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_exec_directory(
1843 const ExecContext
*context
,
1844 const ExecParameters
*params
,
1847 bool manager_is_system
,
1848 ExecDirectoryType type
,
1851 static const int exit_status_table
[_EXEC_DIRECTORY_MAX
] = {
1852 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
1853 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
1854 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
1855 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
1856 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
1863 assert(type
>= 0 && type
< _EXEC_DIRECTORY_MAX
);
1864 assert(exit_status
);
1866 if (!params
->prefix
[type
])
1869 if (manager_is_system
) {
1870 if (!uid_is_valid(uid
))
1872 if (!gid_is_valid(gid
))
1876 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
1877 _cleanup_free_
char *p
;
1879 p
= strjoin(params
->prefix
[type
], "/", *rt
);
1885 r
= mkdir_parents_label(p
, 0755);
1889 r
= mkdir_p_label(p
, context
->directories
[type
].mode
);
1893 r
= chmod_and_chown(p
, context
->directories
[type
].mode
, uid
, gid
);
1901 *exit_status
= exit_status_table
[type
];
1906 static int setup_smack(
1907 const ExecContext
*context
,
1908 const ExecCommand
*command
) {
1916 if (!mac_smack_use())
1919 if (context
->smack_process_label
) {
1920 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1924 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1926 _cleanup_free_
char *exec_label
= NULL
;
1928 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1929 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1932 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
1942 static int compile_read_write_paths(
1943 const ExecContext
*context
,
1944 const ExecParameters
*params
,
1947 _cleanup_strv_free_
char **l
= NULL
;
1949 ExecDirectoryType i
;
1951 /* Compile the list of writable paths. This is the combination of
1952 * the explicitly configured paths, plus all runtime directories. */
1954 if (strv_isempty(context
->read_write_paths
)) {
1955 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++)
1956 if (!strv_isempty(context
->directories
[i
].paths
))
1959 if (i
== _EXEC_DIRECTORY_MAX
) {
1960 *ret
= NULL
; /* NOP if neither is set */
1965 l
= strv_copy(context
->read_write_paths
);
1969 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++) {
1970 if (!params
->prefix
[i
])
1973 STRV_FOREACH(rt
, context
->directories
[i
].paths
) {
1976 s
= strjoin(params
->prefix
[i
], "/", *rt
);
1980 if (strv_consume(&l
, s
) < 0)
1991 static int apply_mount_namespace(
1993 ExecCommand
*command
,
1994 const ExecContext
*context
,
1995 const ExecParameters
*params
,
1996 ExecRuntime
*runtime
) {
1998 _cleanup_strv_free_
char **rw
= NULL
;
1999 char *tmp
= NULL
, *var
= NULL
;
2000 const char *root_dir
= NULL
, *root_image
= NULL
;
2001 NameSpaceInfo ns_info
= {
2002 .ignore_protect_paths
= false,
2003 .private_dev
= context
->private_devices
,
2004 .protect_control_groups
= context
->protect_control_groups
,
2005 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2006 .protect_kernel_modules
= context
->protect_kernel_modules
,
2007 .mount_apivfs
= context
->mount_apivfs
,
2009 bool apply_restrictions
;
2014 /* The runtime struct only contains the parent of the private /tmp,
2015 * which is non-accessible to world users. Inside of it there's a /tmp
2016 * that is sticky, and that's the one we want to use here. */
2018 if (context
->private_tmp
&& runtime
) {
2019 if (runtime
->tmp_dir
)
2020 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2021 if (runtime
->var_tmp_dir
)
2022 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2025 r
= compile_read_write_paths(context
, params
, &rw
);
2029 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2030 root_image
= context
->root_image
;
2033 root_dir
= context
->root_directory
;
2037 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2038 * sandbox info, otherwise enforce it, don't ignore protected paths and
2039 * fail if we are enable to apply the sandbox inside the mount namespace.
2041 if (!context
->dynamic_user
&& root_dir
)
2042 ns_info
.ignore_protect_paths
= true;
2044 apply_restrictions
= (params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
;
2046 r
= setup_namespace(root_dir
, root_image
,
2048 apply_restrictions
? context
->read_only_paths
: NULL
,
2049 apply_restrictions
? context
->inaccessible_paths
: NULL
,
2050 context
->bind_mounts
,
2051 context
->n_bind_mounts
,
2054 apply_restrictions
? context
->protect_home
: PROTECT_HOME_NO
,
2055 apply_restrictions
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2056 context
->mount_flags
,
2057 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2059 /* If we couldn't set up the namespace this is probably due to a
2060 * missing capability. In this case, silently proceeed. */
2061 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2063 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2071 static int apply_working_directory(
2072 const ExecContext
*context
,
2073 const ExecParameters
*params
,
2075 const bool needs_mount_ns
,
2081 assert(exit_status
);
2083 if (context
->working_directory_home
) {
2086 *exit_status
= EXIT_CHDIR
;
2092 } else if (context
->working_directory
)
2093 wd
= context
->working_directory
;
2097 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2098 if (!needs_mount_ns
&& context
->root_directory
)
2099 if (chroot(context
->root_directory
) < 0) {
2100 *exit_status
= EXIT_CHROOT
;
2106 d
= prefix_roota(context
->root_directory
, wd
);
2108 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2109 *exit_status
= EXIT_CHDIR
;
2116 static int setup_keyring(Unit
*u
, const ExecParameters
*p
, uid_t uid
, gid_t gid
) {
2117 key_serial_t keyring
;
2122 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2123 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2124 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2125 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2126 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2127 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2129 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2132 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2133 if (keyring
== -1) {
2134 if (errno
== ENOSYS
