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(
849 int fds
[], unsigned n_fds
) {
853 static const struct pam_conv conv
= {
858 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
859 pam_handle_t
*handle
= NULL
;
861 int pam_code
= PAM_SUCCESS
, r
;
862 char **nv
, **e
= NULL
;
863 bool close_session
= false;
864 pid_t pam_pid
= 0, parent_pid
;
871 /* We set up PAM in the parent process, then fork. The child
872 * will then stay around until killed via PR_GET_PDEATHSIG or
873 * systemd via the cgroup logic. It will then remove the PAM
874 * session again. The parent process will exec() the actual
875 * daemon. We do things this way to ensure that the main PID
876 * of the daemon is the one we initially fork()ed. */
878 r
= barrier_create(&barrier
);
882 if (log_get_max_level() < LOG_DEBUG
)
885 pam_code
= pam_start(name
, user
, &conv
, &handle
);
886 if (pam_code
!= PAM_SUCCESS
) {
892 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
893 if (pam_code
!= PAM_SUCCESS
)
897 STRV_FOREACH(nv
, *env
) {
898 pam_code
= pam_putenv(handle
, *nv
);
899 if (pam_code
!= PAM_SUCCESS
)
903 pam_code
= pam_acct_mgmt(handle
, flags
);
904 if (pam_code
!= PAM_SUCCESS
)
907 pam_code
= pam_open_session(handle
, flags
);
908 if (pam_code
!= PAM_SUCCESS
)
911 close_session
= true;
913 e
= pam_getenvlist(handle
);
915 pam_code
= PAM_BUF_ERR
;
919 /* Block SIGTERM, so that we know that it won't get lost in
922 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
924 parent_pid
= getpid();
933 int sig
, ret
= EXIT_PAM
;
935 /* The child's job is to reset the PAM session on
937 barrier_set_role(&barrier
, BARRIER_CHILD
);
939 /* This string must fit in 10 chars (i.e. the length
940 * of "/sbin/init"), to look pretty in /bin/ps */
941 rename_process("(sd-pam)");
943 /* Make sure we don't keep open the passed fds in this
944 child. We assume that otherwise only those fds are
945 open here that have been opened by PAM. */
946 close_many(fds
, n_fds
);
948 /* Drop privileges - we don't need any to pam_close_session
949 * and this will make PR_SET_PDEATHSIG work in most cases.
950 * If this fails, ignore the error - but expect sd-pam threads
951 * to fail to exit normally */
953 if (maybe_setgroups(0, NULL
) < 0)
954 log_warning_errno(errno
, "Failed to setgroups() in sd-pam: %m");
955 if (setresgid(gid
, gid
, gid
) < 0)
956 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
957 if (setresuid(uid
, uid
, uid
) < 0)
958 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
960 (void) ignore_signals(SIGPIPE
, -1);
962 /* Wait until our parent died. This will only work if
963 * the above setresuid() succeeds, otherwise the kernel
964 * will not allow unprivileged parents kill their privileged
965 * children this way. We rely on the control groups kill logic
966 * to do the rest for us. */
967 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
970 /* Tell the parent that our setup is done. This is especially
971 * important regarding dropping privileges. Otherwise, unit
972 * setup might race against our setresuid(2) call. */
973 barrier_place(&barrier
);
975 /* Check if our parent process might already have
977 if (getppid() == parent_pid
) {
980 assert_se(sigemptyset(&ss
) >= 0);
981 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
984 if (sigwait(&ss
, &sig
) < 0) {
991 assert(sig
== SIGTERM
);
996 /* If our parent died we'll end the session */
997 if (getppid() != parent_pid
) {
998 pam_code
= pam_close_session(handle
, flags
);
999 if (pam_code
!= PAM_SUCCESS
)
1006 pam_end(handle
, pam_code
| flags
);
1010 barrier_set_role(&barrier
, BARRIER_PARENT
);
1012 /* If the child was forked off successfully it will do all the
1013 * cleanups, so forget about the handle here. */
1016 /* Unblock SIGTERM again in the parent */
1017 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1019 /* We close the log explicitly here, since the PAM modules
1020 * might have opened it, but we don't want this fd around. */
1023 /* Synchronously wait for the child to initialize. We don't care for
1024 * errors as we cannot recover. However, warn loudly if it happens. */
1025 if (!barrier_place_and_sync(&barrier
))
1026 log_error("PAM initialization failed");
1034 if (pam_code
!= PAM_SUCCESS
) {
1035 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1036 r
= -EPERM
; /* PAM errors do not map to errno */
1038 log_error_errno(r
, "PAM failed: %m");
1042 pam_code
= pam_close_session(handle
, flags
);
1044 pam_end(handle
, pam_code
| flags
);
1056 static void rename_process_from_path(const char *path
) {
1057 char process_name
[11];
1061 /* This resulting string must fit in 10 chars (i.e. the length
1062 * of "/sbin/init") to look pretty in /bin/ps */
1066 rename_process("(...)");
1072 /* The end of the process name is usually more
1073 * interesting, since the first bit might just be
1079 process_name
[0] = '(';
1080 memcpy(process_name
+1, p
, l
);
1081 process_name
[1+l
] = ')';
1082 process_name
[1+l
+1] = 0;
1084 rename_process(process_name
);
1089 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1090 if (!is_seccomp_available()) {
1092 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1099 static int apply_seccomp(const Unit
* u
, const ExecContext
*c
) {
1100 uint32_t negative_action
, action
;
1101 scmp_filter_ctx
*seccomp
;
1108 if (skip_seccomp_unavailable(u
, "syscall filtering"))
1111 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1113 seccomp
= seccomp_init(c
->syscall_whitelist
? negative_action
: SCMP_ACT_ALLOW
);
1117 if (c
->syscall_archs
) {
1119 SET_FOREACH(id
, c
->syscall_archs
, i
) {
1120 r
= seccomp_arch_add(seccomp
, PTR_TO_UINT32(id
) - 1);
1128 r
= seccomp_add_secondary_archs(seccomp
);
1133 action
= c
->syscall_whitelist
? SCMP_ACT_ALLOW
: negative_action
;
1134 SET_FOREACH(id
, c
->syscall_filter
, i
) {
1135 r
= seccomp_rule_add(seccomp
, action
, PTR_TO_INT(id
) - 1, 0);
1140 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1144 r
= seccomp_load(seccomp
);
1147 seccomp_release(seccomp
);
1151 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1152 scmp_filter_ctx
*seccomp
;
1158 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1161 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1165 r
= seccomp_add_secondary_archs(seccomp
);
1169 if (c
->address_families_whitelist
) {
1170 int af
, first
= 0, last
= 0;
1173 /* If this is a whitelist, we first block the address
1174 * families that are out of range and then everything
1175 * that is not in the set. First, we find the lowest
1176 * and highest address family in the set. */
1178 SET_FOREACH(afp
, c
->address_families
, i
) {
1179 af
= PTR_TO_INT(afp
);
1181 if (af
<= 0 || af
>= af_max())
1184 if (first
== 0 || af
< first
)
