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>
33 #include <sys/socket.h>
35 #include <sys/types.h>
41 #include <security/pam_appl.h>
45 #include <selinux/selinux.h>
53 #include <sys/apparmor.h>
56 #include "sd-messages.h"
59 #include "alloc-util.h"
61 #include "apparmor-util.h"
66 #include "capability-util.h"
69 #include "errno-list.h"
71 #include "exit-status.h"
74 #include "format-util.h"
76 #include "glob-util.h"
83 #include "namespace.h"
84 #include "parse-util.h"
85 #include "path-util.h"
86 #include "process-util.h"
87 #include "rlimit-util.h"
90 #include "seccomp-util.h"
92 #include "securebits.h"
93 #include "selinux-util.h"
94 #include "signal-util.h"
95 #include "smack-util.h"
97 #include "string-table.h"
98 #include "string-util.h"
100 #include "syslog-util.h"
101 #include "terminal-util.h"
103 #include "user-util.h"
105 #include "utmp-wtmp.h"
107 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
108 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
110 /* This assumes there is a 'tty' group */
111 #define TTY_MODE 0620
113 #define SNDBUF_SIZE (8*1024*1024)
115 static int shift_fds(int fds
[], unsigned n_fds
) {
116 int start
, restart_from
;
121 /* Modifies the fds array! (sorts it) */
131 for (i
= start
; i
< (int) n_fds
; i
++) {
134 /* Already at right index? */
138 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
145 /* Hmm, the fd we wanted isn't free? Then
146 * let's remember that and try again from here */
147 if (nfd
!= i
+3 && restart_from
< 0)
151 if (restart_from
< 0)
154 start
= restart_from
;
160 static int flags_fds(const int fds
[], unsigned n_storage_fds
, unsigned n_socket_fds
, bool nonblock
) {
164 n_fds
= n_storage_fds
+ n_socket_fds
;
170 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
171 * O_NONBLOCK only applies to socket activation though. */
173 for (i
= 0; i
< n_fds
; i
++) {
175 if (i
< n_socket_fds
) {
176 r
= fd_nonblock(fds
[i
], nonblock
);
181 /* We unconditionally drop FD_CLOEXEC from the fds,
182 * since after all we want to pass these fds to our
185 r
= fd_cloexec(fds
[i
], false);
193 static const char *exec_context_tty_path(const ExecContext
*context
) {
196 if (context
->stdio_as_fds
)
199 if (context
->tty_path
)
200 return context
->tty_path
;
202 return "/dev/console";
205 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
210 path
= exec_context_tty_path(context
);
212 if (context
->tty_vhangup
) {
213 if (p
&& p
->stdin_fd
>= 0)
214 (void) terminal_vhangup_fd(p
->stdin_fd
);
216 (void) terminal_vhangup(path
);
219 if (context
->tty_reset
) {
220 if (p
&& p
->stdin_fd
>= 0)
221 (void) reset_terminal_fd(p
->stdin_fd
, true);
223 (void) reset_terminal(path
);
226 if (context
->tty_vt_disallocate
&& path
)
227 (void) vt_disallocate(path
);
230 static bool is_terminal_input(ExecInput i
) {
233 EXEC_INPUT_TTY_FORCE
,
234 EXEC_INPUT_TTY_FAIL
);
237 static bool is_terminal_output(ExecOutput o
) {
240 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
241 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
242 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
245 static bool exec_context_needs_term(const ExecContext
*c
) {
248 /* Return true if the execution context suggests we should set $TERM to something useful. */
250 if (is_terminal_input(c
->std_input
))
253 if (is_terminal_output(c
->std_output
))
256 if (is_terminal_output(c
->std_error
))
259 return !!c
->tty_path
;
262 static int open_null_as(int flags
, int nfd
) {
267 fd
= open("/dev/null", flags
|O_NOCTTY
);
272 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
280 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
281 union sockaddr_union sa
= {
282 .un
.sun_family
= AF_UNIX
,
283 .un
.sun_path
= "/run/systemd/journal/stdout",
285 uid_t olduid
= UID_INVALID
;
286 gid_t oldgid
= GID_INVALID
;
289 if (gid
!= GID_INVALID
) {
297 if (uid
!= UID_INVALID
) {
307 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
311 /* If we fail to restore the uid or gid, things will likely
312 fail later on. This should only happen if an LSM interferes. */
314 if (uid
!= UID_INVALID
)
315 (void) seteuid(olduid
);
318 if (gid
!= GID_INVALID
)
319 (void) setegid(oldgid
);
324 static int connect_logger_as(
326 const ExecContext
*context
,
336 assert(output
< _EXEC_OUTPUT_MAX
);
340 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
344 r
= connect_journal_socket(fd
, uid
, gid
);
348 if (shutdown(fd
, SHUT_RD
) < 0) {
353 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
363 context
->syslog_identifier
? context
->syslog_identifier
: ident
,
365 context
->syslog_priority
,
366 !!context
->syslog_level_prefix
,
367 output
== EXEC_OUTPUT_SYSLOG
|| output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
368 output
== EXEC_OUTPUT_KMSG
|| output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
,
369 is_terminal_output(output
));
374 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
379 static int open_terminal_as(const char *path
, mode_t mode
, int nfd
) {
385 fd
= open_terminal(path
, mode
| O_NOCTTY
);
390 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
398 static int fixup_input(ExecInput std_input
, int socket_fd
, bool apply_tty_stdin
) {
400 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
401 return EXEC_INPUT_NULL
;
403 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
404 return EXEC_INPUT_NULL
;
409 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
411 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
412 return EXEC_OUTPUT_INHERIT
;
417 static int setup_input(
418 const ExecContext
*context
,
419 const ExecParameters
*params
,
421 int named_iofds
[3]) {
428 if (params
->stdin_fd
>= 0) {
429 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
432 /* Try to make this the controlling tty, if it is a tty, and reset it */
433 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
434 (void) reset_terminal_fd(STDIN_FILENO
, true);
439 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
443 case EXEC_INPUT_NULL
:
444 return open_null_as(O_RDONLY
, STDIN_FILENO
);
447 case EXEC_INPUT_TTY_FORCE
:
448 case EXEC_INPUT_TTY_FAIL
: {
451 fd
= acquire_terminal(exec_context_tty_path(context
),
452 i
== EXEC_INPUT_TTY_FAIL
,
453 i
== EXEC_INPUT_TTY_FORCE
,
459 if (fd
!= STDIN_FILENO
) {
460 r
= dup2(fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
468 case EXEC_INPUT_SOCKET
:
469 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
471 case EXEC_INPUT_NAMED_FD
:
472 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
473 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
476 assert_not_reached("Unknown input type");
480 static int setup_output(
482 const ExecContext
*context
,
483 const ExecParameters
*params
,
490 dev_t
*journal_stream_dev
,
491 ino_t
*journal_stream_ino
) {
501 assert(journal_stream_dev
);
502 assert(journal_stream_ino
);
504 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
506 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
509 return STDOUT_FILENO
;
512 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
513 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
516 return STDERR_FILENO
;
519 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
520 o
= fixup_output(context
->std_output
, socket_fd
);
522 if (fileno
== STDERR_FILENO
) {
524 e
= fixup_output(context
->std_error
, socket_fd
);
526 /* This expects the input and output are already set up */
528 /* Don't change the stderr file descriptor if we inherit all
529 * the way and are not on a tty */
530 if (e
== EXEC_OUTPUT_INHERIT
&&
531 o
== EXEC_OUTPUT_INHERIT
&&
532 i
== EXEC_INPUT_NULL
&&
533 !is_terminal_input(context
->std_input
) &&
537 /* Duplicate from stdout if possible */
538 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
539 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
543 } else if (o
== EXEC_OUTPUT_INHERIT
) {
544 /* If input got downgraded, inherit the original value */
545 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
546 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
548 /* If the input is connected to anything that's not a /dev/null, inherit that... */
549 if (i
!= EXEC_INPUT_NULL
)
550 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
552 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
556 /* We need to open /dev/null here anew, to get the right access mode. */
557 return open_null_as(O_WRONLY
, fileno
);
562 case EXEC_OUTPUT_NULL
:
563 return open_null_as(O_WRONLY
, fileno
);
565 case EXEC_OUTPUT_TTY
:
566 if (is_terminal_input(i
))
567 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
569 /* We don't reset the terminal if this is just about output */
570 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
572 case EXEC_OUTPUT_SYSLOG
:
573 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
574 case EXEC_OUTPUT_KMSG
:
575 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
576 case EXEC_OUTPUT_JOURNAL
:
577 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
578 r
= connect_logger_as(unit
, context
, o
, ident
, fileno
, uid
, gid
);
580 log_unit_error_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
581 r
= open_null_as(O_WRONLY
, fileno
);
585 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
586 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
587 * services to detect whether they are connected to the journal or not. */
589 if (fstat(fileno
, &st
) >= 0) {