)
2135 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2136 else if (IN_SET(errno
, EACCES
, EPERM
))
2137 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2138 else if (errno
== EDQUOT
)
2139 log_debug_errno(errno
, "Out of kernel keyrings to allocate, ignoring.");
2141 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2146 /* Populate they keyring with the invocation ID by default. */
2147 if (!sd_id128_is_null(u
->invocation_id
)) {
2150 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2152 log_debug_errno(errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2154 if (keyctl(KEYCTL_SETPERM
, key
,
2155 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2156 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2157 return log_error_errno(errno
, "Failed to restrict invocation ID permission: %m");
2161 /* And now, make the keyring owned by the service's user */
2162 if (uid_is_valid(uid
) || gid_is_valid(gid
))
2163 if (keyctl(KEYCTL_CHOWN
, keyring
, uid
, gid
, 0) < 0)
2164 return log_error_errno(errno
, "Failed to change ownership of session keyring: %m");
2169 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
2177 array
[(*n
)++] = pair
[0];
2179 array
[(*n
)++] = pair
[1];
2182 static int close_remaining_fds(
2183 const ExecParameters
*params
,
2184 ExecRuntime
*runtime
,
2185 DynamicCreds
*dcreds
,
2188 int *fds
, unsigned n_fds
) {
2190 unsigned n_dont_close
= 0;
2191 int dont_close
[n_fds
+ 12];
2195 if (params
->stdin_fd
>= 0)
2196 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2197 if (params
->stdout_fd
>= 0)
2198 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2199 if (params
->stderr_fd
>= 0)
2200 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2203 dont_close
[n_dont_close
++] = socket_fd
;
2205 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2206 n_dont_close
+= n_fds
;
2210 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2214 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2216 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2219 if (user_lookup_fd
>= 0)
2220 dont_close
[n_dont_close
++] = user_lookup_fd
;
2222 return close_all_fds(dont_close
, n_dont_close
);
2225 static int send_user_lookup(
2233 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2234 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2237 if (user_lookup_fd
< 0)
2240 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2243 if (writev(user_lookup_fd
,
2245 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2246 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2247 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2253 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2260 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2265 if (!c
->working_directory_home
)
2269 /* Hardcode /root as home directory for UID 0 */
2274 r
= get_home_dir(buf
);
2282 static int exec_child(
2284 ExecCommand
*command
,
2285 const ExecContext
*context
,
2286 const ExecParameters
*params
,
2287 ExecRuntime
*runtime
,
2288 DynamicCreds
*dcreds
,
2293 unsigned n_storage_fds
,
2294 unsigned n_socket_fds
,
2298 char **error_message
) {
2300 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2301 _cleanup_free_
char *mac_selinux_context_net
= NULL
, *home_buffer
= NULL
;
2302 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2303 const char *username
= NULL
, *groupname
= NULL
;
2304 const char *home
= NULL
, *shell
= NULL
;
2305 dev_t journal_stream_dev
= 0;
2306 ino_t journal_stream_ino
= 0;
2307 bool needs_mount_namespace
;
2308 uid_t uid
= UID_INVALID
;
2309 gid_t gid
= GID_INVALID
;
2310 int i
, r
, ngids
= 0;
2312 ExecDirectoryType dt
;
2318 assert(exit_status
);
2319 assert(error_message
);
2320 /* We don't always set error_message, hence it must be initialized */
2321 assert(*error_message
== NULL
);
2323 rename_process_from_path(command
->path
);
2325 /* We reset exactly these signals, since they are the
2326 * only ones we set to SIG_IGN in the main daemon. All
2327 * others we leave untouched because we set them to
2328 * SIG_DFL or a valid handler initially, both of which
2329 * will be demoted to SIG_DFL. */
2330 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2331 SIGNALS_IGNORE
, -1);
2333 if (context
->ignore_sigpipe
)
2334 (void) ignore_signals(SIGPIPE
, -1);
2336 r
= reset_signal_mask();
2338 *exit_status
= EXIT_SIGNAL_MASK
;
2339 *error_message
= strdup("Failed to reset signal mask");
2340 /* If strdup fails, here and below, we will just print the generic error message. */
2344 if (params
->idle_pipe
)
2345 do_idle_pipe_dance(params
->idle_pipe
);
2347 /* Close sockets very early to make sure we don't
2348 * block init reexecution because it cannot bind its
2353 n_fds
= n_storage_fds
+ n_socket_fds
;
2354 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2356 *exit_status
= EXIT_FDS
;
2357 *error_message
= strdup("Failed to close remaining fds");
2361 if (!context
->same_pgrp
)
2363 *exit_status
= EXIT_SETSID
;
2367 exec_context_tty_reset(context
, params
);
2369 if (unit_shall_confirm_spawn(unit
)) {
2370 const char *vc
= params
->confirm_spawn
;
2371 _cleanup_free_
char *cmdline
= NULL
;
2373 cmdline
= exec_command_line(argv
);
2375 *exit_status
= EXIT_CONFIRM
;
2379 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2380 if (r
!= CONFIRM_EXECUTE
) {
2381 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2382 *exit_status
= EXIT_SUCCESS
;
2385 *exit_status
= EXIT_CONFIRM
;
2386 *error_message
= strdup("Execution cancelled");
2391 if (context
->dynamic_user
&& dcreds
) {
2393 /* Make sure we bypass our own NSS module for any NSS checks */
2394 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2395 *exit_status
= EXIT_USER
;
2396 *error_message
= strdup("Failed to update environment");
2400 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2402 *exit_status
= EXIT_USER
;
2403 *error_message
= strdup("Failed to update dynamic user credentials");
2407 if (!uid_is_valid(uid
)) {
2408 *exit_status
= EXIT_USER
;
2409 (void) asprintf(error_message
, "UID validation failed for \""UID_FMT
"\"", uid
);
2410 /* If asprintf fails, here and below, we will just print the generic error message. */
2414 if (!gid_is_valid(gid
)) {
2415 *exit_status
= EXIT_USER
;
2416 (void) asprintf(error_message