1187 if (last
== 0 || af
> last
)
1191 assert((first
== 0) == (last
== 0));
1195 /* No entries in the valid range, block everything */
1196 r
= seccomp_rule_add(
1198 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1206 /* Block everything below the first entry */
1207 r
= seccomp_rule_add(
1209 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1212 SCMP_A0(SCMP_CMP_LT
, first
));
1216 /* Block everything above the last entry */
1217 r
= seccomp_rule_add(
1219 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1222 SCMP_A0(SCMP_CMP_GT
, last
));
1226 /* Block everything between the first and last
1228 for (af
= 1; af
< af_max(); af
++) {
1230 if (set_contains(c
->address_families
, INT_TO_PTR(af
)))
1233 r
= seccomp_rule_add(
1235 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1238 SCMP_A0(SCMP_CMP_EQ
, af
));
1247 /* If this is a blacklist, then generate one rule for
1248 * each address family that are then combined in OR
1251 SET_FOREACH(af
, c
->address_families
, i
) {
1253 r
= seccomp_rule_add(
1255 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1258 SCMP_A0(SCMP_CMP_EQ
, PTR_TO_INT(af
)));
1264 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1268 r
= seccomp_load(seccomp
);
1271 seccomp_release(seccomp
);
1275 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1276 scmp_filter_ctx
*seccomp
;
1281 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1284 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1288 r
= seccomp_add_secondary_archs(seccomp
);
1292 r
= seccomp_rule_add(
1294 SCMP_ACT_ERRNO(EPERM
),
1297 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
|PROT_WRITE
, PROT_EXEC
|PROT_WRITE
));
1301 r
= seccomp_rule_add(
1303 SCMP_ACT_ERRNO(EPERM
),
1306 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
, PROT_EXEC
));
1310 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1314 r
= seccomp_load(seccomp
);
1317 seccomp_release(seccomp
);
1321 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1322 static const int permitted_policies
[] = {
1328 scmp_filter_ctx
*seccomp
;
1330 int r
, p
, max_policy
= 0;
1334 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1337 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1341 r
= seccomp_add_secondary_archs(seccomp
);
1345 /* Determine the highest policy constant we want to allow */
1346 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1347 if (permitted_policies
[i
] > max_policy
)
1348 max_policy
= permitted_policies
[i
];
1350 /* Go through all policies with lower values than that, and block them -- unless they appear in the
1352 for (p
= 0; p
< max_policy
; p
++) {
1355 /* Check if this is in the whitelist. */
1356 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1357 if (permitted_policies
[i
] == p
) {
1365 /* Deny this policy */
1366 r
= seccomp_rule_add(
1368 SCMP_ACT_ERRNO(EPERM
),
1369 SCMP_SYS(sched_setscheduler
),
1371 SCMP_A1(SCMP_CMP_EQ
, p
));
1376 /* Blacklist all other policies, i.e. the ones with higher values. Note that all comparisons are unsigned here,
1377 * hence no need no check for < 0 values. */
1378 r
= seccomp_rule_add(
1380 SCMP_ACT_ERRNO(EPERM
),
1381 SCMP_SYS(sched_setscheduler
),
1383 SCMP_A1(SCMP_CMP_GT
, max_policy
));
1387 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1391 r
= seccomp_load(seccomp
);
1394 seccomp_release(seccomp
);
1398 static int apply_protect_sysctl(Unit
*u
, const ExecContext
*c
) {
1399 scmp_filter_ctx
*seccomp
;
1404 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1405 * let's protect even those systems where this is left on in the kernel. */
1407 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1410 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1414 r
= seccomp_add_secondary_archs(seccomp
);
1418 r
= seccomp_rule_add(
1420 SCMP_ACT_ERRNO(EPERM
),
1426 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1430 r
= seccomp_load(seccomp
);
1433 seccomp_release(seccomp
);
1437 static int apply_private_devices(Unit
*u
, const ExecContext
*c
) {
1438 const SystemCallFilterSet
*set
;
1439 scmp_filter_ctx
*seccomp
;
1441 bool syscalls_found
= false;
1446 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1448 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1451 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1455 r
= seccomp_add_secondary_archs(seccomp
);
1459 for (set
= syscall_filter_sets
; set
->set_name
; set
++)
1460 if (streq(set
->set_name
, "@raw-io")) {
1461 syscalls_found
= true;
1465 /* We should never fail here */
1466 if (!syscalls_found
) {
1471 NULSTR_FOREACH(sys
, set
->value
) {
1475 #ifndef __NR_s390_pci_mmio_read
1476 if (streq(sys
, "s390_pci_mmio_read"))
1479 #ifndef __NR_s390_pci_mmio_write
1480 if (streq(sys
, "s390_pci_mmio_write"))
1487 id
= seccomp_syscall_resolve_name(sys
);
1489 r
= seccomp_rule_add(
1491 SCMP_ACT_ERRNO(EPERM
),
1497 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1501 r
= seccomp_load(seccomp
);
1504 seccomp_release(seccomp
);
1510 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1513 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1514 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1516 if (idle_pipe
[0] >= 0) {
1519 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1521 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1524 /* Signal systemd that we are bored and want to continue. */
1525 n
= write(idle_pipe
[3], "x", 1);
1527 /* Wait for systemd to react to the signal above. */
1528 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1531 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1535 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1538 static int build_environment(
1540 const ExecContext
*c
,
1541 const ExecParameters
*p
,
1544 const char *username
,
1546 dev_t journal_stream_dev
,
1547 ino_t journal_stream_ino
,
1550 _cleanup_strv_free_
char **our_env
= NULL
;
1557 our_env
= new0(char*, 13);
1562 _cleanup_free_
char *joined
= NULL
;
1564 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid()) < 0)
1566 our_env
[n_env
++] = x
;
1568 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1570 our_env
[n_env
++] = x
;
1572 joined
= strv_join(p
->fd_names
, ":");
1576 x
= strjoin("LISTEN_FDNAMES=", joined
, NULL
);
1579 our_env
[n_env
++] = x
;
1582 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1583 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid()) < 0)
1585 our_env
[n_env
++] = x
;
1587 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1589 our_env
[n_env
++] = x
;
1592 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1593 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1594 * check the database directly. */
1595 if (unit_has_name(u
, SPECIAL_DBUS_SERVICE
)) {
1596 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1599 our_env
