r
->n_ref
> 0);
3664 free(r
->var_tmp_dir
);
3665 safe_close_pair(r
->netns_storage_socket
);
3671 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
3680 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
3682 if (rt
->var_tmp_dir
)
3683 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
3685 if (rt
->netns_storage_socket
[0] >= 0) {
3688 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
3692 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
3695 if (rt
->netns_storage_socket
[1] >= 0) {
3698 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
3702 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
3708 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
3715 if (streq(key
, "tmp-dir")) {
3718 r
= exec_runtime_allocate(rt
);
3722 copy
= strdup(value
);
3726 free((*rt
)->tmp_dir
);
3727 (*rt
)->tmp_dir
= copy
;
3729 } else if (streq(key
, "var-tmp-dir")) {
3732 r
= exec_runtime_allocate(rt
);
3736 copy
= strdup(value
);
3740 free((*rt
)->var_tmp_dir
);
3741 (*rt
)->var_tmp_dir
= copy
;
3743 } else if (streq(key
, "netns-socket-0")) {
3746 r
= exec_runtime_allocate(rt
);
3750 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
3751 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
3753 safe_close((*rt
)->netns_storage_socket
[0]);
3754 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
3756 } else if (streq(key
, "netns-socket-1")) {
3759 r
= exec_runtime_allocate(rt
);
3763 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
3764 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
3766 safe_close((*rt
)->netns_storage_socket
[1]);
3767 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
3775 static void *remove_tmpdir_thread(void *p
) {
3776 _cleanup_free_
char *path
= p
;
3778 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
3782 void exec_runtime_destroy(ExecRuntime
*rt
) {
3788 /* If there are multiple users of this, let's leave the stuff around */
3793 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
3795 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
3797 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
3804 if (rt
->var_tmp_dir
) {
3805 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
3807 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
3809 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
3810 free(rt
->var_tmp_dir
);
3813 rt
->var_tmp_dir
= NULL
;
3816 safe_close_pair(rt
->netns_storage_socket
);
3819 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
3820 [EXEC_INPUT_NULL
] = "null",
3821 [EXEC_INPUT_TTY
] = "tty",
3822 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
3823 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
3824 [EXEC_INPUT_SOCKET
] = "socket"
3827 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
3829 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
3830 [EXEC_OUTPUT_INHERIT
] = "inherit",
3831 [EXEC_OUTPUT_NULL
] = "null",
3832 [EXEC_OUTPUT_TTY
] = "tty",
3833 [EXEC_OUTPUT_SYSLOG
] = "syslog",
3834 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
3835 [EXEC_OUTPUT_KMSG
] = "kmsg",
3836 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
3837 [EXEC_OUTPUT_JOURNAL
] = "journal",
3838 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
3839 [EXEC_OUTPUT_SOCKET
] = "socket"
3842 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
3844 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
3845 [EXEC_UTMP_INIT
] = "init",
3846 [EXEC_UTMP_LOGIN
] = "login",
3847 [EXEC_UTMP_USER
] = "user",
3850 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);