, "GID validation failed for \""GID_FMT
"\"", gid
);
2421 username
= dcreds
->user
->name
;
2424 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2426 *exit_status
= EXIT_USER
;
2427 *error_message
= strdup("Failed to determine user credentials");
2431 r
= get_fixed_group(context
, &groupname
, &gid
);
2433 *exit_status
= EXIT_GROUP
;
2434 *error_message
= strdup("Failed to determine group credentials");
2439 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2440 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2441 &supplementary_gids
, &ngids
);
2443 *exit_status
= EXIT_GROUP
;
2444 *error_message
= strdup("Failed to determine supplementary groups");
2448 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2450 *exit_status
= EXIT_USER
;
2451 *error_message
= strdup("Failed to send user credentials to PID1");
2455 user_lookup_fd
= safe_close(user_lookup_fd
);
2457 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2459 *exit_status
= EXIT_CHDIR
;
2460 *error_message
= strdup("Failed to determine $HOME for user");
2464 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2465 * must sure to drop O_NONBLOCK */
2467 (void) fd_nonblock(socket_fd
, false);
2469 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2471 *exit_status
= EXIT_STDIN
;
2472 *error_message
= strdup("Failed to set up stdin");
2476 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2478 *exit_status
= EXIT_STDOUT
;
2479 *error_message
= strdup("Failed to set up stdout");
2483 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2485 *exit_status
= EXIT_STDERR
;
2486 *error_message
= strdup("Failed to set up stderr");
2490 if (params
->cgroup_path
) {
2491 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2493 *exit_status
= EXIT_CGROUP
;
2494 (void) asprintf(error_message
, "Failed to attach to cgroup %s", params
->cgroup_path
);
2499 if (context
->oom_score_adjust_set
) {
2500 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2502 /* When we can't make this change due to EPERM, then
2503 * let's silently skip over it. User namespaces
2504 * prohibit write access to this file, and we
2505 * shouldn't trip up over that. */
2507 sprintf(t
, "%i", context
->oom_score_adjust
);
2508 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2509 if (r
== -EPERM
|| r
== -EACCES
) {
2511 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2514 *exit_status
= EXIT_OOM_ADJUST
;
2515 *error_message
= strdup("Failed to write /proc/self/oom_score_adj");
2520 if (context
->nice_set
)
2521 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2522 *exit_status
= EXIT_NICE
;
2526 if (context
->cpu_sched_set
) {
2527 struct sched_param param
= {
2528 .sched_priority
= context
->cpu_sched_priority
,
2531 r
= sched_setscheduler(0,
2532 context
->cpu_sched_policy
|
2533 (context
->cpu_sched_reset_on_fork
?
2534 SCHED_RESET_ON_FORK
: 0),
2537 *exit_status
= EXIT_SETSCHEDULER
;
2542 if (context
->cpuset
)
2543 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2544 *exit_status
= EXIT_CPUAFFINITY
;
2548 if (context
->ioprio_set
)
2549 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2550 *exit_status
= EXIT_IOPRIO
;
2554 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2555 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2556 *exit_status
= EXIT_TIMERSLACK
;
2560 if (context
->personality
!= PERSONALITY_INVALID
)
2561 if (personality(context
->personality
) < 0) {
2562 *exit_status
= EXIT_PERSONALITY
;
2566 if (context
->utmp_id
)
2567 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0),
2569 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2570 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2574 if (context
->user
) {
2575 r
= chown_terminal(STDIN_FILENO
, uid
);
2577 *exit_status
= EXIT_STDIN
;
2582 /* If delegation is enabled we'll pass ownership of the cgroup
2583 * (but only in systemd's own controller hierarchy!) to the
2584 * user of the new process. */
2585 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2586 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2588 *exit_status
= EXIT_CGROUP
;
2593 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2595 *exit_status
= EXIT_CGROUP
;
2600 for (dt
= 0; dt
< _EXEC_DIRECTORY_MAX
; dt
++) {
2601 r
= setup_exec_directory(context
, params
, uid
, gid
, MANAGER_IS_SYSTEM(unit
->manager
), dt
, exit_status
);
2606 r
= build_environment(
2618 *exit_status
= EXIT_MEMORY
;
2622 r
= build_pass_environment(context
, &pass_env
);
2624 *exit_status
= EXIT_MEMORY
;
2628 accum_env
= strv_env_merge(5,
2629 params
->environment
,
2632 context
->environment
,
2636 *exit_status
= EXIT_MEMORY
;
2639 accum_env
= strv_env_clean(accum_env
);
2641 (void) umask(context
->umask
);
2643 r
= setup_keyring(unit
, params
, uid
, gid
);
2645 *exit_status
= EXIT_KEYRING
;
2649 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2650 if (context
->pam_name
&& username
) {
2651 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2653 *exit_status
= EXIT_PAM
;
2659 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2660 r
= setup_netns(runtime
->netns_storage_socket
);
2662 *exit_status
= EXIT_NETWORK
;
2667 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2668 if (needs_mount_namespace
) {
2669 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
2671 *exit_status
= EXIT_NAMESPACE
;
2676 /* Apply just after mount namespace setup */
2677 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
2681 /* Drop groups as early as possbile */
2682 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2683 r
= enforce_groups(context
, gid
, supplementary_gids
, ngids
);
2685 *exit_status
= EXIT_GROUP
;
2691 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) &&
2692 mac_selinux_use() &&
2693 params
->selinux_context_net
&&
2695 !command
->privileged
) {
2697 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2699 *exit_status
= EXIT_SELINUX_CONTEXT
;
2705 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2706 r
= setup_private_users(uid
, gid
);
2708 *exit_status
= EXIT_USER
;
2713 /* We repeat the fd closing here, to make sure that
2714 * nothing is leaked from the PAM modules. Note that
2715 * we are more aggressive this time since socket_fd
2716 * and the netns fds we don't need anymore. The custom