[n_env
++] = x
;
1603 x
= strappend("HOME=", home
);
1606 our_env
[n_env
++] = x
;
1610 x
= strappend("LOGNAME=", username
);
1613 our_env
[n_env
++] = x
;
1615 x
= strappend("USER=", username
);
1618 our_env
[n_env
++] = x
;
1622 x
= strappend("SHELL=", shell
);
1625 our_env
[n_env
++] = x
;
1628 if (exec_context_needs_term(c
)) {
1629 const char *tty_path
, *term
= NULL
;
1631 tty_path
= exec_context_tty_path(c
);
1633 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1634 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1635 * passes to PID 1 ends up all the way in the console login shown. */
1637 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1638 term
= getenv("TERM");
1640 term
= default_term_for_tty(tty_path
);
1642 x
= strappend("TERM=", term
);
1645 our_env
[n_env
++] = x
;
1648 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1649 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1652 our_env
[n_env
++] = x
;
1655 our_env
[n_env
++] = NULL
;
1656 assert(n_env
<= 12);
1664 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1665 _cleanup_strv_free_
char **pass_env
= NULL
;
1666 size_t n_env
= 0, n_bufsize
= 0;
1669 STRV_FOREACH(i
, c
->pass_environment
) {
1670 _cleanup_free_
char *x
= NULL
;
1676 x
= strjoin(*i
, "=", v
, NULL
);
1679 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1681 pass_env
[n_env
++] = x
;
1682 pass_env
[n_env
] = NULL
;
1692 static bool exec_needs_mount_namespace(
1693 const ExecContext
*context
,
1694 const ExecParameters
*params
,
1695 ExecRuntime
*runtime
) {
1700 if (!strv_isempty(context
->read_write_paths
) ||
1701 !strv_isempty(context
->read_only_paths
) ||
1702 !strv_isempty(context
->inaccessible_paths
))
1705 if (context
->mount_flags
!= 0)
1708 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1711 if (context
->private_devices
||
1712 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1713 context
->protect_home
!= PROTECT_HOME_NO
||
1714 context
->protect_kernel_tunables
||
1715 context
->protect_control_groups
)
1721 static int setup_private_users(uid_t uid
, gid_t gid
) {
1722 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1723 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1724 _cleanup_close_
int unshare_ready_fd
= -1;
1725 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1731 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1732 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1733 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1734 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1735 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1736 * continues execution normally. */
1738 if (uid
!= 0 && uid_is_valid(uid
))
1740 "0 0 1\n" /* Map root → root */
1741 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1742 uid
, uid
); /* The case where the above is the same */
1744 uid_map
= strdup("0 0 1\n");
1748 if (gid
!= 0 && gid_is_valid(gid
))
1750 "0 0 1\n" /* Map root → root */
1751 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1754 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1758 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1760 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1761 if (unshare_ready_fd
< 0)
1764 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1766 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1774 _cleanup_close_
int fd
= -1;
1778 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1779 * here, after the parent opened its own user namespace. */
1782 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1784 /* Wait until the parent unshared the user namespace */
1785 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1790 /* Disable the setgroups() system call in the child user namespace, for good. */
1791 a
= procfs_file_alloca(ppid
, "setgroups");
1792 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1794 if (errno
!= ENOENT
) {
1799 /* If the file is missing the kernel is too old, let's continue anyway. */
1801 if (write(fd
, "deny\n", 5) < 0) {
1806 fd
= safe_close(fd
);
1809 /* First write the GID map */
1810 a
= procfs_file_alloca(ppid
, "gid_map");
1811 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1816 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1820 fd
= safe_close(fd
);
1822 /* The write the UID map */
1823 a
= procfs_file_alloca(ppid
, "uid_map");
1824 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1829 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1834 _exit(EXIT_SUCCESS
);
1837 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1838 _exit(EXIT_FAILURE
);
1841 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1843 if (unshare(CLONE_NEWUSER
) < 0)
1846 /* Let the child know that the namespace is ready now */
1847 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1850 /* Try to read an error code from the child */
1851 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1854 if (n
== sizeof(r
)) { /* an error code was sent to us */
1859 if (n
!= 0) /* on success we should have read 0 bytes */
1862 r
= wait_for_terminate(pid
, &si
);
1867 /* If something strange happened with the child, let's consider this fatal, too */
1868 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1874 static int setup_runtime_directory(
1875 const ExecContext
*context
,
1876 const ExecParameters
*params
,
1886 STRV_FOREACH(rt
, context
->runtime_directory
) {
1887 _cleanup_free_
char *p
;
1889 p
= strjoin(params
->runtime_prefix
, "/", *rt
, NULL
);
1893 r
= mkdir_p_label(p
, context
->runtime_directory_mode
);
1897 r
= chmod_and_chown(p
, context
->runtime_directory_mode
, uid
, gid
);
1905 static int setup_smack(
1906 const ExecContext
*context
,
1907 const ExecCommand
*command
) {
1915 if (!mac_smack_use())
1918 if (context
->smack_process_label
) {
1919 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1923 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1925 _cleanup_free_
char *exec_label
= NULL
;
1927 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1928 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1931 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
1941 static int compile_read_write_paths(
1942 const ExecContext
*context
,
1943 const ExecParameters
*params
,
1946 _cleanup_strv_free_
char **l
= NULL
;
1949 /* Compile the list of writable paths. This is the combination of the explicitly configured paths, plus all
1950 * runtime directories. */
1952 if (strv_isempty(context
->read_write_paths
) &&
1953 strv_isempty(context
->runtime_directory
)) {
1954 *ret
= NULL
; /* NOP if neither is set */
1958 l
= strv_copy(context
->read_write_paths
);
1962 STRV_FOREACH(rt
, context
->runtime_directory
) {
1965 s
= strjoin(params
->runtime_prefix
, "/", *rt
, NULL
