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 "securebits-util.h"
94 #include "selinux-util.h"
95 #include "signal-util.h"
96 #include "smack-util.h"
98 #include "string-table.h"
99 #include "string-util.h"
101 #include "syslog-util.h"
102 #include "terminal-util.h"
104 #include "user-util.h"
106 #include "utmp-wtmp.h"
108 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
109 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
111 /* This assumes there is a 'tty' group */
112 #define TTY_MODE 0620
114 #define SNDBUF_SIZE (8*1024*1024)
116 static int shift_fds(int fds
[], unsigned n_fds
) {
117 int start
, restart_from
;
122 /* Modifies the fds array! (sorts it) */
132 for (i
= start
; i
< (int) n_fds
; i
++) {
135 /* Already at right index? */
139 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
146 /* Hmm, the fd we wanted isn't free? Then
147 * let's remember that and try again from here */
148 if (nfd
!= i
+3 && restart_from
< 0)
152 if (restart_from
< 0)
155 start
= restart_from
;
161 static int flags_fds(const int fds
[], unsigned n_storage_fds
, unsigned n_socket_fds
, bool nonblock
) {
165 n_fds
= n_storage_fds
+ n_socket_fds
;
171 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
172 * O_NONBLOCK only applies to socket activation though. */
174 for (i
= 0; i
< n_fds
; i
++) {
176 if (i
< n_socket_fds
) {
177 r
= fd_nonblock(fds
[i
], nonblock
);
182 /* We unconditionally drop FD_CLOEXEC from the fds,
183 * since after all we want to pass these fds to our
186 r
= fd_cloexec(fds
[i
], false);
194 static const char *exec_context_tty_path(const ExecContext
*context
) {
197 if (context
->stdio_as_fds
)
200 if (context
->tty_path
)
201 return context
->tty_path
;
203 return "/dev/console";
206 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
211 path
= exec_context_tty_path(context
);
213 if (context
->tty_vhangup
) {
214 if (p
&& p
->stdin_fd
>= 0)
215 (void) terminal_vhangup_fd(p
->stdin_fd
);
217 (void) terminal_vhangup(path
);
220 if (context
->tty_reset
) {
221 if (p
&& p
->stdin_fd
>= 0)
222 (void) reset_terminal_fd(p
->stdin_fd
, true);
224 (void) reset_terminal(path
);
227 if (context
->tty_vt_disallocate
&& path
)
228 (void) vt_disallocate(path
);
231 static bool is_terminal_input(ExecInput i
) {
234 EXEC_INPUT_TTY_FORCE
,
235 EXEC_INPUT_TTY_FAIL
);
238 static bool is_terminal_output(ExecOutput o
) {
241 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
242 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
243 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
246 static bool exec_context_needs_term(const ExecContext
*c
) {
249 /* Return true if the execution context suggests we should set $TERM to something useful. */
251 if (is_terminal_input(c
->std_input
))
254 if (is_terminal_output(c
->std_output
))
257 if (is_terminal_output(c
->std_error
))
260 return !!c
->tty_path
;
263 static int open_null_as(int flags
, int nfd
) {
268 fd
= open("/dev/null", flags
|O_NOCTTY
);
273 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
281 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
282 static const union sockaddr_union sa
= {
283 .un
.sun_family
= AF_UNIX
,
284 .un
.sun_path
= "/run/systemd/journal/stdout",
286 uid_t olduid
= UID_INVALID
;
287 gid_t oldgid
= GID_INVALID
;
290 if (gid_is_valid(gid
)) {
293 if (setegid(gid
) < 0)
297 if (uid_is_valid(uid
)) {
300 if (seteuid(uid
) < 0) {
306 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
308 /* If we fail to restore the uid or gid, things will likely
309 fail later on. This should only happen if an LSM interferes. */
311 if (uid_is_valid(uid
))
312 (void) seteuid(olduid
);
315 if (gid_is_valid(gid
))
316 (void) setegid(oldgid
);
321 static int connect_logger_as(
323 const ExecContext
*context
,
333 assert(output
< _EXEC_OUTPUT_MAX
);
337 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
341 r
= connect_journal_socket(fd
, uid
, gid
);
345 if (shutdown(fd
, SHUT_RD
) < 0) {
350 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
360 context
->syslog_identifier
?: ident
,
361 MANAGER_IS_SYSTEM(unit
->manager
) ? unit
->id
: "",
362 context
->syslog_priority
,
363 !!context
->syslog_level_prefix
,
364 output
== EXEC_OUTPUT_SYSLOG
|| output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
365 output
== EXEC_OUTPUT_KMSG
|| output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
,
366 is_terminal_output(output
));
371 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
376 static int open_terminal_as(const char *path
, mode_t mode
, int nfd
) {
382 fd
= open_terminal(path
, mode
| O_NOCTTY
);
387 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
395 static int fixup_input(ExecInput std_input
, int socket_fd
, bool apply_tty_stdin
) {
397 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
398 return EXEC_INPUT_NULL
;
400 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
401 return EXEC_INPUT_NULL
;
406 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