590 *journal_stream_dev
= st
.st_dev
;
591 *journal_stream_ino
= st
.st_ino
;
596 case EXEC_OUTPUT_SOCKET
:
597 assert(socket_fd
>= 0);
598 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
600 case EXEC_OUTPUT_NAMED_FD
:
601 (void) fd_nonblock(named_iofds
[fileno
], false);
602 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
605 assert_not_reached("Unknown error type");
609 static int chown_terminal(int fd
, uid_t uid
) {
614 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
618 /* This might fail. What matters are the results. */
619 (void) fchown(fd
, uid
, -1);
620 (void) fchmod(fd
, TTY_MODE
);
622 if (fstat(fd
, &st
) < 0)
625 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
631 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
632 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
635 assert(_saved_stdin
);
636 assert(_saved_stdout
);
638 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
642 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
643 if (saved_stdout
< 0)
646 fd
= acquire_terminal(vc
, false, false, false, DEFAULT_CONFIRM_USEC
);
650 r
= chown_terminal(fd
, getuid());
654 r
= reset_terminal_fd(fd
, true);
658 if (dup2(fd
, STDIN_FILENO
) < 0)
661 if (dup2(fd
, STDOUT_FILENO
) < 0)
668 *_saved_stdin
= saved_stdin
;
669 *_saved_stdout
= saved_stdout
;
671 saved_stdin
= saved_stdout
= -1;
676 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
679 if (err
== -ETIMEDOUT
)
680 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
683 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
687 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
688 _cleanup_close_
int fd
= -1;
692 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
696 write_confirm_error_fd(err
, fd
, u
);
699 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
703 assert(saved_stdout
);
707 if (*saved_stdin
>= 0)
708 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
711 if (*saved_stdout
>= 0)
712 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
715 *saved_stdin
= safe_close(*saved_stdin
);
716 *saved_stdout
= safe_close(*saved_stdout
);
722 CONFIRM_PRETEND_FAILURE
= -1,
723 CONFIRM_PRETEND_SUCCESS
= 0,
727 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
728 int saved_stdout
= -1, saved_stdin
= -1, r
;
729 _cleanup_free_
char *e
= NULL
;
732 /* For any internal errors, assume a positive response. */
733 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
735 write_confirm_error(r
, vc
, u
);
736 return CONFIRM_EXECUTE
;
739 /* confirm_spawn might have been disabled while we were sleeping. */
740 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
745 e
= ellipsize(cmdline
, 60, 100);
753 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
755 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
762 printf("Resuming normal execution.\n");
763 manager_disable_confirm_spawn();
767 unit_dump(u
, stdout
, " ");
768 continue; /* ask again */
770 printf("Failing execution.\n");
771 r
= CONFIRM_PRETEND_FAILURE
;
774 printf(" c - continue, proceed without asking anymore\n"
775 " D - dump, show the state of the unit\n"
776 " f - fail, don't execute the command and pretend it failed\n"
778 " i - info, show a short summary of the unit\n"
779 " j - jobs, show jobs that are in progress\n"
780 " s - skip, don't execute the command and pretend it succeeded\n"
781 " y - yes, execute the command\n");
782 continue; /* ask again */
784 printf(" Description: %s\n"
787 u
->id
, u
->description
, cmdline
);
788 continue; /* ask again */
790 manager_dump_jobs(u
->manager
, stdout
, " ");
791 continue; /* ask again */
793 /* 'n' was removed in favor of 'f'. */
794 printf("Didn't understand 'n', did you mean 'f'?\n");
795 continue; /* ask again */
797 printf("Skipping execution.\n");
798 r
= CONFIRM_PRETEND_SUCCESS
;
804 assert_not_reached("Unhandled choice");
810 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
814 static int get_fixed_user(const ExecContext
*c
, const char **user
,
815 uid_t
*uid
, gid_t
*gid
,
816 const char **home
, const char **shell
) {
825 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
826 * (i.e. are "/" or "/bin/nologin"). */
829 r
= get_user_creds_clean(&name
, uid
, gid
, home
, shell
);
837 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
847 r
= get_group_creds(&name
, gid
);
855 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
856 const char *group
, gid_t gid
,
857 gid_t
**supplementary_gids
, int *ngids
) {
861 bool keep_groups
= false;
862 gid_t
*groups
= NULL
;
863 _cleanup_free_ gid_t
*l_gids
= NULL
;
868 * If user is given, then lookup GID and supplementary groups list.
869 * We avoid NSS lookups for gid=0. Also we have to initialize groups
870 * here and as early as possible so we keep the list of supplementary
871 * groups of the caller.
873 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
874 /* First step, initialize groups from /etc/groups */
875 if (initgroups(user
, gid
) < 0)
881 if (!c
->supplementary_groups
)
885 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
886 * be positive, otherwise fail.
889 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
890 if (ngroups_max
<= 0) {
894 return -EOPNOTSUPP
; /* For all other values */
897 l_gids
= new(gid_t
, ngroups_max
);
903 * Lookup the list of groups that the user belongs to, we
904 * avoid NSS lookups here too for gid=0.
907 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
912 STRV_FOREACH(i
, c
->supplementary_groups
) {
915 if (k
>= ngroups_max
)
919 r
= get_group_creds(&g
, l_gids
+k
);
927 * Sets ngids to zero to drop all supplementary groups, happens
928 * when we are under root and SupplementaryGroups= is empty.
935 /* Otherwise get the final list of supplementary groups */
936 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
940 *supplementary_gids
= groups
;
948 static int enforce_groups(const ExecContext
*context
, gid_t gid
,
949 gid_t
*supplementary_gids
, int ngids
) {
954 /* Handle SupplementaryGroups= even if it is empty */
955 if (context
->supplementary_groups
) {
956 r
= maybe_setgroups(ngids
, supplementary_gids
);
961 if (gid_is_valid(gid
)) {
962 /* Then set our gids */
963 if (setresgid(gid
, gid
, gid
) < 0)
970 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
973 if (!uid_is_valid(uid
))
976 /* Sets (but doesn't look up) the uid and make sure we keep the
977 * capabilities while doing so. */
979 if (context
->capability_ambient_set
!= 0) {
981 /* First step: If we need to keep capabilities but
982 * drop privileges we need to make sure we keep our
983 * caps, while we drop privileges. */
985 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
987 if (prctl(PR_GET_SECUREBITS
) != sb
)
988 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
993 /* Second step: actually set the uids */
994 if (setresuid(uid
, uid
, uid
) < 0)
997 /* At this point we should have all necessary capabilities but
998 are otherwise a normal user. However, the caps might got
999 corrupted due to the setresuid() so we need clean them up
1000 later. This is done outside of this call. */
1007 static int null_conv(
1009 const struct pam_message
**msg
,
1010 struct pam_response
**resp
,
1011 void *appdata_ptr
) {
1013 /* We don't support conversations */
1015 return PAM_CONV_ERR
;
1020 static int setup_pam(
1027 int fds
[], unsigned n_fds
) {
1031 static const struct pam_conv conv
= {
1036 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1037 pam_handle_t
*handle
= NULL
;
1039 int pam_code
= PAM_SUCCESS
, r
;
1040 char **nv
, **e
= NULL
;
1041 bool close_session
= false;
1042 pid_t pam_pid
= 0, parent_pid
;
1049 /* We set up PAM in the parent process, then fork. The child
1050 * will then stay around until killed via PR_GET_PDEATHSIG or
1051 * systemd via the cgroup logic. It will then remove the PAM
1052 * session again. The parent process will exec() the actual
1053 * daemon. We do things this way to ensure that the main PID
1054 * of the daemon is the one we initially fork()ed. */
1056 r
= barrier_create(&barrier
);
1060 if (log_get_max_level() < LOG_DEBUG
)
1061 flags
|= PAM_SILENT
;
1063 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1064 if (pam_code
!= PAM_SUCCESS
) {
1070 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1071 if (pam_code
!= PAM_SUCCESS
)
1075 STRV_FOREACH(nv
, *env
) {
1076 pam_code
= pam_putenv(handle
, *nv
);
1077 if (pam_code
!= PAM_SUCCESS
)
1081 pam_code
= pam_acct_mgmt(handle
, flags
);
1082 if (pam_code
!= PAM_SUCCESS
)
1085 pam_code
= pam_open_session(handle
, flags
);
1086 if (pam_code
!= PAM_SUCCESS
)
1089 close_session
= true;
1091 e
= pam_getenvlist(handle
);
1093 pam_code
= PAM_BUF_ERR
;
1097 /* Block SIGTERM, so that we know that it won't get lost in
1100 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1102 parent_pid
= getpid();
1111 int sig
, ret
= EXIT_PAM
;
1113 /* The child's job is to reset the PAM session on
1115 barrier_set_role(&barrier
, BARRIER_CHILD
);
1117 /* This string must fit in 10 chars (i.e. the length
1118 * of "/sbin/init"), to look pretty in /bin/ps */
1119 rename_process("(sd-pam)");
1121 /* Make sure we don't keep open the passed fds in this
1122 child. We assume that otherwise only those fds are
1123 open here that have been opened by PAM. */
1124 close_many(fds
, n_fds
);
1126 /* Drop privileges - we don't need any to pam_close_session
1127 * and this will make PR_SET_PDEATHSIG work in most cases.