2717 * endpoint fd was needed to upload the policy and can
2718 * now be closed as well. */
2719 r
= close_all_fds(fds
, n_fds
);
2721 r
= shift_fds(fds
, n_fds
);
2723 r
= flags_fds(fds
, n_storage_fds
, n_socket_fds
, context
->non_blocking
);
2725 *exit_status
= EXIT_FDS
;
2729 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2731 int secure_bits
= context
->secure_bits
;
2733 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2735 if (!context
->rlimit
[i
])
2738 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2740 *exit_status
= EXIT_LIMITS
;
2745 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2746 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2747 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2748 *exit_status
= EXIT_LIMITS
;
2753 if (!cap_test_all(context
->capability_bounding_set
)) {
2754 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2756 *exit_status
= EXIT_CAPABILITIES
;
2757 *error_message
= strdup("Failed to drop capabilities");
2762 /* This is done before enforce_user, but ambient set
2763 * does not survive over setresuid() if keep_caps is not set. */
2764 if (context
->capability_ambient_set
!= 0) {
2765 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2767 *exit_status
= EXIT_CAPABILITIES
;
2768 *error_message
= strdup("Failed to apply ambient capabilities (before UID change)");
2773 if (context
->user
) {
2774 r
= enforce_user(context
, uid
);
2776 *exit_status
= EXIT_USER
;
2777 (void) asprintf(error_message
, "Failed to change UID to "UID_FMT
, uid
);
2780 if (context
->capability_ambient_set
!= 0) {
2782 /* Fix the ambient capabilities after user change. */
2783 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2785 *exit_status
= EXIT_CAPABILITIES
;
2786 *error_message
= strdup("Failed to apply ambient capabilities (after UID change)");
2790 /* If we were asked to change user and ambient capabilities
2791 * were requested, we had to add keep-caps to the securebits
2792 * so that we would maintain the inherited capability set
2793 * through the setresuid(). Make sure that the bit is added
2794 * also to the context secure_bits so that we don't try to
2795 * drop the bit away next. */
2797 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2801 /* Apply the MAC contexts late, but before seccomp syscall filtering, as those should really be last to
2802 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
2803 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
2804 * are restricted. */
2807 if (mac_selinux_use()) {
2808 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2811 r
= setexeccon(exec_context
);
2813 *exit_status
= EXIT_SELINUX_CONTEXT
;
2814 (void) asprintf(error_message
, "Failed to set SELinux context to %s", exec_context
);
2821 r
= setup_smack(context
, command
);
2823 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2824 *error_message
= strdup("Failed to set SMACK process label");
2828 #ifdef HAVE_APPARMOR
2829 if (context
->apparmor_profile
&& mac_apparmor_use()) {
2830 r
= aa_change_onexec(context
->apparmor_profile
);
2831 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2832 *exit_status
= EXIT_APPARMOR_PROFILE
;
2833 (void) asprintf(error_message
,
2834 "Failed to prepare AppArmor profile change to %s",
2835 context
->apparmor_profile
);
2841 /* PR_GET_SECUREBITS is not privileged, while
2842 * PR_SET_SECUREBITS is. So to suppress
2843 * potential EPERMs we'll try not to call
2844 * PR_SET_SECUREBITS unless necessary. */
2845 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2846 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2847 *exit_status
= EXIT_SECUREBITS
;
2848 *error_message
= strdup("Failed to set secure bits");
2852 if (context_has_no_new_privileges(context
))
2853 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2854 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2855 *error_message
= strdup("Failed to disable new privileges");
2860 r
= apply_address_families(unit
, context
);
2862 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2863 *error_message
= strdup("Failed to restrict address families");
2867 r
= apply_memory_deny_write_execute(unit
, context
);
2869 *exit_status
= EXIT_SECCOMP
;
2870 *error_message
= strdup("Failed to disable writing to executable memory");
2874 r
= apply_restrict_realtime(unit
, context
);
2876 *exit_status
= EXIT_SECCOMP
;
2877 *error_message
= strdup("Failed to apply realtime restrictions");
2881 r
= apply_restrict_namespaces(unit
, context
);
2883 *exit_status
= EXIT_SECCOMP
;
2884 *error_message
= strdup("Failed to apply namespace restrictions");
2888 r
= apply_protect_sysctl(unit
, context
);
2890 *exit_status
= EXIT_SECCOMP
;
2891 *error_message
= strdup("Failed to apply sysctl restrictions");
2895 r
= apply_protect_kernel_modules(unit
, context
);
2897 *exit_status
= EXIT_SECCOMP
;
2898 *error_message
= strdup("Failed to apply module loading restrictions");
2902 r
= apply_private_devices(unit
, context
);
2904 *exit_status
= EXIT_SECCOMP
;
2905 *error_message
= strdup("Failed to set up private devices");
2909 r
= apply_syscall_archs(unit
, context
);
2911 *exit_status
= EXIT_SECCOMP
;
2912 *error_message
= strdup("Failed to apply syscall architecture restrictions");
2916 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
2917 * by the filter as little as possible. */
2918 r
= apply_syscall_filter(unit
, context
);
2920 *exit_status
= EXIT_SECCOMP
;
2921 *error_message
= strdup("Failed to apply syscall filters");
2927 final_argv
= replace_env_argv(argv
, accum_env
);
2929 *exit_status
= EXIT_MEMORY
;
2930 *error_message
= strdup("Failed to prepare process arguments");
2934 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2935 _cleanup_free_
char *line
;
2937 line
= exec_command_line(final_argv
);
2940 log_struct(LOG_DEBUG
,
2941 "EXECUTABLE=%s", command
->path
,
2942 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2949 execve(command
->path
, final_argv
, accum_env
);
2950 *exit_status
= EXIT_EXEC
;
2954 int exec_spawn(Unit
*unit
,
2955 ExecCommand
*command
,
2956 const ExecContext
*context
,
2957 const ExecParameters
*params
,
2958 ExecRuntime
*runtime
,
2959 DynamicCreds
*dcreds
,
2962 _cleanup_strv_free_
char **files_env
= NULL
;
2964 unsigned n_storage_fds
= 0, n_socket_fds
= 0;
2965 _cleanup_free_