408 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
409 return EXEC_OUTPUT_INHERIT
;
414 static int setup_input(
415 const ExecContext
*context
,
416 const ExecParameters
*params
,
418 int named_iofds
[3]) {
425 if (params
->stdin_fd
>= 0) {
426 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
429 /* Try to make this the controlling tty, if it is a tty, and reset it */
430 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
431 (void) reset_terminal_fd(STDIN_FILENO
, true);
436 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
440 case EXEC_INPUT_NULL
:
441 return open_null_as(O_RDONLY
, STDIN_FILENO
);
444 case EXEC_INPUT_TTY_FORCE
:
445 case EXEC_INPUT_TTY_FAIL
: {
448 fd
= acquire_terminal(exec_context_tty_path(context
),
449 i
== EXEC_INPUT_TTY_FAIL
,
450 i
== EXEC_INPUT_TTY_FORCE
,
456 if (fd
!= STDIN_FILENO
) {
457 r
= dup2(fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
465 case EXEC_INPUT_SOCKET
:
466 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
468 case EXEC_INPUT_NAMED_FD
:
469 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
470 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
473 assert_not_reached("Unknown input type");
477 static int setup_output(
479 const ExecContext
*context
,
480 const ExecParameters
*params
,
487 dev_t
*journal_stream_dev
,
488 ino_t
*journal_stream_ino
) {
498 assert(journal_stream_dev
);
499 assert(journal_stream_ino
);
501 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
503 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
506 return STDOUT_FILENO
;
509 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
510 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
513 return STDERR_FILENO
;
516 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
517 o
= fixup_output(context
->std_output
, socket_fd
);
519 if (fileno
== STDERR_FILENO
) {
521 e
= fixup_output(context
->std_error
, socket_fd
);
523 /* This expects the input and output are already set up */
525 /* Don't change the stderr file descriptor if we inherit all
526 * the way and are not on a tty */
527 if (e
== EXEC_OUTPUT_INHERIT
&&
528 o
== EXEC_OUTPUT_INHERIT
&&
529 i
== EXEC_INPUT_NULL
&&
530 !is_terminal_input(context
->std_input
) &&
534 /* Duplicate from stdout if possible */
535 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
536 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
540 } else if (o
== EXEC_OUTPUT_INHERIT
) {
541 /* If input got downgraded, inherit the original value */
542 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
543 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
545 /* If the input is connected to anything that's not a /dev/null, inherit that... */
546 if (i
!= EXEC_INPUT_NULL
)
547 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
549 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
553 /* We need to open /dev/null here anew, to get the right access mode. */
554 return open_null_as(O_WRONLY
, fileno
);
559 case EXEC_OUTPUT_NULL
:
560 return open_null_as(O_WRONLY
, fileno
);
562 case EXEC_OUTPUT_TTY
:
563 if (is_terminal_input(i
))
564 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
566 /* We don't reset the terminal if this is just about output */
567 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
569 case EXEC_OUTPUT_SYSLOG
:
570 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
571 case EXEC_OUTPUT_KMSG
:
572 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
573 case EXEC_OUTPUT_JOURNAL
:
574 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
575 r
= connect_logger_as(unit
, context
, o
, ident
, fileno
, uid
, gid
);
577 log_unit_error_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
578 r
= open_null_as(O_WRONLY
, fileno
);
582 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
583 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
584 * services to detect whether they are connected to the journal or not. */
586 if (fstat(fileno
, &st
) >= 0) {
587 *journal_stream_dev
= st
.st_dev
;
588 *journal_stream_ino
= st
.st_ino
;
593 case EXEC_OUTPUT_SOCKET
:
594 assert(socket_fd
>= 0);
595 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
597 case EXEC_OUTPUT_NAMED_FD
:
598 (void) fd_nonblock(named_iofds
[fileno
], false);
599 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
602 assert_not_reached("Unknown error type");
606 static int chown_terminal(int fd
, uid_t uid
) {
611 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
615 /* This might fail. What matters are the results. */
616 (void) fchown(fd
, uid
, -1);
617 (void) fchmod(fd
, TTY_MODE
);
619 if (fstat(fd
, &st
) < 0)
622 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
628 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
629 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
632 assert(_saved_stdin
);
633 assert(_saved_stdout
);
635 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
639 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
640 if (saved_stdout