1128 * If this fails, ignore the error - but expect sd-pam threads
1129 * to fail to exit normally */
1131 r
= maybe_setgroups(0, NULL
);
1133 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1134 if (setresgid(gid
, gid
, gid
) < 0)
1135 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1136 if (setresuid(uid
, uid
, uid
) < 0)
1137 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1139 (void) ignore_signals(SIGPIPE
, -1);
1141 /* Wait until our parent died. This will only work if
1142 * the above setresuid() succeeds, otherwise the kernel
1143 * will not allow unprivileged parents kill their privileged
1144 * children this way. We rely on the control groups kill logic
1145 * to do the rest for us. */
1146 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1149 /* Tell the parent that our setup is done. This is especially
1150 * important regarding dropping privileges. Otherwise, unit
1151 * setup might race against our setresuid(2) call.
1153 * If the parent aborted, we'll detect this below, hence ignore
1154 * return failure here. */
1155 (void) barrier_place(&barrier
);
1157 /* Check if our parent process might already have died? */
1158 if (getppid() == parent_pid
) {
1161 assert_se(sigemptyset(&ss
) >= 0);
1162 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1165 if (sigwait(&ss
, &sig
) < 0) {
1172 assert(sig
== SIGTERM
);
1177 /* If our parent died we'll end the session */
1178 if (getppid() != parent_pid
) {
1179 pam_code
= pam_close_session(handle
, flags
);
1180 if (pam_code
!= PAM_SUCCESS
)
1187 pam_end(handle
, pam_code
| flags
);
1191 barrier_set_role(&barrier
, BARRIER_PARENT
);
1193 /* If the child was forked off successfully it will do all the
1194 * cleanups, so forget about the handle here. */
1197 /* Unblock SIGTERM again in the parent */
1198 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1200 /* We close the log explicitly here, since the PAM modules
1201 * might have opened it, but we don't want this fd around. */
1204 /* Synchronously wait for the child to initialize. We don't care for
1205 * errors as we cannot recover. However, warn loudly if it happens. */
1206 if (!barrier_place_and_sync(&barrier
))
1207 log_error("PAM initialization failed");
1215 if (pam_code
!= PAM_SUCCESS
) {
1216 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1217 r
= -EPERM
; /* PAM errors do not map to errno */
1219 log_error_errno(r
, "PAM failed: %m");
1223 pam_code
= pam_close_session(handle
, flags
);
1225 pam_end(handle
, pam_code
| flags
);
1237 static void rename_process_from_path(const char *path
) {
1238 char process_name
[11];
1242 /* This resulting string must fit in 10 chars (i.e. the length
1243 * of "/sbin/init") to look pretty in /bin/ps */
1247 rename_process("(...)");
1253 /* The end of the process name is usually more
1254 * interesting, since the first bit might just be
1260 process_name
[0] = '(';
1261 memcpy(process_name
+1, p
, l
);
1262 process_name
[1+l
] = ')';
1263 process_name
[1+l
+1] = 0;
1265 rename_process(process_name
);
1268 static bool context_has_address_families(const ExecContext
*c
) {
1271 return c
->address_families_whitelist
||
1272 !set_isempty(c
->address_families
);
1275 static bool context_has_syscall_filters(const ExecContext
*c
) {
1278 return c
->syscall_whitelist
||
1279 !set_isempty(c
->syscall_filter
);
1282 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1285 if (c
->no_new_privileges
)
1288 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1291 /* We need NNP if we have any form of seccomp and are unprivileged */
1292 return context_has_address_families(c
) ||
1293 c
->memory_deny_write_execute
||
1294 c
->restrict_realtime
||
1295 exec_context_restrict_namespaces_set(c
) ||
1296 c
->protect_kernel_tunables
||
1297 c
->protect_kernel_modules
||
1298 c
->private_devices
||
1299 context_has_syscall_filters(c
) ||
1300 !set_isempty(c
->syscall_archs
);
1305 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1307 if (is_seccomp_available())
1311 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1316 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
) {
1317 uint32_t negative_action
, default_action
, action
;
1322 if (!context_has_syscall_filters(c
))
1325 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1328 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1330 if (c
->syscall_whitelist
) {
1331 default_action
= negative_action
;
1332 action
= SCMP_ACT_ALLOW
;
1334 default_action
= SCMP_ACT_ALLOW
;
1335 action
= negative_action
;
1338 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
);
1341 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1345 if (set_isempty(c
->syscall_archs
))
1348 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1351 return seccomp_restrict_archs(c
->syscall_archs
);
1354 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1358 if (!context_has_address_families(c
))
1361 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1364 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1367 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1371 if (!c
->memory_deny_write_execute
)
1374 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1377 return seccomp_memory_deny_write_execute();
1380 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1384 if (!c
->restrict_realtime
)
1387 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1390 return seccomp_restrict_realtime();
1393 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1397 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1398 * let's protect even those systems where this is left on in the kernel. */
1400 if (!c
->protect_kernel_tunables
)
1403 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1406 return seccomp_protect_sysctl();
1409 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1413 /* Turn off module syscalls on ProtectKernelModules=yes */
1415 if (!c
->protect_kernel_modules
)
1418 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1421 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
));
1424 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1428 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1430 if (!c
->private_devices
)
1433 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1436 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
));
1439 static int apply_restrict_namespaces(Unit
*u
, const ExecContext
*c
) {
1443 if (!exec_context_restrict_namespaces_set(c
))
1446 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1449 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1454 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1457 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1458 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1460 if (idle_pipe
[0] >= 0) {
1463 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1465 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1468 /* Signal systemd that we are bored and want to continue. */
1469 n
= write(idle_pipe
[3], "x", 1);
1471 /* Wait for systemd to react to the signal above. */
1472 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1475 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1479 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1482 static int build_environment(
1484 const ExecContext
*c
,
1485 const ExecParameters
*p
,
1488 const char *username
,
1490 dev_t journal_stream_dev
,
1491 ino_t journal_stream_ino
,
1494 _cleanup_strv_free_
char **our_env
= NULL
;
1502 our_env
= new0(char*, 14);
1507 _cleanup_free_
char *joined
= NULL
;
1509 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid()) < 0)
1511 our_env
[n_env
++] = x
;
1513 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1515 our_env
[n_env
++] = x
;
1517 joined
= strv_join(p
->fd_names
, ":");
1521 x
= strjoin("LISTEN_FDNAMES=", joined
);
1524 our_env
[n_env
++] = x
;
1527 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1528 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid()) < 0)
1530 our_env
[n_env
++] = x
;
1532 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1534 our_env
[n_env
++] = x
;
1537 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1538 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1539 * check the database directly. */
1540 if (unit_has_name(u
, SPECIAL_DBUS_SERVICE
)) {
1541 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1544 our_env
[n_env
++] = x
;
1548 x
= strappend("HOME=", home
);
1551 our_env
[n_env
++] = x
;
1555 x
= strappend("LOGNAME=", username
);
1558 our_env
[n_env
++] = x
;
1560 x
= strappend("USER=", username
);
1563 our_env
[n_env
++] = x
;
1567 x
= strappend("SHELL=", shell
);
1570 our_env
[n_env
++] = x
;
1573 if (!sd_id128_is_null(u
->invocation_id
)) {
1574 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1577 our_env
[n_env
++] = x
;
1580 if (exec_context_needs_term(c
)) {
1581 const char *tty_path
, *term
= NULL
;
1583 tty_path
= exec_context_tty_path(c
);
1585 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1586 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1587 * passes to PID 1 ends up all the way in the console login shown. */
1589 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1590 term
= getenv("TERM");
1592 term
= default_term_for_tty(tty_path
);
1594 x
= strappend("TERM=", term
);
1597 our_env
[n_env
++] = x
;
1600 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1601 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1604 our_env
[n_env
++] = x
;
1607 our_env
[n_env
++] = NULL
;
1608 assert(n_env
<= 12);
1616 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1617 _cleanup_strv_free_
char **pass_env
= NULL
;
1618 size_t n_env
= 0, n_bufsize
= 0;
1621 STRV_FOREACH(i
, c
->pass_environment
) {
1622 _cleanup_free_
char *x
= NULL
;
1628 x
= strjoin(*i
, "=", v
);
1631 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1633 pass_env
[n_env
++] = x
;
1634 pass_env
[n_env
] = NULL
;
1644 static bool exec_needs_mount_namespace(
1645 const ExecContext
*context
,
1646 const ExecParameters
*params
,
1647 ExecRuntime
*runtime
) {
1652 if (context
->root_image
)
1655 if (!strv_isempty(context
->read_write_paths
) ||
1656 !strv_isempty(context
->read_only_paths
) ||
1657 !strv_isempty(context
->inaccessible_paths
))
1660 if (context
->n_bind_mounts
> 0)
1663 if (context
->mount_flags
!= 0)