);
1969 if (strv_consume(&l
, s
) < 0)
1979 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
1987 array
[(*n
)++] = pair
[0];
1989 array
[(*n
)++] = pair
[1];
1992 static int close_remaining_fds(
1993 const ExecParameters
*params
,
1994 ExecRuntime
*runtime
,
1995 DynamicCreds
*dcreds
,
1998 int *fds
, unsigned n_fds
) {
2000 unsigned n_dont_close
= 0;
2001 int dont_close
[n_fds
+ 12];
2005 if (params
->stdin_fd
>= 0)
2006 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2007 if (params
->stdout_fd
>= 0)
2008 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2009 if (params
->stderr_fd
>= 0)
2010 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2013 dont_close
[n_dont_close
++] = socket_fd
;
2015 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2016 n_dont_close
+= n_fds
;
2020 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2024 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2026 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2029 if (user_lookup_fd
>= 0)
2030 dont_close
[n_dont_close
++] = user_lookup_fd
;
2032 return close_all_fds(dont_close
, n_dont_close
);
2035 static bool context_has_address_families(const ExecContext
*c
) {
2038 return c
->address_families_whitelist
||
2039 !set_isempty(c
->address_families
);
2042 static bool context_has_syscall_filters(const ExecContext
*c
) {
2045 return c
->syscall_whitelist
||
2046 !set_isempty(c
->syscall_filter
) ||
2047 !set_isempty(c
->syscall_archs
);
2050 static bool context_has_no_new_privileges(const ExecContext
*c
) {
2053 if (c
->no_new_privileges
)
2056 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
2059 return context_has_address_families(c
) || /* we need NNP if we have any form of seccomp and are unprivileged */
2060 c
->memory_deny_write_execute
||
2061 c
->restrict_realtime
||
2062 c
->protect_kernel_tunables
||
2063 context_has_syscall_filters(c
);
2066 static int send_user_lookup(
2074 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2075 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2078 if (user_lookup_fd
< 0)
2081 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2084 if (writev(user_lookup_fd
,
2086 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2087 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2088 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2094 static int exec_child(
2096 ExecCommand
*command
,
2097 const ExecContext
*context
,
2098 const ExecParameters
*params
,
2099 ExecRuntime
*runtime
,
2100 DynamicCreds
*dcreds
,
2103 int *fds
, unsigned n_fds
,
2108 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2109 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2110 const char *username
= NULL
, *home
= NULL
, *shell
= NULL
, *wd
;
2111 dev_t journal_stream_dev
= 0;
2112 ino_t journal_stream_ino
= 0;
2113 bool needs_mount_namespace
;
2114 uid_t uid
= UID_INVALID
;
2115 gid_t gid
= GID_INVALID
;
2122 assert(exit_status
);
2124 rename_process_from_path(command
->path
);
2126 /* We reset exactly these signals, since they are the
2127 * only ones we set to SIG_IGN in the main daemon. All
2128 * others we leave untouched because we set them to
2129 * SIG_DFL or a valid handler initially, both of which
2130 * will be demoted to SIG_DFL. */
2131 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2132 SIGNALS_IGNORE
, -1);
2134 if (context
->ignore_sigpipe
)
2135 (void) ignore_signals(SIGPIPE
, -1);
2137 r
= reset_signal_mask();
2139 *exit_status
= EXIT_SIGNAL_MASK
;
2143 if (params
->idle_pipe
)
2144 do_idle_pipe_dance(params
->idle_pipe
);
2146 /* Close sockets very early to make sure we don't
2147 * block init reexecution because it cannot bind its
2152 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2154 *exit_status
= EXIT_FDS
;
2158 if (!context
->same_pgrp
)
2160 *exit_status
= EXIT_SETSID
;
2164 exec_context_tty_reset(context
, params
);
2166 if (params
->flags
& EXEC_CONFIRM_SPAWN
) {
2169 r
= ask_for_confirmation(&response
, argv
);
2170 if (r
== -ETIMEDOUT
)
2171 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
2173 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-r
));
2174 else if (response
== 's') {
2175 write_confirm_message("Skipping execution.\n");
2176 *exit_status
= EXIT_CONFIRM
;
2178 } else if (response
== 'n') {
2179 write_confirm_message("Failing execution.\n");
2185 if (context
->dynamic_user
&& dcreds
) {
2187 /* Make sure we bypass our own NSS module for any NSS checks */
2188 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2189 *exit_status
= EXIT_USER
;
2193 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2195 *exit_status
= EXIT_USER
;
2199 if (!uid_is_valid(uid
) || !gid_is_valid(gid
)) {
2200 *exit_status
= EXIT_USER
;
2205 username
= dcreds
->user
->name
;
2208 if (context
->user
) {
2209 username
= context
->user
;
2210 r
= get_user_creds_clean(&username
, &uid
, &gid
, &home
, &shell
);
2212 *exit_status
= EXIT_USER
;
2216 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
2217 * (i.e. are "/" or "/bin/nologin"). */
2220 if (context
->group
) {
2221 const char *g
= context
->group
;
2223 r
= get_group_creds(&g
, &gid
);
2225 *exit_status
= EXIT_GROUP
;
2231 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2233 *exit_status
= EXIT_USER
;
2237 user_lookup_fd
= safe_close(user_lookup_fd
);
2239 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2240 * must sure to drop O_NONBLOCK */
2242 (void) fd_nonblock(socket_fd
, false);
2244 r
= setup_input(context
, params
, socket_fd
);
2246 *exit_status
= EXIT_STDIN
;
2250 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2252 *exit_status
= EXIT_STDOUT
;
2256 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2258 *exit_status
= EXIT_STDERR
;
2262 if (params
->cgroup_path
) {
2263 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2265 *exit_status
= EXIT_CGROUP
;
2270 if (context
->oom_score_adjust_set
) {
2271 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2273 /* When we can't make this change due to EPERM, then
2274 * let's silently skip over it. User namespaces
2275 * prohibit write access to this file, and we
2276 * shouldn't trip up over that. */
2278 sprintf(t
, "%i", context
->oom_score_adjust
);
2279 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2280 if (r
== -EPERM
|| r
== -EACCES
) {
2282 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2285 *exit_status
= EXIT_OOM_ADJUST
;
2290 if (context
->nice_set
)
2291 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2292 *exit_status
= EXIT_NICE
;
2296 if (context
->cpu_sched_set
) {
2297 struct sched_param param
= {
2298 .sched_priority
= context
->cpu_sched_priority
,
2301 r
= sched_setscheduler(0,
2302 context
->cpu_sched_policy
|
2303 (context
->cpu_sched_reset_on_fork
?