char *line
= NULL
;
2967 int named_iofds
[3] = { -1, -1, -1 };
2976 assert(params
->fds
|| (params
->n_storage_fds
+ params
->n_socket_fds
<= 0));
2978 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2979 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2980 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
2982 if (params
->n_socket_fds
> 1) {
2983 log_unit_error(unit
, "Got more than one socket.");
2987 if (params
->n_socket_fds
== 0) {
2988 log_unit_error(unit
, "Got no socket.");
2992 socket_fd
= params
->fds
[0];
2996 n_storage_fds
= params
->n_storage_fds
;
2997 n_socket_fds
= params
->n_socket_fds
;
3000 r
= exec_context_named_iofds(unit
, context
, params
, named_iofds
);
3002 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3004 r
= exec_context_load_environment(unit
, context
, &files_env
);
3006 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3008 argv
= params
->argv
?: command
->argv
;
3009 line
= exec_command_line(argv
);
3013 log_struct(LOG_DEBUG
,
3014 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3015 "EXECUTABLE=%s", command
->path
,
3020 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3024 _cleanup_free_
char *error_message
= NULL
;
3026 r
= exec_child(unit
,
3039 unit
->manager
->user_lookup_fds
[1],
3045 log_struct_errno(LOG_ERR
, r
,
3046 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3048 LOG_UNIT_MESSAGE(unit
, "%s: %m",
3050 "EXECUTABLE=%s", command
->path
,
3052 else if (r
== -ENOENT
&& command
->ignore
)
3053 log_struct_errno(LOG_INFO
, r
,
3054 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3056 LOG_UNIT_MESSAGE(unit
, "Skipped spawning %s: %m",
3058 "EXECUTABLE=%s", command
->path
,
3061 log_struct_errno(LOG_ERR
, r
,
3062 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3064 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3065 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3067 "EXECUTABLE=%s", command
->path
,
3074 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3076 /* We add the new process to the cgroup both in the child (so
3077 * that we can be sure that no user code is ever executed
3078 * outside of the cgroup) and in the parent (so that we can be
3079 * sure that when we kill the cgroup the process will be
3081 if (params
->cgroup_path
)
3082 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3084 exec_status_start(&command
->exec_status
, pid
);
3090 void exec_context_init(ExecContext
*c
) {
3091 ExecDirectoryType i
;
3096 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3097 c
->cpu_sched_policy
= SCHED_OTHER
;
3098 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3099 c
->syslog_level_prefix
= true;
3100 c
->ignore_sigpipe
= true;
3101 c
->timer_slack_nsec
= NSEC_INFINITY
;
3102 c
->personality
= PERSONALITY_INVALID
;
3103 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++)
3104 c
->directories
[i
].mode
= 0755;
3105 c
->capability_bounding_set
= CAP_ALL
;
3106 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
3109 void exec_context_done(ExecContext
*c
) {
3111 ExecDirectoryType i
;
3115 c
->environment
= strv_free(c
->environment
);
3116 c
->environment_files
= strv_free(c
->environment_files
);
3117 c
->pass_environment
= strv_free(c
->pass_environment
);
3119 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
3120 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
3122 for (l
= 0; l
< 3; l
++)
3123 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3125 c
->working_directory
= mfree(c
->working_directory
);
3126 c
->root_directory
= mfree(c
->root_directory
);
3127 c
->root_image
= mfree(c
->root_image
);
3128 c
->tty_path
= mfree(c
->tty_path
);
3129 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3130 c
->user
= mfree(c
->user
);
3131 c
->group
= mfree(c
->group
);
3133 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3135 c
->pam_name
= mfree(c
->pam_name
);
3137 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3138 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3139 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3141 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3144 CPU_FREE(c
->cpuset
);
3146 c
->utmp_id
= mfree(c
->utmp_id
);
3147 c
->selinux_context
= mfree(c
->selinux_context
);
3148 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3149 c
->smack_process_label
= mfree(c
->smack_process_label
);
3151 c
->syscall_filter
= set_free(c
->syscall_filter
);
3152 c
->syscall_archs
= set_free(c
->syscall_archs
);
3153 c
->address_families
= set_free(c
->address_families
);
3155 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++)
3156 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3159 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
3164 if (!runtime_prefix
)
3167 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3168 _cleanup_free_
char *p
;
3170 p
= strjoin(runtime_prefix
, "/", *i
);
3174 /* We execute this synchronously, since we need to be
3175 * sure this is gone when we start the service
3177 (void) rm_rf(p
, REMOVE_ROOT
);
3183 void exec_command_done(ExecCommand
*c
) {
3186 c
->path
= mfree(c
->path
);
3188 c
->argv
= strv_free(c
->argv
);
3191 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3194 for (i
= 0; i
< n
; i
++)
3195 exec_command_done(c
+i
);
3198 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3202 LIST_REMOVE(command
, c
, i
);
3203 exec_command_done(i
);
3210 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3213 for (i
= 0; i
< n
; i
++)
3214 c
[i
] = exec_command_free_list(c
[i
]);
3217 typedef struct InvalidEnvInfo
{
3222 static void invalid_env(const char *p
, void *userdata
) {
3223 InvalidEnvInfo
*info
= userdata
;
3225 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3228 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3233 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3235 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3237 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3239 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3241 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3243 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3249 int exec_context_named_iofds(Unit
*unit
, const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3250 unsigned i
, targets
;
3251 const char* stdio_fdname
[3];