1666 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1669 if (context
->private_devices
||
1670 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1671 context
->protect_home
!= PROTECT_HOME_NO
||
1672 context
->protect_kernel_tunables
||
1673 context
->protect_kernel_modules
||
1674 context
->protect_control_groups
)
1677 if (context
->mount_apivfs
&& (context
->root_image
|| context
->root_directory
))
1683 static int setup_private_users(uid_t uid
, gid_t gid
) {
1684 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1685 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1686 _cleanup_close_
int unshare_ready_fd
= -1;
1687 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1693 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1694 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1695 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1696 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1697 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1698 * continues execution normally. */
1700 if (uid
!= 0 && uid_is_valid(uid
)) {
1701 r
= asprintf(&uid_map
,
1702 "0 0 1\n" /* Map root → root */
1703 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1708 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1713 if (gid
!= 0 && gid_is_valid(gid
)) {
1714 r
= asprintf(&gid_map
,
1715 "0 0 1\n" /* Map root → root */
1716 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1721 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1726 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1728 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1729 if (unshare_ready_fd
< 0)
1732 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1734 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1742 _cleanup_close_
int fd
= -1;
1746 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1747 * here, after the parent opened its own user namespace. */
1750 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1752 /* Wait until the parent unshared the user namespace */
1753 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1758 /* Disable the setgroups() system call in the child user namespace, for good. */
1759 a
= procfs_file_alloca(ppid
, "setgroups");
1760 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1762 if (errno
!= ENOENT
) {
1767 /* If the file is missing the kernel is too old, let's continue anyway. */
1769 if (write(fd
, "deny\n", 5) < 0) {
1774 fd
= safe_close(fd
);
1777 /* First write the GID map */
1778 a
= procfs_file_alloca(ppid
, "gid_map");
1779 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1784 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1788 fd
= safe_close(fd
);
1790 /* The write the UID map */
1791 a
= procfs_file_alloca(ppid
, "uid_map");
1792 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1797 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1802 _exit(EXIT_SUCCESS
);
1805 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1806 _exit(EXIT_FAILURE
);
1809 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1811 if (unshare(CLONE_NEWUSER
) < 0)
1814 /* Let the child know that the namespace is ready now */
1815 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1818 /* Try to read an error code from the child */
1819 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1822 if (n
== sizeof(r
)) { /* an error code was sent to us */
1827 if (n
!= 0) /* on success we should have read 0 bytes */
1830 r
= wait_for_terminate(pid
, &si
);
1835 /* If something strange happened with the child, let's consider this fatal, too */
1836 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1842 static int setup_runtime_directory(
1843 const ExecContext
*context
,
1844 const ExecParameters
*params
,
1854 STRV_FOREACH(rt
, context
->runtime_directory
) {
1855 _cleanup_free_
char *p
;
1857 p
= strjoin(params
->runtime_prefix
, "/", *rt
);
1861 r
= mkdir_p_label(p
, context
->runtime_directory_mode
);
1865 r
= chmod_and_chown(p
, context
->runtime_directory_mode
, uid
, gid
);
1873 static int setup_smack(
1874 const ExecContext
*context
,
1875 const ExecCommand
*command
) {
1883 if (!mac_smack_use())
1886 if (context
->smack_process_label
) {
1887 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1891 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1893 _cleanup_free_
char *exec_label
= NULL
;
1895 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1896 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1899 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
1909 static int compile_read_write_paths(
1910 const ExecContext
*context
,
1911 const ExecParameters
*params
,
1914 _cleanup_strv_free_
char **l
= NULL
;
1917 /* Compile the list of writable paths. This is the combination of
1918 * the explicitly configured paths, plus all runtime directories. */
1920 if (strv_isempty(context
->read_write_paths
) &&
1921 strv_isempty(context
->runtime_directory
)) {
1922 *ret
= NULL
; /* NOP if neither is set */
1926 l
= strv_copy(context
->read_write_paths
);
1930 STRV_FOREACH(rt
, context
->runtime_directory
) {
1933 s
= strjoin(params
->runtime_prefix
, "/", *rt
);
1937 if (strv_consume(&l
, s
) < 0)
1947 static int apply_mount_namespace(
1949 ExecCommand
*command
,
1950 const ExecContext
*context
,
1951 const ExecParameters
*params
,
1952 ExecRuntime
*runtime
) {
1954 _cleanup_strv_free_
char **rw
= NULL
;
1955 char *tmp
= NULL
, *var
= NULL
;
1956 const char *root_dir
= NULL
, *root_image
= NULL
;
1957 NameSpaceInfo ns_info
= {
1958 .ignore_protect_paths
= false,
1959 .private_dev
= context
->private_devices
,
1960 .protect_control_groups
= context
->protect_control_groups
,
1961 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
1962 .protect_kernel_modules
= context
->protect_kernel_modules
,
1963 .mount_apivfs
= context
->mount_apivfs
,
1965 bool apply_restrictions
;
1970 /* The runtime struct only contains the parent of the private /tmp,
1971 * which is non-accessible to world users. Inside of it there's a /tmp
1972 * that is sticky, and that's the one we want to use here. */
1974 if (context
->private_tmp
&& runtime
) {
1975 if (runtime
->tmp_dir
)
1976 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
1977 if (runtime
->var_tmp_dir
)
1978 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
1981 r
= compile_read_write_paths(context
, params
, &rw
);
1985 if (params
->flags
& EXEC_APPLY_CHROOT
) {
1986 root_image
= context
->root_image
;
1989 root_dir
= context
->root_directory
;
1993 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
1994 * sandbox info, otherwise enforce it, don't ignore protected paths and
1995 * fail if we are enable to apply the sandbox inside the mount namespace.
1997 if (!context
->dynamic_user
&& root_dir
)
1998 ns_info
.ignore_protect_paths
= true;
2000 apply_restrictions
= (params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
;
2002 r
= setup_namespace(root_dir
, root_image
,
2004 apply_restrictions
? context
->read_only_paths
: NULL
,
2005 apply_restrictions
? context
->inaccessible_paths
: NULL
,
2006 context
->bind_mounts
,
2007 context
->n_bind_mounts
,
2010 apply_restrictions
? context
->protect_home
: PROTECT_HOME_NO
,
2011 apply_restrictions
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2012 context
->mount_flags
,
2013 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2015 /* If we couldn't set up the namespace this is probably due to a
2016 * missing capability. In this case, silently proceeed. */
2017 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2019 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2027 static int apply_working_directory(
2028 const ExecContext
*context
,
2029 const ExecParameters
*params
,
2031 const bool needs_mount_ns
,
2037 assert(exit_status
);
2039 if (context
->working_directory_home
) {
2042 *exit_status
= EXIT_CHDIR
;
2048 } else if (context
->working_directory
)
2049 wd
= context
->working_directory
;
2053 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2054 if (!needs_mount_ns
&& context
->root_directory
)
2055 if (chroot(context
->root_directory
) < 0) {
2056 *exit_status
= EXIT_CHROOT
;
2062 d
= prefix_roota(context
->root_directory
, wd
);
2064 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2065 *exit_status
= EXIT_CHDIR
;
2072 static int setup_keyring(Unit
*u
, const ExecParameters
*p
, uid_t uid
, gid_t gid
) {
2073 key_serial_t keyring
;
2078 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2079 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2080 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2081 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2082 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2083 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2085 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2088 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2089 if (keyring
== -1) {
2090 if (errno
== ENOSYS
)
2091 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2092 else if (IN_SET(errno
, EACCES
, EPERM
))
2093 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2094 else if (errno
== EDQUOT
)
2095 log_debug_errno(errno
, "Out of kernel keyrings to allocate, ignoring.");
2097 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2102 /* Populate they keyring with the invocation ID by default. */
2103 if (!sd_id128_is_null(u
->invocation_id
)) {
2106 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2108 log_debug_errno(errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2110 if (keyctl(KEYCTL_SETPERM
, key
,
2111 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2112 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2113 return log_error_errno(errno
, "Failed to restrict invocation ID permission: %m");
2117 /* And now, make the keyring owned by the service's user */
2118 if (uid_is_valid(uid
) || gid_is_valid(gid
))
2119 if (keyctl(KEYCTL_CHOWN
, keyring
, uid
, gid
, 0) < 0)
2120 return log_error_errno(errno
, "Failed to change ownership of session keyring: %m");
2125 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
2133 array
[(*n
)++] = pair
[0];
2135 array
[(*n
)++] = pair
[1];
2138 static int close_remaining_fds(
2139 const ExecParameters
*params
,
2140 ExecRuntime
*runtime
,
2141 DynamicCreds
*dcreds
,
2144 int *fds
, unsigned n_fds
) {
2146 unsigned n_dont_close
= 0;
2147 int dont_close