2304 SCHED_RESET_ON_FORK
: 0),
2307 *exit_status
= EXIT_SETSCHEDULER
;
2312 if (context
->cpuset
)
2313 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2314 *exit_status
= EXIT_CPUAFFINITY
;
2318 if (context
->ioprio_set
)
2319 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2320 *exit_status
= EXIT_IOPRIO
;
2324 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2325 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2326 *exit_status
= EXIT_TIMERSLACK
;
2330 if (context
->personality
!= PERSONALITY_INVALID
)
2331 if (personality(context
->personality
) < 0) {
2332 *exit_status
= EXIT_PERSONALITY
;
2336 if (context
->utmp_id
)
2337 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0), context
->tty_path
,
2338 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2339 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2341 username
? "root" : context
->user
);
2343 if (context
->user
&& is_terminal_input(context
->std_input
)) {
2344 r
= chown_terminal(STDIN_FILENO
, uid
);
2346 *exit_status
= EXIT_STDIN
;
2351 /* If delegation is enabled we'll pass ownership of the cgroup
2352 * (but only in systemd's own controller hierarchy!) to the
2353 * user of the new process. */
2354 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2355 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2357 *exit_status
= EXIT_CGROUP
;
2362 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2364 *exit_status
= EXIT_CGROUP
;
2369 if (!strv_isempty(context
->runtime_directory
) && params
->runtime_prefix
) {
2370 r
= setup_runtime_directory(context
, params
, uid
, gid
);
2372 *exit_status
= EXIT_RUNTIME_DIRECTORY
;
2377 r
= build_environment(
2389 *exit_status
= EXIT_MEMORY
;
2393 r
= build_pass_environment(context
, &pass_env
);
2395 *exit_status
= EXIT_MEMORY
;
2399 accum_env
= strv_env_merge(5,
2400 params
->environment
,
2403 context
->environment
,
2407 *exit_status
= EXIT_MEMORY
;
2410 accum_env
= strv_env_clean(accum_env
);
2412 (void) umask(context
->umask
);
2414 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2415 r
= setup_smack(context
, command
);
2417 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2421 if (context
->pam_name
&& username
) {
2422 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2424 *exit_status
= EXIT_PAM
;
2430 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2431 r
= setup_netns(runtime
->netns_storage_socket
);
2433 *exit_status
= EXIT_NETWORK
;
2438 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2439 if (needs_mount_namespace
) {
2440 _cleanup_free_
char **rw
= NULL
;
2441 char *tmp
= NULL
, *var
= NULL
;
2443 /* The runtime struct only contains the parent
2444 * of the private /tmp, which is
2445 * non-accessible to world users. Inside of it
2446 * there's a /tmp that is sticky, and that's
2447 * the one we want to use here. */
2449 if (context
->private_tmp
&& runtime
) {
2450 if (runtime
->tmp_dir
)
2451 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2452 if (runtime
->var_tmp_dir
)
2453 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2456 r
= compile_read_write_paths(context
, params
, &rw
);
2458 *exit_status
= EXIT_NAMESPACE
;
2462 r
= setup_namespace(
2463 (params
->flags
& EXEC_APPLY_CHROOT
) ? context
->root_directory
: NULL
,
2465 context
->read_only_paths
,
2466 context
->inaccessible_paths
,
2469 context
->private_devices
,
2470 context
->protect_kernel_tunables
,
2471 context
->protect_control_groups
,
2472 context
->protect_home
,
2473 context
->protect_system
,
2474 context
->mount_flags
);
2476 /* If we couldn't set up the namespace this is
2477 * probably due to a missing capability. In this case,
2478 * silently proceeed. */
2479 if (r
== -EPERM
|| r
== -EACCES
) {
2481 log_unit_debug_errno(unit
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2484 *exit_status
= EXIT_NAMESPACE
;
2489 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2490 r
= enforce_groups(context
, username
, gid
);
2492 *exit_status
= EXIT_GROUP
;
2497 if (context
->working_directory_home
)
2499 else if (context
->working_directory
)
2500 wd
= context
->working_directory
;
2504 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2505 if (!needs_mount_namespace
&& context
->root_directory
)
2506 if (chroot(context
->root_directory
) < 0) {
2507 *exit_status
= EXIT_CHROOT
;
2511 if (chdir(wd
) < 0 &&
2512 !context
->working_directory_missing_ok
) {
2513 *exit_status
= EXIT_CHDIR
;
2519 d
= strjoina(strempty(context
->root_directory
), "/", strempty(wd
));
2521 !context
->working_directory_missing_ok
) {
2522 *exit_status
= EXIT_CHDIR
;
2528 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) &&
2529 mac_selinux_use() &&
2530 params
->selinux_context_net
&&
2532 !command
->privileged
) {
2534 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2536 *exit_status
= EXIT_SELINUX_CONTEXT
;
2542 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2543 r
= setup_private_users(uid
, gid
);
2545 *exit_status
= EXIT_USER
;
2550 /* We repeat the fd closing here, to make sure that
2551 * nothing is leaked from the PAM modules. Note that
2552 * we are more aggressive this time since socket_fd
2553 * and the netns fds we don't need anymore. The custom
2554 * endpoint fd was needed to upload the policy and can
2555 * now be closed as well. */
2556 r
= close_all_fds(fds
, n_fds
);
2558 r
= shift_fds(fds
, n_fds
);
2560 r
= flags_fds(fds
, n_fds
, context
->non_blocking
);
2562 *exit_status
= EXIT_FDS
;
2566 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2568 int secure_bits
= context
->secure_bits
;
2570 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2572 if (!context
->rlimit
[i
])
2575 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2577 *exit_status
= EXIT_LIMITS
;
2582 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2583 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2584 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2585 *exit_status
= EXIT_LIMITS
;
2590 if (!cap_test_all(context
->capability_bounding_set
)) {
2591 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2593 *exit_status
= EXIT_CAPABILITIES
;
2598 /* This is done before enforce_user, but ambient set
2599 * does not survive over setresuid() if keep_caps is not set. */
2600 if (context
->capability_ambient_set
!= 0) {
2601 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2603 *exit_status
= EXIT_CAPABILITIES
;
2608 if (context
->user
) {
2609 r
= enforce_user(context
, uid
);
2611 *exit_status
= EXIT_USER
;
2614 if (context
->capability_ambient_set
!= 0) {
2616 /* Fix the ambient capabilities after user change. */
2617 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2619 *exit_status
= EXIT_CAPABILITIES
;
2623 /* If we were asked to change user and ambient capabilities
2624 * were requested, we had to add keep-caps to the securebits
2625 * so that we would maintain the inherited capability set
2626 * through the setresuid(). Make sure that the bit is added
2627 * also to the context secure_bits so that we don't try to
2628 * drop the bit away next. */
2630 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2634 /* PR_GET_SECUREBITS is not privileged, while
2635 * PR_SET_SECUREBITS is. So to suppress
2636 * potential EPERMs we'll try not to call
2637 * PR_SET_SECUREBITS unless necessary. */
2638 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2639 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2640 *exit_status