3257 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3258 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3259 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3261 for (i
= 0; i
< 3; i
++)
3262 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3264 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3266 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3267 if (named_iofds
[STDIN_FILENO
] < 0 &&
3268 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3269 stdio_fdname
[STDIN_FILENO
] &&
3270 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3272 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3275 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3276 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3277 stdio_fdname
[STDOUT_FILENO
] &&
3278 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3280 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3283 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3284 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3285 stdio_fdname
[STDERR_FILENO
] &&
3286 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3288 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3292 return targets
== 0 ? 0 : -ENOENT
;
3295 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
3296 char **i
, **r
= NULL
;
3301 STRV_FOREACH(i
, c
->environment_files
) {
3305 bool ignore
= false;
3307 _cleanup_globfree_ glob_t pglob
= {};
3316 if (!path_is_absolute(fn
)) {
3324 /* Filename supports globbing, take all matching files */
3325 k
= safe_glob(fn
, 0, &pglob
);
3334 /* When we don't match anything, -ENOENT should be returned */
3335 assert(pglob
.gl_pathc
> 0);
3337 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
3338 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3346 /* Log invalid environment variables with filename */
3348 InvalidEnvInfo info
= {
3350 .path
= pglob
.gl_pathv
[n
]
3353 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3361 m
= strv_env_merge(2, r
, p
);
3377 static bool tty_may_match_dev_console(const char *tty
) {
3378 _cleanup_free_
char *active
= NULL
;
3384 if (startswith(tty
, "/dev/"))
3387 /* trivial identity? */
3388 if (streq(tty
, "console"))
3391 console
= resolve_dev_console(&active
);
3392 /* if we could not resolve, assume it may */
3396 /* "tty0" means the active VC, so it may be the same sometimes */
3397 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3400 bool exec_context_may_touch_console(ExecContext
*ec
) {
3402 return (ec
->tty_reset
||
3404 ec
->tty_vt_disallocate
||
3405 is_terminal_input(ec
->std_input
) ||
3406 is_terminal_output(ec
->std_output
) ||
3407 is_terminal_output(ec
->std_error
)) &&
3408 tty_may_match_dev_console(exec_context_tty_path(ec
));
3411 static void strv_fprintf(FILE *f
, char **l
) {
3417 fprintf(f
, " %s", *g
);
3420 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3423 ExecDirectoryType dt
;
3429 prefix
= strempty(prefix
);
3433 "%sWorkingDirectory: %s\n"
3434 "%sRootDirectory: %s\n"
3435 "%sNonBlocking: %s\n"
3436 "%sPrivateTmp: %s\n"
3437 "%sPrivateDevices: %s\n"
3438 "%sProtectKernelTunables: %s\n"
3439 "%sProtectKernelModules: %s\n"
3440 "%sProtectControlGroups: %s\n"
3441 "%sPrivateNetwork: %s\n"
3442 "%sPrivateUsers: %s\n"
3443 "%sProtectHome: %s\n"
3444 "%sProtectSystem: %s\n"
3445 "%sMountAPIVFS: %s\n"
3446 "%sIgnoreSIGPIPE: %s\n"
3447 "%sMemoryDenyWriteExecute: %s\n"
3448 "%sRestrictRealtime: %s\n",
3450 prefix
, c
->working_directory
? c
->working_directory
: "/",
3451 prefix
, c
->root_directory
? c
->root_directory
: "/",
3452 prefix
, yes_no(c
->non_blocking
),
3453 prefix
, yes_no(c
->private_tmp
),
3454 prefix
, yes_no(c
->private_devices
),
3455 prefix
, yes_no(c
->protect_kernel_tunables
),
3456 prefix
, yes_no(c
->protect_kernel_modules
),
3457 prefix
, yes_no(c
->protect_control_groups
),
3458 prefix
, yes_no(c
->private_network
),
3459 prefix
, yes_no(c
->private_users
),
3460 prefix
, protect_home_to_string(c
->protect_home
),
3461 prefix
, protect_system_to_string(c
->protect_system
),
3462 prefix
, yes_no(c
->mount_apivfs
),
3463 prefix
, yes_no(c
->ignore_sigpipe
),
3464 prefix
, yes_no(c
->memory_deny_write_execute
),
3465 prefix
, yes_no(c
->restrict_realtime
));
3468 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3470 STRV_FOREACH(e
, c
->environment
)
3471 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3473 STRV_FOREACH(e
, c
->environment_files
)
3474 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3476 STRV_FOREACH(e
, c
->pass_environment
)
3477 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3479 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
3481 for (dt
= 0; dt
< _EXEC_DIRECTORY_MAX
; dt
++) {
3482 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
3484 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
3485 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
3493 if (c
->oom_score_adjust_set
)
3495 "%sOOMScoreAdjust: %i\n",
3496 prefix
, c
->oom_score_adjust
);
3498 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3500 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3501 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3502 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3503 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3506 if (c
->ioprio_set
) {
3507 _cleanup_free_
char *class_str
= NULL
;
3509 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3511 "%sIOSchedulingClass: %s\n"
3512 "%sIOPriority: %i\n",
3513 prefix
, strna(class_str
),
3514 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
3517 if (c
->cpu_sched_set
) {
3518 _cleanup_free_
char *policy_str
= NULL
;
3520 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3522 "%sCPUSchedulingPolicy: %s\n"
3523 "%sCPUSchedulingPriority: %i\n"
3524 "%sCPUSchedulingResetOnFork: %s\n",
3525 prefix
, strna(policy_str
),
3526 prefix
, c
->cpu_sched_priority
,
3527 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3531 fprintf(f
, "%sCPUAffinity:", prefix
);
3532 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3533 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3534 fprintf(f
, " %u", i
);
3538 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3539 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3542 "%sStandardInput: %s\n"