[n_fds
+ 12];
2151 if (params
->stdin_fd
>= 0)
2152 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2153 if (params
->stdout_fd
>= 0)
2154 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2155 if (params
->stderr_fd
>= 0)
2156 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2159 dont_close
[n_dont_close
++] = socket_fd
;
2161 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2162 n_dont_close
+= n_fds
;
2166 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2170 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2172 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2175 if (user_lookup_fd
>= 0)
2176 dont_close
[n_dont_close
++] = user_lookup_fd
;
2178 return close_all_fds(dont_close
, n_dont_close
);
2181 static int send_user_lookup(
2189 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2190 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2193 if (user_lookup_fd
< 0)
2196 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2199 if (writev(user_lookup_fd
,
2201 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2202 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2203 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2209 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2216 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2221 if (!c
->working_directory_home
)
2225 /* Hardcode /root as home directory for UID 0 */
2230 r
= get_home_dir(buf
);
2238 static int exec_child(
2240 ExecCommand
*command
,
2241 const ExecContext
*context
,
2242 const ExecParameters
*params
,
2243 ExecRuntime
*runtime
,
2244 DynamicCreds
*dcreds
,
2249 unsigned n_storage_fds
,
2250 unsigned n_socket_fds
,
2254 char **error_message
) {
2256 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2257 _cleanup_free_
char *mac_selinux_context_net
= NULL
, *home_buffer
= NULL
;
2258 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2259 const char *username
= NULL
, *groupname
= NULL
;
2260 const char *home
= NULL
, *shell
= NULL
;
2261 dev_t journal_stream_dev
= 0;
2262 ino_t journal_stream_ino
= 0;
2263 bool needs_mount_namespace
;
2264 uid_t uid
= UID_INVALID
;
2265 gid_t gid
= GID_INVALID
;
2266 int i
, r
, ngids
= 0;
2273 assert(exit_status
);
2274 assert(error_message
);
2275 /* We don't always set error_message, hence it must be initialized */
2276 assert(*error_message
== NULL
);
2278 rename_process_from_path(command
->path
);
2280 /* We reset exactly these signals, since they are the
2281 * only ones we set to SIG_IGN in the main daemon. All
2282 * others we leave untouched because we set them to
2283 * SIG_DFL or a valid handler initially, both of which
2284 * will be demoted to SIG_DFL. */
2285 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2286 SIGNALS_IGNORE
, -1);
2288 if (context
->ignore_sigpipe
)
2289 (void) ignore_signals(SIGPIPE
, -1);
2291 r
= reset_signal_mask();
2293 *exit_status
= EXIT_SIGNAL_MASK
;
2294 *error_message
= strdup("Failed to reset signal mask");
2295 /* If strdup fails, here and below, we will just print the generic error message. */
2299 if (params
->idle_pipe
)
2300 do_idle_pipe_dance(params
->idle_pipe
);
2302 /* Close sockets very early to make sure we don't
2303 * block init reexecution because it cannot bind its
2308 n_fds
= n_storage_fds
+ n_socket_fds
;
2309 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2311 *exit_status
= EXIT_FDS
;
2312 *error_message
= strdup("Failed to close remaining fds");
2316 if (!context
->same_pgrp
)
2318 *exit_status
= EXIT_SETSID
;
2322 exec_context_tty_reset(context
, params
);
2324 if (unit_shall_confirm_spawn(unit
)) {
2325 const char *vc
= params
->confirm_spawn
;
2326 _cleanup_free_
char *cmdline
= NULL
;
2328 cmdline
= exec_command_line(argv
);
2330 *exit_status
= EXIT_CONFIRM
;
2334 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2335 if (r
!= CONFIRM_EXECUTE
) {
2336 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2337 *exit_status
= EXIT_SUCCESS
;
2340 *exit_status
= EXIT_CONFIRM
;
2341 *error_message
= strdup("Execution cancelled");
2346 if (context
->dynamic_user
&& dcreds
) {
2348 /* Make sure we bypass our own NSS module for any NSS checks */
2349 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2350 *exit_status
= EXIT_USER
;
2351 *error_message
= strdup("Failed to update environment");
2355 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2357 *exit_status
= EXIT_USER
;
2358 *error_message
= strdup("Failed to update dynamic user credentials");
2362 if (!uid_is_valid(uid
)) {
2363 *exit_status
= EXIT_USER
;
2364 (void) asprintf(error_message
, "UID validation failed for \""UID_FMT
"\"", uid
);
2365 /* If asprintf fails, here and below, we will just print the generic error message. */
2369 if (!gid_is_valid(gid
)) {
2370 *exit_status
= EXIT_USER
;
2371 (void) asprintf(error_message
, "GID validation failed for \""GID_FMT
"\"", gid
);
2376 username
= dcreds
->user
->name
;
2379 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2381 *exit_status
= EXIT_USER
;
2382 *error_message
= strdup("Failed to determine user credentials");
2386 r
= get_fixed_group(context
, &groupname
, &gid
);
2388 *exit_status
= EXIT_GROUP
;
2389 *error_message
= strdup("Failed to determine group credentials");
2394 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2395 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2396 &supplementary_gids
, &ngids
);
2398 *exit_status
= EXIT_GROUP
;
2399 *error_message
= strdup("Failed to determine supplementary groups");
2403 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2405 *exit_status
= EXIT_USER
;
2406 *error_message
= strdup("Failed to send user credentials to PID1");
2410 user_lookup_fd
= safe_close(user_lookup_fd
);
2412 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2414 *exit_status
= EXIT_CHDIR
;
2415 *error_message
= strdup("Failed to determine $HOME for user");
2419 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2420 * must sure to drop O_NONBLOCK */
2422 (void) fd_nonblock(socket_fd
, false);
2424 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2426 *exit_status
= EXIT_STDIN
;
2427 *error_message
= strdup("Failed to set up stdin");
2431 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2433 *exit_status
= EXIT_STDOUT
;
2434 *error_message
= strdup("Failed to set up stdout");
2438 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2440 *exit_status
= EXIT_STDERR
;
2441 *error_message
= strdup("Failed to set up stderr");
2445 if (params
->cgroup_path
) {
2446 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2448 *exit_status
= EXIT_CGROUP
;
2449 (void) asprintf(error_message
, "Failed to attach to cgroup %s", params
->cgroup_path
);
2454 if (context
->oom_score_adjust_set
) {
2455 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2457 /* When we can't make this change due to EPERM, then
2458 * let's silently skip over it. User namespaces
2459 * prohibit write access to this file, and we
2460 * shouldn't trip up over that. */
2462 sprintf(t
, "%i", context
->oom_score_adjust
);
2463 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2464 if (r
== -EPERM
|| r
== -EACCES
) {
2466 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2469 *exit_status
= EXIT_OOM_ADJUST
;
2470 *error_message
= strdup("Failed to write /proc/self/oom_score_adj");
2475 if (context
->nice_set
)
2476 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2477 *exit_status
= EXIT_NICE
;
2481 if (context
->cpu_sched_set
) {
2482 struct sched_param param
= {
2483 .sched_priority
= context
->cpu_sched_priority
,
2486 r
= sched_setscheduler(0,
2487 context
->cpu_sched_policy
|
2488 (context
->cpu_sched_reset_on_fork
?
2489 SCHED_RESET_ON_FORK
: 0),
2492 *exit_status
= EXIT_SETSCHEDULER
;
2497 if (context
->cpuset
)
2498 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2499 *exit_status
= EXIT_CPUAFFINITY
;
2503 if (context
->ioprio_set
)
2504 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2505 *exit_status
= EXIT_IOPRIO
;
2509 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2510 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2511 *exit_status
= EXIT_TIMERSLACK
;
2515 if (context
->personality
!= PERSONALITY_INVALID
)
2516 if (personality(context
->personality
) < 0) {
2517 *exit_status
= EXIT_PERSONALITY
;
2521 if (context
->utmp_id
)
2522 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0),
2524 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2525 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2529 if (context
->user
) {
2530 r
= chown_terminal(STDIN_FILENO
, uid
);
2532 *exit_status
= EXIT_STDIN
;
2537 /* If delegation is enabled we'll pass ownership of the cgroup
2538 * (but only in systemd's own controller hierarchy!) to the
2539 * user of the new process. */
2540 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2541 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2543 *exit_status
= EXIT_CGROUP
;
2548 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2550 *exit_status
= EXIT_CGROUP
;
2555 if (!strv_isempty(context
->runtime_directory
) && params
->runtime_prefix
) {
2556 r
= setup_runtime_directory(context
, params
, uid
, gid
);
2558 *exit_status
= EXIT_RUNTIME_DIRECTORY
;
2563 r
= build_environment(
2575 *exit_status
= EXIT_MEMORY
;
2579 r
= build_pass_environment(context
, &pass_env
);
2581 *exit_status
= EXIT_MEMORY
;
2585 accum_env
= strv_env_merge(5,
2586 params
->environment
,
2589 context
->environment
,
2593 *exit_status
= EXIT_MEMORY
;
2596 accum_env
= strv_env_clean(accum_env
);
2598 (void) umask(context
->umask
);
2600 r
= setup_keyring(unit
, params
, uid
, gid
);
2602 *exit_status
= EXIT_KEYRING
;
2606 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2607 if (context
->pam_name
&& username
) {
2608 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2610 *exit_status
= EXIT_PAM
;
2616 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2617 r
= setup_netns(runtime
->netns_storage_socket
);
2619 *exit_status
= EXIT_NETWORK
;
2624 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2625 if (needs_mount_namespace
) {
2626 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
2628 *exit_status
= EXIT_NAMESPACE
;
2633 /* Apply just after mount namespace setup */
2634 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
2638 /* Drop groups as early as possbile */