= EXIT_SECUREBITS
;
2644 if (context_has_no_new_privileges(context
))
2645 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2646 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2651 if (context_has_address_families(context
)) {
2652 r
= apply_address_families(unit
, context
);
2654 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2659 if (context
->memory_deny_write_execute
) {
2660 r
= apply_memory_deny_write_execute(unit
, context
);
2662 *exit_status
= EXIT_SECCOMP
;
2667 if (context
->restrict_realtime
) {
2668 r
= apply_restrict_realtime(unit
, context
);
2670 *exit_status
= EXIT_SECCOMP
;
2675 if (context
->protect_kernel_tunables
) {
2676 r
= apply_protect_sysctl(unit
, context
);
2678 *exit_status
= EXIT_SECCOMP
;
2683 if (context
->private_devices
) {
2684 r
= apply_private_devices(unit
, context
);
2686 *exit_status
= EXIT_SECCOMP
;
2691 if (context_has_syscall_filters(context
)) {
2692 r
= apply_seccomp(unit
, context
);
2694 *exit_status
= EXIT_SECCOMP
;
2701 if (mac_selinux_use()) {
2702 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2705 r
= setexeccon(exec_context
);
2707 *exit_status
= EXIT_SELINUX_CONTEXT
;
2714 #ifdef HAVE_APPARMOR
2715 if (context
->apparmor_profile
&& mac_apparmor_use()) {
2716 r
= aa_change_onexec(context
->apparmor_profile
);
2717 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2718 *exit_status
= EXIT_APPARMOR_PROFILE
;
2725 final_argv
= replace_env_argv(argv
, accum_env
);
2727 *exit_status
= EXIT_MEMORY
;
2731 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2732 _cleanup_free_
char *line
;
2734 line
= exec_command_line(final_argv
);
2737 log_struct(LOG_DEBUG
,
2739 "EXECUTABLE=%s", command
->path
,
2740 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2746 execve(command
->path
, final_argv
, accum_env
);
2747 *exit_status
= EXIT_EXEC
;
2751 int exec_spawn(Unit
*unit
,
2752 ExecCommand
*command
,
2753 const ExecContext
*context
,
2754 const ExecParameters
*params
,
2755 ExecRuntime
*runtime
,
2756 DynamicCreds
*dcreds
,
2759 _cleanup_strv_free_
char **files_env
= NULL
;
2760 int *fds
= NULL
; unsigned n_fds
= 0;
2761 _cleanup_free_
char *line
= NULL
;
2771 assert(params
->fds
|| params
->n_fds
<= 0);
2773 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2774 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2775 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
2777 if (params
->n_fds
!= 1) {
2778 log_unit_error(unit
, "Got more than one socket.");
2782 socket_fd
= params
->fds
[0];
2786 n_fds
= params
->n_fds
;
2789 r
= exec_context_load_environment(unit
, context
, &files_env
);
2791 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
2793 argv
= params
->argv
?: command
->argv
;
2794 line
= exec_command_line(argv
);
2798 log_struct(LOG_DEBUG
,
2800 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
2801 "EXECUTABLE=%s", command
->path
,
2805 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
2810 r
= exec_child(unit
,
2820 unit
->manager
->user_lookup_fds
[1],
2824 log_struct_errno(LOG_ERR
, r
,
2825 LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED
),
2827 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
2828 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
2830 "EXECUTABLE=%s", command
->path
,
2837 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
2839 /* We add the new process to the cgroup both in the child (so
2840 * that we can be sure that no user code is ever executed
2841 * outside of the cgroup) and in the parent (so that we can be
2842 * sure that when we kill the cgroup the process will be
2844 if (params
->cgroup_path
)
2845 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
2847 exec_status_start(&command
->exec_status
, pid
);
2853 void exec_context_init(ExecContext
*c
) {
2857 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
2858 c
->cpu_sched_policy
= SCHED_OTHER
;
2859 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
2860 c
->syslog_level_prefix
= true;
2861 c
->ignore_sigpipe
= true;
2862 c
->timer_slack_nsec
= NSEC_INFINITY
;
2863 c
->personality
= PERSONALITY_INVALID
;
2864 c
->runtime_directory_mode
= 0755;
2865 c
->capability_bounding_set
= CAP_ALL
;
2868 void exec_context_done(ExecContext
*c
) {
2873 c
->environment
= strv_free(c
->environment
);
2874 c
->environment_files
= strv_free(c
->environment_files
);
2875 c
->pass_environment
= strv_free(c
->pass_environment
);
2877 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
2878 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
2880 c
->working_directory
= mfree(c
->working_directory
);
2881 c
->root_directory
= mfree(c
->root_directory
);
2882 c
->tty_path
= mfree(c
->tty_path
);
2883 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
2884 c
->user
= mfree(c
->user
);
2885 c
->group
= mfree(c
->group
);
2887 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
2889 c
->pam_name
= mfree(c
->pam_name
);
2891 c
->read_only_paths
= strv_free(c
->read_only_paths
);
2892 c
->read_write_paths
= strv_free(c
->read_write_paths
);
2893 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
2896 CPU_FREE(c
->cpuset
);
2898 c
->utmp_id
= mfree(c
->utmp_id
);
2899 c
->selinux_context
= mfree(c
->selinux_context
);
2900 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
2902 c
->syscall_filter
= set_free(c
->syscall_filter
);
2903 c
->syscall_archs
= set_free(c
->syscall_archs
);
2904 c
->address_families
= set_free(c
->address_families
);
2906 c
->runtime_directory
= strv_free(c
->runtime_directory
);
2909 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
2914 if (!runtime_prefix
)
2917 STRV_FOREACH(i
, c
->runtime_directory
) {
2918 _cleanup_free_
char *p
;
2920 p
= strjoin(runtime_prefix
, "/", *i
, NULL
);
2924 /* We execute this synchronously, since we need to be
2925 * sure this is gone when we start the service
2927 (void) rm_rf(p
, REMOVE_ROOT
);
2933 void exec_command_done(ExecCommand
*c
) {
2936 c
->path
= mfree(c
->path
);
2938 c
->argv
= strv_free(c
->argv
);
2941 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
2944 for (i
= 0; i
< n
; i
++)
2945 exec_command_done(c
+i
);
2948 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
2952 LIST_REMOVE(command
, c
, i
);
2953 exec_command_done(i
);
2960 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
2963 for (i
= 0; i
< n
; i
++)
2964 c
[i
] = exec_command_free_list(c
[i
]);
2967 typedef struct InvalidEnvInfo
{
2972 static void invalid_env(const char *p
, void *userdata
) {
2973 InvalidEnvInfo
*info
= userdata
;
2975 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
2978 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
2979 char **i
, **r
= NULL
;
2984 STRV_FOREACH(i
, c
->environment_files
) {
2987 bool ignore
= false;
2989 _cleanup_globfree_ glob_t pglob
= {};
2999 if (!path_is_absolute(fn
)) {
3007 /* Filename supports globbing, take all matching files */
3009 if (glob(fn
, 0, NULL
, &pglob
) != 0) {
3014 return errno
> 0 ? -errno
: -EINVAL
;
3016 count
= pglob
.gl_pathc
;
3024 for (n
= 0; n
< count
; n
++) {
3025 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3033 /* Log invalid environment variables with filename */
3035 InvalidEnvInfo info
= {
3037 .path
= pglob
.gl_pathv
[n
]
3040 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3048 m
= strv_env_merge(2, r
, p
);
3064 static bool tty_may_match_dev_console(const char *tty