3543 "%sStandardOutput: %s\n"
3544 "%sStandardError: %s\n",
3545 prefix
, exec_input_to_string(c
->std_input
),
3546 prefix
, exec_output_to_string(c
->std_output
),
3547 prefix
, exec_output_to_string(c
->std_error
));
3553 "%sTTYVHangup: %s\n"
3554 "%sTTYVTDisallocate: %s\n",
3555 prefix
, c
->tty_path
,
3556 prefix
, yes_no(c
->tty_reset
),
3557 prefix
, yes_no(c
->tty_vhangup
),
3558 prefix
, yes_no(c
->tty_vt_disallocate
));
3560 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3561 c
->std_output
== EXEC_OUTPUT_KMSG
||
3562 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3563 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3564 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3565 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3566 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3567 c
->std_error
== EXEC_OUTPUT_KMSG
||
3568 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3569 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3570 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3571 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3573 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3575 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3576 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3579 "%sSyslogFacility: %s\n"
3580 "%sSyslogLevel: %s\n",
3581 prefix
, strna(fac_str
),
3582 prefix
, strna(lvl_str
));
3586 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
3588 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
3589 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
3590 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
3591 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
3592 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
3593 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
3595 if (c
->capability_bounding_set
!= CAP_ALL
) {
3597 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
3599 for (l
= 0; l
<= cap_last_cap(); l
++)
3600 if (c
->capability_bounding_set
& (UINT64_C(1) << l
))
3601 fprintf(f
, " %s", strna(capability_to_name(l
)));
3606 if (c
->capability_ambient_set
!= 0) {
3608 fprintf(f
, "%sAmbientCapabilities:", prefix
);
3610 for (l
= 0; l
<= cap_last_cap(); l
++)
3611 if (c
->capability_ambient_set
& (UINT64_C(1) << l
))
3612 fprintf(f
, " %s", strna(capability_to_name(l
)));
3618 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3620 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3622 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3624 if (strv_length(c
->supplementary_groups
) > 0) {
3625 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3626 strv_fprintf(f
, c
->supplementary_groups
);
3631 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3633 if (strv_length(c
->read_write_paths
) > 0) {
3634 fprintf(f
, "%sReadWritePaths:", prefix
);
3635 strv_fprintf(f
, c
->read_write_paths
);
3639 if (strv_length(c
->read_only_paths
) > 0) {
3640 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3641 strv_fprintf(f
, c
->read_only_paths
);
3645 if (strv_length(c
->inaccessible_paths
) > 0) {
3646 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3647 strv_fprintf(f
, c
->inaccessible_paths
);
3651 if (c
->n_bind_mounts
> 0)
3652 for (i
= 0; i
< c
->n_bind_mounts
; i
++) {
3653 fprintf(f
, "%s%s: %s:%s:%s\n", prefix
,
3654 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
3655 c
->bind_mounts
[i
].source
,
3656 c
->bind_mounts
[i
].destination
,
3657 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
3662 "%sUtmpIdentifier: %s\n",
3663 prefix
, c
->utmp_id
);
3665 if (c
->selinux_context
)
3667 "%sSELinuxContext: %s%s\n",
3668 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3670 if (c
->apparmor_profile
)
3672 "%sAppArmorProfile: %s%s\n",
3673 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3675 if (c
->smack_process_label
)
3677 "%sSmackProcessLabel: %s%s\n",
3678 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
3680 if (c
->personality
!= PERSONALITY_INVALID
)
3682 "%sPersonality: %s\n",
3683 prefix
, strna(personality_to_string(c
->personality
)));
3685 if (c
->syscall_filter
) {
3693 "%sSystemCallFilter: ",
3696 if (!c
->syscall_whitelist
)
3700 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3701 _cleanup_free_
char *name
= NULL
;
3708 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3709 fputs(strna(name
), f
);
3716 if (c
->syscall_archs
) {
3723 "%sSystemCallArchitectures:",
3727 SET_FOREACH(id
, c
->syscall_archs
, j
)
3728 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3733 if (exec_context_restrict_namespaces_set(c
)) {
3734 _cleanup_free_
char *s
= NULL
;
3736 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
3738 fprintf(f
, "%sRestrictNamespaces: %s\n",
3742 if (c
->syscall_errno
> 0)
3744 "%sSystemCallErrorNumber: %s\n",
3745 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3747 if (c
->apparmor_profile
)
3749 "%sAppArmorProfile: %s%s\n",
3750 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3753 bool exec_context_maintains_privileges(ExecContext
*c
) {
3756 /* Returns true if the process forked off would run under
3757 * an unchanged UID or as root. */
3762 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3768 int exec_context_get_effective_ioprio(ExecContext
*c
) {
3776 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
3778 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
3783 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3788 dual_timestamp_get(&s
->start_timestamp
);
3791 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3794 if (s
->pid
&& s
->pid
!= pid
)
3798 dual_timestamp_get(&s
->exit_timestamp
);
3804 if (context
->utmp_id
)
3805 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3807 exec_context_tty_reset(context
, NULL
);
3811 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3812 char buf
[FORMAT_TIMESTAMP_MAX
];
3820 prefix
= strempty(prefix
);
3823 "%sPID: "PID_FMT
"\n",
3826 if (dual_timestamp_is_set(&s
->start_timestamp
))
3828 "%sStart Timestamp: %s\n",
3829 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3831 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3833 "%sExit Timestamp: %s\n"
3835 "%sExit Status: %i\n",
3836 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3837 prefix
, sigchld_code_to_string(s