2639 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2640 r
= enforce_groups(context
, gid
, supplementary_gids
, ngids
);
2642 *exit_status
= EXIT_GROUP
;
2648 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) &&
2649 mac_selinux_use() &&
2650 params
->selinux_context_net
&&
2652 !command
->privileged
) {
2654 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2656 *exit_status
= EXIT_SELINUX_CONTEXT
;
2662 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2663 r
= setup_private_users(uid
, gid
);
2665 *exit_status
= EXIT_USER
;
2670 /* We repeat the fd closing here, to make sure that
2671 * nothing is leaked from the PAM modules. Note that
2672 * we are more aggressive this time since socket_fd
2673 * and the netns fds we don't need anymore. The custom
2674 * endpoint fd was needed to upload the policy and can
2675 * now be closed as well. */
2676 r
= close_all_fds(fds
, n_fds
);
2678 r
= shift_fds(fds
, n_fds
);
2680 r
= flags_fds(fds
, n_storage_fds
, n_socket_fds
, context
->non_blocking
);
2682 *exit_status
= EXIT_FDS
;
2686 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2688 int secure_bits
= context
->secure_bits
;
2690 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2692 if (!context
->rlimit
[i
])
2695 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2697 *exit_status
= EXIT_LIMITS
;
2702 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2703 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2704 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2705 *exit_status
= EXIT_LIMITS
;
2710 if (!cap_test_all(context
->capability_bounding_set
)) {
2711 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2713 *exit_status
= EXIT_CAPABILITIES
;
2714 *error_message
= strdup("Failed to drop capabilities");
2719 /* This is done before enforce_user, but ambient set
2720 * does not survive over setresuid() if keep_caps is not set. */
2721 if (context
->capability_ambient_set
!= 0) {
2722 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2724 *exit_status
= EXIT_CAPABILITIES
;
2725 *error_message
= strdup("Failed to apply ambient capabilities (before UID change)");
2730 if (context
->user
) {
2731 r
= enforce_user(context
, uid
);
2733 *exit_status
= EXIT_USER
;
2734 (void) asprintf(error_message
, "Failed to change UID to "UID_FMT
, uid
);
2737 if (context
->capability_ambient_set
!= 0) {
2739 /* Fix the ambient capabilities after user change. */
2740 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2742 *exit_status
= EXIT_CAPABILITIES
;
2743 *error_message
= strdup("Failed to apply ambient capabilities (after UID change)");
2747 /* If we were asked to change user and ambient capabilities
2748 * were requested, we had to add keep-caps to the securebits
2749 * so that we would maintain the inherited capability set
2750 * through the setresuid(). Make sure that the bit is added
2751 * also to the context secure_bits so that we don't try to
2752 * drop the bit away next. */
2754 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2758 /* Apply the MAC contexts late, but before seccomp syscall filtering, as those should really be last to
2759 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
2760 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
2761 * are restricted. */
2764 if (mac_selinux_use()) {
2765 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2768 r
= setexeccon(exec_context
);
2770 *exit_status
= EXIT_SELINUX_CONTEXT
;
2771 (void) asprintf(error_message
, "Failed to set SELinux context to %s", exec_context
);
2778 r
= setup_smack(context
, command
);
2780 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2781 *error_message
= strdup("Failed to set SMACK process label");
2785 #ifdef HAVE_APPARMOR
2786 if (context
->apparmor_profile
&& mac_apparmor_use()) {
2787 r
= aa_change_onexec(context
->apparmor_profile
);
2788 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2789 *exit_status
= EXIT_APPARMOR_PROFILE
;
2790 (void) asprintf(error_message
,
2791 "Failed to prepare AppArmor profile change to %s",
2792 context
->apparmor_profile
);
2798 /* PR_GET_SECUREBITS is not privileged, while
2799 * PR_SET_SECUREBITS is. So to suppress
2800 * potential EPERMs we'll try not to call
2801 * PR_SET_SECUREBITS unless necessary. */
2802 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2803 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2804 *exit_status
= EXIT_SECUREBITS
;
2805 *error_message
= strdup("Failed to set secure bits");
2809 if (context_has_no_new_privileges(context
))
2810 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2811 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2812 *error_message
= strdup("Failed to disable new privileges");
2817 r
= apply_address_families(unit
, context
);
2819 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2820 *error_message
= strdup("Failed to restrict address families");
2824 r
= apply_memory_deny_write_execute(unit
, context
);
2826 *exit_status
= EXIT_SECCOMP
;
2827 *error_message
= strdup("Failed to disable writing to executable memory");
2831 r
= apply_restrict_realtime(unit
, context
);
2833 *exit_status
= EXIT_SECCOMP
;
2834 *error_message
= strdup("Failed to apply realtime restrictions");
2838 r
= apply_restrict_namespaces(unit
, context
);
2840 *exit_status
= EXIT_SECCOMP
;
2841 *error_message
= strdup("Failed to apply namespace restrictions");
2845 r
= apply_protect_sysctl(unit
, context
);
2847 *exit_status
= EXIT_SECCOMP
;
2848 *error_message
= strdup("Failed to apply sysctl restrictions");
2852 r
= apply_protect_kernel_modules(unit
, context
);
2854 *exit_status
= EXIT_SECCOMP
;
2855 *error_message
= strdup("Failed to apply module loading restrictions");
2859 r
= apply_private_devices(unit
, context
);
2861 *exit_status
= EXIT_SECCOMP
;
2862 *error_message
= strdup("Failed to set up private devices");
2866 r
= apply_syscall_archs(unit
, context
);
2868 *exit_status
= EXIT_SECCOMP
;
2869 *error_message
= strdup("Failed to apply syscall architecture restrictions");
2873 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
2874 * by the filter as little as possible. */
2875 r
= apply_syscall_filter(unit
, context
);
2877 *exit_status
= EXIT_SECCOMP
;
2878 *error_message
= strdup("Failed to apply syscall filters");
2884 final_argv
= replace_env_argv(argv
, accum_env
);
2886 *exit_status
= EXIT_MEMORY
;
2887 *error_message
= strdup("Failed to prepare process arguments");
2891 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2892 _cleanup_free_
char *line
;
2894 line
= exec_command_line(final_argv
);
2897 log_struct(LOG_DEBUG
,
2898 "EXECUTABLE=%s", command
->path
,
2899 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2906 execve(command
->path
, final_argv
, accum_env
);
2907 *exit_status
= EXIT_EXEC
;
2911 int exec_spawn(Unit
*unit
,
2912 ExecCommand
*command
,
2913 const ExecContext
*context
,
2914 const ExecParameters
*params
,
2915 ExecRuntime
*runtime
,
2916 DynamicCreds
*dcreds
,
2919 _cleanup_strv_free_
char **files_env
= NULL
;
2921 unsigned n_storage_fds
= 0, n_socket_fds
= 0;
2922 _cleanup_free_
char *line
= NULL
;
2924 int named_iofds
[3] = { -1, -1, -1 };
2933 assert(params
->fds
|| (params
->n_storage_fds
+ params
->n_socket_fds
<= 0));
2935 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2936 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2937 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
2939 if (params
->n_socket_fds
> 1) {
2940 log_unit_error(unit
, "Got more than one socket.");
2944 if (params
->n_socket_fds
== 0) {
2945 log_unit_error(unit
, "Got no socket.");
2949 socket_fd
= params
->fds
[0];
2953 n_storage_fds
= params
->n_storage_fds
;
2954 n_socket_fds
= params
->n_socket_fds
;
2957 r
= exec_context_named_iofds(unit
, context
, params
, named_iofds
);
2959 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
2961 r
= exec_context_load_environment(unit
, context
, &files_env
);
2963 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
2965 argv
= params
->argv
?: command
->argv
;
2966 line
= exec_command_line(argv
);
2970 log_struct(LOG_DEBUG
,
2971 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
2972 "EXECUTABLE=%s", command
->path
,
2977 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
2981 _cleanup_free_
char *error_message
= NULL
;
2983 r
= exec_child(unit
,
2996 unit
->manager
->user_lookup_fds
[1],
3002 log_struct_errno(LOG_ERR
, r
,
3003 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3005 LOG_UNIT_MESSAGE(unit
, "%s: %m",
3007 "EXECUTABLE=%s", command
->path
,
3009 else if (r
== -ENOENT
&& command
->ignore
)
3010 log_struct_errno(LOG_INFO
, r
,
3011 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3013 LOG_UNIT_MESSAGE(unit
, "Skipped spawning %s: %m",
3015 "EXECUTABLE=%s", command
->path
,
3018 log_struct_errno(LOG_ERR
, r
,
3019 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3021 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3022 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3024 "EXECUTABLE=%s", command
->path
,
3031 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3033 /* We add the new process to the cgroup both in the child (so
3034 * that we can be sure that no user code is ever executed
3035 * outside of the cgroup) and in the parent (so that we can be
3036 * sure that when we kill the cgroup the process will be
3038 if (params
->cgroup_path
)
3039 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3041 exec_status_start(&command
->exec_status
, pid
);
3047 void exec_context_init(ExecContext
*c
) {
3051 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3052 c
->cpu_sched_policy
= SCHED_OTHER
;
3053 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3054 c
->syslog_level_prefix
= true;
3055 c
->ignore_sigpipe
= true;
3056 c
->timer_slack_nsec
= NSEC_INFINITY
;
3057 c
->personality
= PERSONALITY_INVALID
;
3058 c
->runtime_directory_mode
= 0755;
3059 c
->capability_bounding_set
= CAP_ALL
;
3060 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
3063 void exec_context_done(ExecContext
*c
) {
3068 c
->environment
= strv_free(c
->environment
);
3069 c
->environment_files
= strv_free(c
->environment_files
);
3070 c
->pass_environment
= strv_free(c
->pass_environment
);
3072 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
3073 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