) {
3065 _cleanup_free_
char *active
= NULL
;
3071 if (startswith(tty
, "/dev/"))
3074 /* trivial identity? */
3075 if (streq(tty
, "console"))
3078 console
= resolve_dev_console(&active
);
3079 /* if we could not resolve, assume it may */
3083 /* "tty0" means the active VC, so it may be the same sometimes */
3084 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3087 bool exec_context_may_touch_console(ExecContext
*ec
) {
3089 return (ec
->tty_reset
||
3091 ec
->tty_vt_disallocate
||
3092 is_terminal_input(ec
->std_input
) ||
3093 is_terminal_output(ec
->std_output
) ||
3094 is_terminal_output(ec
->std_error
)) &&
3095 tty_may_match_dev_console(exec_context_tty_path(ec
));
3098 static void strv_fprintf(FILE *f
, char **l
) {
3104 fprintf(f
, " %s", *g
);
3107 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3114 prefix
= strempty(prefix
);
3118 "%sWorkingDirectory: %s\n"
3119 "%sRootDirectory: %s\n"
3120 "%sNonBlocking: %s\n"
3121 "%sPrivateTmp: %s\n"
3122 "%sPrivateDevices: %s\n"
3123 "%sProtectKernelTunables: %s\n"
3124 "%sProtectControlGroups: %s\n"
3125 "%sPrivateNetwork: %s\n"
3126 "%sPrivateUsers: %s\n"
3127 "%sProtectHome: %s\n"
3128 "%sProtectSystem: %s\n"
3129 "%sIgnoreSIGPIPE: %s\n"
3130 "%sMemoryDenyWriteExecute: %s\n"
3131 "%sRestrictRealtime: %s\n",
3133 prefix
, c
->working_directory
? c
->working_directory
: "/",
3134 prefix
, c
->root_directory
? c
->root_directory
: "/",
3135 prefix
, yes_no(c
->non_blocking
),
3136 prefix
, yes_no(c
->private_tmp
),
3137 prefix
, yes_no(c
->private_devices
),
3138 prefix
, yes_no(c
->protect_kernel_tunables
),
3139 prefix
, yes_no(c
->protect_control_groups
),
3140 prefix
, yes_no(c
->private_network
),
3141 prefix
, yes_no(c
->private_users
),
3142 prefix
, protect_home_to_string(c
->protect_home
),
3143 prefix
, protect_system_to_string(c
->protect_system
),
3144 prefix
, yes_no(c
->ignore_sigpipe
),
3145 prefix
, yes_no(c
->memory_deny_write_execute
),
3146 prefix
, yes_no(c
->restrict_realtime
));
3148 STRV_FOREACH(e
, c
->environment
)
3149 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3151 STRV_FOREACH(e
, c
->environment_files
)
3152 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3154 STRV_FOREACH(e
, c
->pass_environment
)
3155 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3157 fprintf(f
, "%sRuntimeDirectoryMode: %04o\n", prefix
, c
->runtime_directory_mode
);
3159 STRV_FOREACH(d
, c
->runtime_directory
)
3160 fprintf(f
, "%sRuntimeDirectory: %s\n", prefix
, *d
);
3167 if (c
->oom_score_adjust_set
)
3169 "%sOOMScoreAdjust: %i\n",
3170 prefix
, c
->oom_score_adjust
);
3172 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3174 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3175 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3176 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3177 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3180 if (c
->ioprio_set
) {
3181 _cleanup_free_
char *class_str
= NULL
;
3183 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3185 "%sIOSchedulingClass: %s\n"
3186 "%sIOPriority: %i\n",
3187 prefix
, strna(class_str
),
3188 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
3191 if (c
->cpu_sched_set
) {
3192 _cleanup_free_
char *policy_str
= NULL
;
3194 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3196 "%sCPUSchedulingPolicy: %s\n"
3197 "%sCPUSchedulingPriority: %i\n"
3198 "%sCPUSchedulingResetOnFork: %s\n",
3199 prefix
, strna(policy_str
),
3200 prefix
, c
->cpu_sched_priority
,
3201 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3205 fprintf(f
, "%sCPUAffinity:", prefix
);
3206 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3207 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3208 fprintf(f
, " %u", i
);
3212 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3213 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3216 "%sStandardInput: %s\n"
3217 "%sStandardOutput: %s\n"
3218 "%sStandardError: %s\n",
3219 prefix
, exec_input_to_string(c
->std_input
),
3220 prefix
, exec_output_to_string(c
->std_output
),
3221 prefix
, exec_output_to_string(c
->std_error
));
3227 "%sTTYVHangup: %s\n"
3228 "%sTTYVTDisallocate: %s\n",
3229 prefix
, c
->tty_path
,
3230 prefix
, yes_no(c
->tty_reset
),
3231 prefix
, yes_no(c
->tty_vhangup
),
3232 prefix
, yes_no(c
->tty_vt_disallocate
));
3234 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3235 c
->std_output
== EXEC_OUTPUT_KMSG
||
3236 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3237 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3238 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3239 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3240 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3241 c
->std_error
== EXEC_OUTPUT_KMSG
||
3242 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3243 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3244 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3245 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3247 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3249 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3250 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3253 "%sSyslogFacility: %s\n"
3254 "%sSyslogLevel: %s\n",
3255 prefix
, strna(fac_str
),
3256 prefix
, strna(lvl_str
));
3260 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
3262 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
3263 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
3264 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
3265 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
3266 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
3267 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
3269 if (c
->capability_bounding_set
!= CAP_ALL
) {
3271 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
3273 for (l
= 0; l
<= cap_last_cap(); l
++)
3274 if (c
->capability_bounding_set
& (UINT64_C(1) << l
))
3275 fprintf(f
, " %s", strna(capability_to_name(l
)));
3280 if (c
->capability_ambient_set
!= 0) {
3282 fprintf(f
, "%sAmbientCapabilities:", prefix
);
3284 for (l
= 0; l
<= cap_last_cap(); l
++)
3285 if (c
->capability_ambient_set
& (UINT64_C(1) << l
))
3286 fprintf(f
, " %s", strna(capability_to_name(l
)));
3292 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3294 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3296 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3298 if (strv_length(c
->supplementary_groups
) > 0) {
3299 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3300 strv_fprintf(f
, c
->supplementary_groups
);
3305 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3307 if (strv_length(c
->read_write_paths
) > 0) {
3308 fprintf(f
, "%sReadWritePaths:", prefix
);
3309 strv_fprintf(f
, c
->read_write_paths
);
3313 if (strv_length(c
->read_only_paths
) > 0) {
3314 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3315 strv_fprintf(f
, c
->read_only_paths
);
3319 if (strv_length(c
->inaccessible_paths
) > 0) {
3320 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3321 strv_fprintf(f
, c
->inaccessible_paths
);
3327 "%sUtmpIdentifier: %s\n",
3328 prefix
, c
->utmp_id
);
3330 if (c
->selinux_context
)
3332 "%sSELinuxContext: %s%s\n",
3333 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3335 if (c
->personality
!= PERSONALITY_INVALID
)
3337 "%sPersonality: %s\n",
3338 prefix
, strna(personality_to_string(c
->personality