->code
),
3841 char *exec_command_line(char **argv
) {
3849 STRV_FOREACH(a
, argv
)
3857 STRV_FOREACH(a
, argv
) {
3864 if (strpbrk(*a
, WHITESPACE
)) {
3875 /* FIXME: this doesn't really handle arguments that have
3876 * spaces and ticks in them */
3881 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3882 _cleanup_free_
char *cmd
= NULL
;
3883 const char *prefix2
;
3888 prefix
= strempty(prefix
);
3889 prefix2
= strjoina(prefix
, "\t");
3891 cmd
= exec_command_line(c
->argv
);
3893 "%sCommand Line: %s\n",
3894 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3896 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3899 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3902 prefix
= strempty(prefix
);
3904 LIST_FOREACH(command
, c
, c
)
3905 exec_command_dump(c
, f
, prefix
);
3908 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3915 /* It's kind of important, that we keep the order here */
3916 LIST_FIND_TAIL(command
, *l
, end
);
3917 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3922 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3930 l
= strv_new_ap(path
, ap
);
3951 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3952 _cleanup_strv_free_
char **l
= NULL
;
3960 l
= strv_new_ap(path
, ap
);
3966 r
= strv_extend_strv(&c
->argv
, l
, false);
3974 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3979 *rt
= new0(ExecRuntime
, 1);
3984 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3989 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
3999 if (!c
->private_network
&& !c
->private_tmp
)
4002 r
= exec_runtime_allocate(rt
);
4006 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
4007 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
4011 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
4012 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
4020 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
4022 assert(r
->n_ref
> 0);
4028 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
4033 assert(r
->n_ref
> 0);
4040 free(r
->var_tmp_dir
);
4041 safe_close_pair(r
->netns_storage_socket
);
4045 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
4054 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
4056 if (rt
->var_tmp_dir
)
4057 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
4059 if (rt
->netns_storage_socket
[0] >= 0) {
4062 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4066 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
4069 if (rt
->netns_storage_socket
[1] >= 0) {
4072 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
4076 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
4082 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
4089 if (streq(key
, "tmp-dir")) {
4092 r
= exec_runtime_allocate(rt
);
4096 copy
= strdup(value
);
4100 free((*rt
)->tmp_dir
);
4101 (*rt
)->tmp_dir
= copy
;
4103 } else if (streq(key
, "var-tmp-dir")) {
4106 r
= exec_runtime_allocate(rt
);
4110 copy
= strdup(value
);
4114 free((*rt
)->var_tmp_dir
);
4115 (*rt
)->var_tmp_dir
= copy
;
4117 } else if (streq(key
, "netns-socket-0")) {
4120 r
= exec_runtime_allocate(rt
);
4124 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4125 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4127 safe_close((*rt
)->netns_storage_socket
[0]);
4128 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4130 } else if (streq(key
, "netns-socket-1")) {
4133 r
= exec_runtime_allocate(rt
);
4137 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4138 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4140 safe_close((*rt
)->netns_storage_socket
[1]);
4141 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
4149 static void *remove_tmpdir_thread(void *p
) {
4150 _cleanup_free_
char *path
= p
;
4152 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4156 void exec_runtime_destroy(ExecRuntime
*rt
) {
4162 /* If there are multiple users of this, let's leave the stuff around */
4167 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4169 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4171 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4178 if (rt
->var_tmp_dir
) {
4179 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4181 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4183 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4184 free(rt
->var_tmp_dir
);
4187 rt
->var_tmp_dir
= NULL
;
4190 safe_close_pair(rt
->netns_storage_socket
);
4193 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4194 [EXEC_INPUT_NULL
] = "null",
4195 [EXEC_INPUT_TTY
] = "tty",
4196 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4197 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4198 [EXEC_INPUT_SOCKET
] = "socket",
4199 [EXEC_INPUT_NAMED_FD
] = "fd",
4202 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4204 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4205 [EXEC_OUTPUT_INHERIT
] = "inherit",
4206 [EXEC_OUTPUT_NULL
] = "null",
4207 [EXEC_OUTPUT_TTY
] = "tty",
4208 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4209 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4210 [EXEC_OUTPUT_KMSG
] = "kmsg",
4211 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4212 [EXEC_OUTPUT_JOURNAL
] = "journal",
4213 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4214 [EXEC_OUTPUT_SOCKET
] = "socket",
4215 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4218 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4220 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4221 [EXEC_UTMP_INIT
] = "init",
4222 [EXEC_UTMP_LOGIN
] = "login",
4223 [EXEC_UTMP_USER
] = "user",
4226 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
4228 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
4229 [EXEC_PRESERVE_NO
] = "no",
4230 [EXEC_PRESERVE_YES
] = "yes",
4231 [EXEC_PRESERVE_RESTART
] = "restart",
4234 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
4236 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_MAX
] = {
4237 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
4238 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
4239 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
4240 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
4241 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
4244 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);