3075 for (l
= 0; l
< 3; l
++)
3076 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3078 c
->working_directory
= mfree(c
->working_directory
);
3079 c
->root_directory
= mfree(c
->root_directory
);
3080 c
->root_image
= mfree(c
->root_image
);
3081 c
->tty_path
= mfree(c
->tty_path
);
3082 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3083 c
->user
= mfree(c
->user
);
3084 c
->group
= mfree(c
->group
);
3086 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3088 c
->pam_name
= mfree(c
->pam_name
);
3090 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3091 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3092 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3094 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3097 CPU_FREE(c
->cpuset
);
3099 c
->utmp_id
= mfree(c
->utmp_id
);
3100 c
->selinux_context
= mfree(c
->selinux_context
);
3101 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3103 c
->syscall_filter
= set_free(c
->syscall_filter
);
3104 c
->syscall_archs
= set_free(c
->syscall_archs
);
3105 c
->address_families
= set_free(c
->address_families
);
3107 c
->runtime_directory
= strv_free(c
->runtime_directory
);
3110 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
3115 if (!runtime_prefix
)
3118 STRV_FOREACH(i
, c
->runtime_directory
) {
3119 _cleanup_free_
char *p
;
3121 p
= strjoin(runtime_prefix
, "/", *i
);
3125 /* We execute this synchronously, since we need to be
3126 * sure this is gone when we start the service
3128 (void) rm_rf(p
, REMOVE_ROOT
);
3134 void exec_command_done(ExecCommand
*c
) {
3137 c
->path
= mfree(c
->path
);
3139 c
->argv
= strv_free(c
->argv
);
3142 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3145 for (i
= 0; i
< n
; i
++)
3146 exec_command_done(c
+i
);
3149 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3153 LIST_REMOVE(command
, c
, i
);
3154 exec_command_done(i
);
3161 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3164 for (i
= 0; i
< n
; i
++)
3165 c
[i
] = exec_command_free_list(c
[i
]);
3168 typedef struct InvalidEnvInfo
{
3173 static void invalid_env(const char *p
, void *userdata
) {
3174 InvalidEnvInfo
*info
= userdata
;
3176 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3179 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3184 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3186 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3188 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3190 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3192 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3194 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3200 int exec_context_named_iofds(Unit
*unit
, const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3201 unsigned i
, targets
;
3202 const char* stdio_fdname
[3];
3208 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3209 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3210 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3212 for (i
= 0; i
< 3; i
++)
3213 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3215 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3217 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3218 if (named_iofds
[STDIN_FILENO
] < 0 &&
3219 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3220 stdio_fdname
[STDIN_FILENO
] &&
3221 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3223 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3226 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3227 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3228 stdio_fdname
[STDOUT_FILENO
] &&
3229 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3231 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3234 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3235 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3236 stdio_fdname
[STDERR_FILENO
] &&
3237 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3239 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3243 return targets
== 0 ? 0 : -ENOENT
;
3246 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
3247 char **i
, **r
= NULL
;
3252 STRV_FOREACH(i
, c
->environment_files
) {
3256 bool ignore
= false;
3258 _cleanup_globfree_ glob_t pglob
= {};
3267 if (!path_is_absolute(fn
)) {
3275 /* Filename supports globbing, take all matching files */
3276 k
= safe_glob(fn
, 0, &pglob
);
3285 /* When we don't match anything, -ENOENT should be returned */
3286 assert(pglob
.gl_pathc
> 0);
3288 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
3289 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3297 /* Log invalid environment variables with filename */
3299 InvalidEnvInfo info
= {
3301 .path
= pglob
.gl_pathv
[n
]
3304 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3312 m
= strv_env_merge(2, r
, p
);
3328 static bool tty_may_match_dev_console(const char *tty
) {
3329 _cleanup_free_
char *active
= NULL
;
3335 if (startswith(tty
, "/dev/"))
3338 /* trivial identity? */
3339 if (streq(tty
, "console"))
3342 console
= resolve_dev_console(&active
);
3343 /* if we could not resolve, assume it may */
3347 /* "tty0" means the active VC, so it may be the same sometimes */
3348 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3351 bool exec_context_may_touch_console(ExecContext
*ec
) {
3353 return (ec
->tty_reset
||
3355 ec
->tty_vt_disallocate
||
3356 is_terminal_input(ec
->std_input
) ||
3357 is_terminal_output(ec
->std_output
) ||
3358 is_terminal_output(ec
->std_error
)) &&
3359 tty_may_match_dev_console(exec_context_tty_path(ec
));
3362 static void strv_fprintf(FILE *f
, char **l
) {
3368 fprintf(f
, " %s", *g
);
3371 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3379 prefix
= strempty(prefix
);
3383 "%sWorkingDirectory: %s\n"
3384 "%sRootDirectory: %s\n"
3385 "%sNonBlocking: %s\n"
3386 "%sPrivateTmp: %s\n"
3387 "%sPrivateDevices: %s\n"
3388 "%sProtectKernelTunables: %s\n"
3389 "%sProtectKernelModules: %s\n"
3390 "%sProtectControlGroups: %s\n"
3391 "%sPrivateNetwork: %s\n"
3392 "%sPrivateUsers: %s\n"
3393 "%sProtectHome: %s\n"
3394 "%sProtectSystem: %s\n"
3395 "%sMountAPIVFS: %s\n"
3396 "%sIgnoreSIGPIPE: %s\n"
3397 "%sMemoryDenyWriteExecute: %s\n"
3398 "%sRestrictRealtime: %s\n",
3400 prefix
, c
->working_directory
? c
->working_directory
: "/",
3401 prefix
, c
->root_directory
? c
->root_directory
: "/",
3402 prefix
, yes_no(c
->non_blocking
),
3403 prefix
, yes_no(c
->private_tmp
),
3404 prefix
, yes_no(c
->private_devices
),
3405 prefix
, yes_no(c
->protect_kernel_tunables
),
3406 prefix
, yes_no(c
->protect_kernel_modules
),
3407 prefix
, yes_no(c
->protect_control_groups
),
3408 prefix
, yes_no(c
->private_network
),
3409 prefix
, yes_no(c
->private_users
),
3410 prefix
, protect_home_to_string(c
->protect_home
),
3411 prefix
, protect_system_to_string(c
->protect_system
),
3412 prefix
, yes_no(c
->mount_apivfs
),
3413 prefix
, yes_no(c
->ignore_sigpipe
),
3414 prefix
, yes_no(c
->memory_deny_write_execute
),
3415 prefix
, yes_no(c
->restrict_realtime
));
3418 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3420 STRV_FOREACH(e
, c
->environment
)
3421 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3423 STRV_FOREACH(e
, c
->environment_files
)
3424 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3426 STRV_FOREACH(e
, c
->pass_environment
)
3427 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3429 fprintf(f
, "%sRuntimeDirectoryMode: %04o\n", prefix
, c
->runtime_directory_mode
);
3431 STRV_FOREACH(d
, c
->runtime_directory
)
3432 fprintf(f
, "%sRuntimeDirectory: %s\n", prefix
, *d
);
3439 if (c
->oom_score_adjust_set
)
3441 "%sOOMScoreAdjust: %i\n",
3442 prefix
, c
->oom_score_adjust
);
3444 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3446 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3447 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3448 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3449 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3452 if (c
->ioprio_set
) {
3453 _cleanup_free_
char *class_str
= NULL
;
3455 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3457 "%sIOSchedulingClass: %s\n"
3458 "%sIOPriority: %i\n",
3459 prefix
, strna(class_str
),
3460 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
3463 if (c
->cpu_sched_set
) {
3464 _cleanup_free_
char *policy_str
= NULL
;
3466 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3468 "%sCPUSchedulingPolicy: %s\n"
3469 "%sCPUSchedulingPriority: %i\n"
3470 "%sCPUSchedulingResetOnFork: %s\n",
3471 prefix
, strna(policy_str
),
3472 prefix
, c
->cpu_sched_priority
,
3473 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3477 fprintf(f
, "%sCPUAffinity:", prefix
);
3478 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3479 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3480 fprintf(f
, " %u", i
);
3484 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3485 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3488 "%sStandardInput: %s\n"
3489 "%sStandardOutput: %s\n"
3490 "%sStandardError: %s\n",
3491 prefix
, exec_input_to_string(c
->std_input
),
3492 prefix
, exec_output_to_string(c
->std_output
),
3493 prefix
, exec_output_to_string(c
->std_error
));
3499 "%sTTYVHangup: %s\n"
3500 "%sTTYVTDisallocate: %s\n",
3501 prefix
, c
->tty_path
,
3502 prefix
, yes_no(c
->tty_reset
),
3503 prefix
, yes_no(c
->tty_vhangup
),
3504 prefix
, yes_no(c
->tty_vt_disallocate
));
3506 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3507 c
->std_output
== EXEC_OUTPUT_KMSG
||
3508 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3509 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3510 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3511 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3512 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3513 c
->std_error
== EXEC_OUTPUT_KMSG
||
3514 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3515 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3516 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3517 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3519 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3521 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3522 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3525 "%sSyslogFacility: %s\n"
3526 "%sSyslogLevel: %s\n",
3527 prefix