)));
3340 if (c
->syscall_filter
) {
3348 "%sSystemCallFilter: ",
3351 if (!c
->syscall_whitelist
)
3355 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3356 _cleanup_free_
char *name
= NULL
;
3363 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3364 fputs(strna(name
), f
);
3371 if (c
->syscall_archs
) {
3378 "%sSystemCallArchitectures:",
3382 SET_FOREACH(id
, c
->syscall_archs
, j
)
3383 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3388 if (c
->syscall_errno
> 0)
3390 "%sSystemCallErrorNumber: %s\n",
3391 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3393 if (c
->apparmor_profile
)
3395 "%sAppArmorProfile: %s%s\n",
3396 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3399 bool exec_context_maintains_privileges(ExecContext
*c
) {
3402 /* Returns true if the process forked off would run under
3403 * an unchanged UID or as root. */
3408 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3414 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3419 dual_timestamp_get(&s
->start_timestamp
);
3422 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3425 if (s
->pid
&& s
->pid
!= pid
)
3429 dual_timestamp_get(&s
->exit_timestamp
);
3435 if (context
->utmp_id
)
3436 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3438 exec_context_tty_reset(context
, NULL
);
3442 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3443 char buf
[FORMAT_TIMESTAMP_MAX
];
3451 prefix
= strempty(prefix
);
3454 "%sPID: "PID_FMT
"\n",
3457 if (dual_timestamp_is_set(&s
->start_timestamp
))
3459 "%sStart Timestamp: %s\n",
3460 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3462 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3464 "%sExit Timestamp: %s\n"
3466 "%sExit Status: %i\n",
3467 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3468 prefix
, sigchld_code_to_string(s
->code
),
3472 char *exec_command_line(char **argv
) {
3480 STRV_FOREACH(a
, argv
)
3483 if (!(n
= new(char, k
)))
3487 STRV_FOREACH(a
, argv
) {
3494 if (strpbrk(*a
, WHITESPACE
)) {
3505 /* FIXME: this doesn't really handle arguments that have
3506 * spaces and ticks in them */
3511 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3512 _cleanup_free_
char *cmd
= NULL
;
3513 const char *prefix2
;
3518 prefix
= strempty(prefix
);
3519 prefix2
= strjoina(prefix
, "\t");
3521 cmd
= exec_command_line(c
->argv
);
3523 "%sCommand Line: %s\n",
3524 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3526 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3529 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3532 prefix
= strempty(prefix
);
3534 LIST_FOREACH(command
, c
, c
)
3535 exec_command_dump(c
, f
, prefix
);
3538 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3545 /* It's kind of important, that we keep the order here */
3546 LIST_FIND_TAIL(command
, *l
, end
);
3547 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3552 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3560 l
= strv_new_ap(path
, ap
);
3581 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3582 _cleanup_strv_free_
char **l
= NULL
;
3590 l
= strv_new_ap(path
, ap
);
3596 r
= strv_extend_strv(&c
->argv
, l
, false);
3604 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3609 *rt
= new0(ExecRuntime
, 1);
3614 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3619 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
3629 if (!c
->private_network
&& !c
->private_tmp
)
3632 r
= exec_runtime_allocate(rt
);
3636 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
3637 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
3641 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
3642 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
3650 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
3652 assert(r
->n_ref
> 0);
3658 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
3663 assert(r
->n_ref
> 0);
3670 free(r
->var_tmp_dir
);
3671 safe_close_pair(r
->netns_storage_socket
);
3677 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
3686 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
3688 if (rt
->var_tmp_dir
)
3689 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
3691 if (rt
->netns_storage_socket
[0] >= 0) {
3694 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
3698 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
3701 if (rt
->netns_storage_socket
[1] >= 0) {
3704 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
3708 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
3714 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
3721 if (streq(key
, "tmp-dir")) {
3724 r
= exec_runtime_allocate(rt
);
3728 copy
= strdup(value
);
3732 free((*rt
)->tmp_dir
);
3733 (*rt
)->tmp_dir
= copy
;
3735 } else if (streq(key
, "var-tmp-dir")) {
3738 r
= exec_runtime_allocate(rt
);
3742 copy
= strdup(value
);
3746 free((*rt
)->var_tmp_dir
);
3747 (*rt
)->var_tmp_dir
= copy
;
3749 } else if (streq(key
, "netns-socket-0")) {
3752 r
= exec_runtime_allocate(rt
);
3756 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
3757 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
3759 safe_close((*rt
)->netns_storage_socket
[0]);
3760 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
3762 } else if (streq(key
, "netns-socket-1")) {
3765 r
= exec_runtime_allocate(rt
);
3769 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
3770 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
3772 safe_close((*rt
)->netns_storage_socket
[1]);
3773 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
3781 static void *remove_tmpdir_thread(void *p
) {
3782 _cleanup_free_
char *path
= p
;
3784 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
3788 void exec_runtime_destroy(ExecRuntime
*rt
) {
3794 /* If there are multiple users of this, let's leave the stuff around */
3799 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
3801 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
3803 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
3810 if (rt
->var_tmp_dir
) {
3811 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
3813 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
3815 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
3816 free(rt
->var_tmp_dir
);
3819 rt
->var_tmp_dir
= NULL
;
3822 safe_close_pair(rt
->netns_storage_socket
);
3825 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
3826 [EXEC_INPUT_NULL
] = "null",
3827 [EXEC_INPUT_TTY
] = "tty",
3828 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
3829 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
3830 [EXEC_INPUT_SOCKET
] = "socket"
3833 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
3835 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
3836 [EXEC_OUTPUT_INHERIT
] = "inherit",
3837 [EXEC_OUTPUT_NULL
] = "null",
3838 [EXEC_OUTPUT_TTY
] = "tty",
3839 [EXEC_OUTPUT_SYSLOG
] = "syslog",
3840 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
3841 [EXEC_OUTPUT_KMSG
] = "kmsg",
3842 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
3843 [EXEC_OUTPUT_JOURNAL
] = "journal",
3844 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
3845 [EXEC_OUTPUT_SOCKET
] = "socket"
3848 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
3850 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
3851 [EXEC_UTMP_INIT
] = "init",
3852 [EXEC_UTMP_LOGIN
] = "login",
3853 [EXEC_UTMP_USER
] = "user",
3856 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);