, strna(fac_str
),
3528 prefix
, strna(lvl_str
));
3532 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
3534 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
3535 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
3536 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
3537 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
3538 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
3539 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
3541 if (c
->capability_bounding_set
!= CAP_ALL
) {
3543 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
3545 for (l
= 0; l
<= cap_last_cap(); l
++)
3546 if (c
->capability_bounding_set
& (UINT64_C(1) << l
))
3547 fprintf(f
, " %s", strna(capability_to_name(l
)));
3552 if (c
->capability_ambient_set
!= 0) {
3554 fprintf(f
, "%sAmbientCapabilities:", prefix
);
3556 for (l
= 0; l
<= cap_last_cap(); l
++)
3557 if (c
->capability_ambient_set
& (UINT64_C(1) << l
))
3558 fprintf(f
, " %s", strna(capability_to_name(l
)));
3564 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3566 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3568 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3570 if (strv_length(c
->supplementary_groups
) > 0) {
3571 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3572 strv_fprintf(f
, c
->supplementary_groups
);
3577 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3579 if (strv_length(c
->read_write_paths
) > 0) {
3580 fprintf(f
, "%sReadWritePaths:", prefix
);
3581 strv_fprintf(f
, c
->read_write_paths
);
3585 if (strv_length(c
->read_only_paths
) > 0) {
3586 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3587 strv_fprintf(f
, c
->read_only_paths
);
3591 if (strv_length(c
->inaccessible_paths
) > 0) {
3592 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3593 strv_fprintf(f
, c
->inaccessible_paths
);
3597 if (c
->n_bind_mounts
> 0)
3598 for (i
= 0; i
< c
->n_bind_mounts
; i
++) {
3599 fprintf(f
, "%s%s: %s:%s:%s\n", prefix
,
3600 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
3601 c
->bind_mounts
[i
].source
,
3602 c
->bind_mounts
[i
].destination
,
3603 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
3608 "%sUtmpIdentifier: %s\n",
3609 prefix
, c
->utmp_id
);
3611 if (c
->selinux_context
)
3613 "%sSELinuxContext: %s%s\n",
3614 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3616 if (c
->personality
!= PERSONALITY_INVALID
)
3618 "%sPersonality: %s\n",
3619 prefix
, strna(personality_to_string(c
->personality
)));
3621 if (c
->syscall_filter
) {
3629 "%sSystemCallFilter: ",
3632 if (!c
->syscall_whitelist
)
3636 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3637 _cleanup_free_
char *name
= NULL
;
3644 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3645 fputs(strna(name
), f
);
3652 if (c
->syscall_archs
) {
3659 "%sSystemCallArchitectures:",
3663 SET_FOREACH(id
, c
->syscall_archs
, j
)
3664 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3669 if (exec_context_restrict_namespaces_set(c
)) {
3670 _cleanup_free_
char *s
= NULL
;
3672 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
3674 fprintf(f
, "%sRestrictNamespaces: %s\n",
3678 if (c
->syscall_errno
> 0)
3680 "%sSystemCallErrorNumber: %s\n",
3681 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3683 if (c
->apparmor_profile
)
3685 "%sAppArmorProfile: %s%s\n",
3686 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3689 bool exec_context_maintains_privileges(ExecContext
*c
) {
3692 /* Returns true if the process forked off would run under
3693 * an unchanged UID or as root. */
3698 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3704 int exec_context_get_effective_ioprio(ExecContext
*c
) {
3712 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
3714 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
3719 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3724 dual_timestamp_get(&s
->start_timestamp
);
3727 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3730 if (s
->pid
&& s
->pid
!= pid
)
3734 dual_timestamp_get(&s
->exit_timestamp
);
3740 if (context
->utmp_id
)
3741 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3743 exec_context_tty_reset(context
, NULL
);
3747 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3748 char buf
[FORMAT_TIMESTAMP_MAX
];
3756 prefix
= strempty(prefix
);
3759 "%sPID: "PID_FMT
"\n",
3762 if (dual_timestamp_is_set(&s
->start_timestamp
))
3764 "%sStart Timestamp: %s\n",
3765 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3767 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3769 "%sExit Timestamp: %s\n"
3771 "%sExit Status: %i\n",
3772 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3773 prefix
, sigchld_code_to_string(s
->code
),
3777 char *exec_command_line(char **argv
) {
3785 STRV_FOREACH(a
, argv
)
3793 STRV_FOREACH(a
, argv
) {
3800 if (strpbrk(*a
, WHITESPACE
)) {
3811 /* FIXME: this doesn't really handle arguments that have
3812 * spaces and ticks in them */
3817 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3818 _cleanup_free_
char *cmd
= NULL
;
3819 const char *prefix2
;
3824 prefix
= strempty(prefix
);
3825 prefix2
= strjoina(prefix
, "\t");
3827 cmd
= exec_command_line(c
->argv
);
3829 "%sCommand Line: %s\n",
3830 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3832 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3835 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3838 prefix
= strempty(prefix
);
3840 LIST_FOREACH(command
, c
, c
)
3841 exec_command_dump(c
, f
, prefix
);
3844 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3851 /* It's kind of important, that we keep the order here */
3852 LIST_FIND_TAIL(command
, *l
, end
);
3853 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3858 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3866 l
= strv_new_ap(path
, ap
);
3887 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3888 _cleanup_strv_free_
char **l
= NULL
;
3896 l
= strv_new_ap(path
, ap
);
3902 r
= strv_extend_strv(&c
->argv
, l
, false);
3910 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3915 *rt
= new0(ExecRuntime
, 1);
3920 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3925 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
3935 if (!c
->private_network
&& !c
->private_tmp
)
3938 r
= exec_runtime_allocate(rt
);
3942 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
3943 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
3947 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
3948 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
3956 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
3958 assert(r
->n_ref
> 0);
3964 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
3969 assert(r
->n_ref
> 0);
3976 free(r
->var_tmp_dir
);
3977 safe_close_pair(r
->netns_storage_socket
);
3981 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
3990 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
3992 if (rt
->var_tmp_dir
)
3993 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
3995 if (rt
->netns_storage_socket
[0] >= 0) {
3998 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4002 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
4005 if (rt
->netns_storage_socket
[1] >= 0) {
4008 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
4012 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
4018 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
4025 if (streq(key
, "tmp-dir")) {
4028 r
= exec_runtime_allocate(rt
);
4032 copy
= strdup(value
);
4036 free((*rt
)->tmp_dir
);
4037 (*rt
)->tmp_dir
= copy
;
4039 } else if (streq(key
, "var-tmp-dir")) {
4042 r
= exec_runtime_allocate(rt
);
4046 copy
= strdup(value
);
4050 free((*rt
)->var_tmp_dir
);
4051 (*rt
)->var_tmp_dir
= copy
;
4053 } else if (streq(key
, "netns-socket-0")) {
4056 r
= exec_runtime_allocate(rt
);
4060 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4061 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4063 safe_close((*rt
)->netns_storage_socket
[0]);
4064 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4066 } else if (streq(key
, "netns-socket-1")) {
4069 r
= exec_runtime_allocate(rt
);
4073 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4074 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4076 safe_close((*rt
)->netns_storage_socket
[1]);
4077 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
4085 static void *remove_tmpdir_thread(void *p
) {
4086 _cleanup_free_
char *path
= p
;
4088 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4092 void exec_runtime_destroy(ExecRuntime
*rt
) {
4098 /* If there are multiple users of this, let's leave the stuff around */
4103 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4105 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4107 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4114 if (rt
->var_tmp_dir
) {
4115 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4117 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4119 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4120 free(rt
->var_tmp_dir
);
4123 rt
->var_tmp_dir
= NULL
;
4126 safe_close_pair(rt
->netns_storage_socket
);
4129 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4130 [EXEC_INPUT_NULL
] = "null",
4131 [EXEC_INPUT_TTY
] = "tty",
4132 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4133 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4134 [EXEC_INPUT_SOCKET
] = "socket",
4135 [EXEC_INPUT_NAMED_FD
] = "fd",
4138 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4140 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4141 [EXEC_OUTPUT_INHERIT
] = "inherit",
4142 [EXEC_OUTPUT_NULL
] = "null",
4143 [EXEC_OUTPUT_TTY
] = "tty",
4144 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4145 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4146 [EXEC_OUTPUT_KMSG
] = "kmsg",
4147 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4148 [EXEC_OUTPUT_JOURNAL
] = "journal",
4149 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4150 [EXEC_OUTPUT_SOCKET
] = "socket",
4151 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4154 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4156 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4157 [EXEC_UTMP_INIT
] = "init",
4158 [EXEC_UTMP_LOGIN
] = "login",
4159 [EXEC_UTMP_USER
] = "user",
4162 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);