< 0)
643 fd
= acquire_terminal(vc
, false, false, false, DEFAULT_CONFIRM_USEC
);
647 r
= chown_terminal(fd
, getuid());
651 r
= reset_terminal_fd(fd
, true);
655 if (dup2(fd
, STDIN_FILENO
) < 0)
658 if (dup2(fd
, STDOUT_FILENO
) < 0)
665 *_saved_stdin
= saved_stdin
;
666 *_saved_stdout
= saved_stdout
;
668 saved_stdin
= saved_stdout
= -1;
673 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
676 if (err
== -ETIMEDOUT
)
677 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
680 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
684 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
685 _cleanup_close_
int fd
= -1;
689 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
693 write_confirm_error_fd(err
, fd
, u
);
696 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
700 assert(saved_stdout
);
704 if (*saved_stdin
>= 0)
705 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
708 if (*saved_stdout
>= 0)
709 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
712 *saved_stdin
= safe_close(*saved_stdin
);
713 *saved_stdout
= safe_close(*saved_stdout
);
719 CONFIRM_PRETEND_FAILURE
= -1,
720 CONFIRM_PRETEND_SUCCESS
= 0,
724 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
725 int saved_stdout
= -1, saved_stdin
= -1, r
;
726 _cleanup_free_
char *e
= NULL
;
729 /* For any internal errors, assume a positive response. */
730 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
732 write_confirm_error(r
, vc
, u
);
733 return CONFIRM_EXECUTE
;
736 /* confirm_spawn might have been disabled while we were sleeping. */
737 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
742 e
= ellipsize(cmdline
, 60, 100);
750 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
752 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
759 printf("Resuming normal execution.\n");
760 manager_disable_confirm_spawn();
764 unit_dump(u
, stdout
, " ");
765 continue; /* ask again */
767 printf("Failing execution.\n");
768 r
= CONFIRM_PRETEND_FAILURE
;
771 printf(" c - continue, proceed without asking anymore\n"
772 " D - dump, show the state of the unit\n"
773 " f - fail, don't execute the command and pretend it failed\n"
775 " i - info, show a short summary of the unit\n"
776 " j - jobs, show jobs that are in progress\n"
777 " s - skip, don't execute the command and pretend it succeeded\n"
778 " y - yes, execute the command\n");
779 continue; /* ask again */
781 printf(" Description: %s\n"
784 u
->id
, u
->description
, cmdline
);
785 continue; /* ask again */
787 manager_dump_jobs(u
->manager
, stdout
, " ");
788 continue; /* ask again */
790 /* 'n' was removed in favor of 'f'. */
791 printf("Didn't understand 'n', did you mean 'f'?\n");
792 continue; /* ask again */
794 printf("Skipping execution.\n");
795 r
= CONFIRM_PRETEND_SUCCESS
;
801 assert_not_reached("Unhandled choice");
807 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
811 static int get_fixed_user(const ExecContext
*c
, const char **user
,
812 uid_t
*uid
, gid_t
*gid
,
813 const char **home
, const char **shell
) {
822 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
823 * (i.e. are "/" or "/bin/nologin"). */
826 r
= get_user_creds_clean(&name
, uid
, gid
, home
, shell
);
834 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
844 r
= get_group_creds(&name
, gid
);
852 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
853 const char *group
, gid_t gid
,
854 gid_t
**supplementary_gids
, int *ngids
) {
858 bool keep_groups
= false;
859 gid_t
*groups
= NULL
;
860 _cleanup_free_ gid_t
*l_gids
= NULL
;
865 * If user is given, then lookup GID and supplementary groups list.
866 * We avoid NSS lookups for gid=0. Also we have to initialize groups
867 * here and as early as possible so we keep the list of supplementary
868 * groups of the caller.
870 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
871 /* First step, initialize groups from /etc/groups */
872 if (initgroups(user
, gid
) < 0)
878 if (!c
->supplementary_groups
)
882 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
883 * be positive, otherwise fail.
886 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
887 if (ngroups_max
<= 0) {
891 return -EOPNOTSUPP
; /* For all other values */
894 l_gids
= new(gid_t
, ngroups_max
);
900 * Lookup the list of groups that the user belongs to, we
901 * avoid NSS lookups here too for gid=0.
904 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
909 STRV_FOREACH(i
, c
->supplementary_groups
) {
912 if (k
>= ngroups_max
)
916 r
= get_group_creds(&g
, l_gids
+k
);
924 * Sets ngids to zero to drop all supplementary groups, happens
925 * when we are under root and SupplementaryGroups= is empty.
932 /* Otherwise get the final list of supplementary groups */
933 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
937 *supplementary_gids
= groups
;
945 static int enforce_groups(const ExecContext
*context
, gid_t gid
,
946 gid_t
*supplementary_gids
, int ngids
) {
951 /* Handle SupplementaryGroups= even if it is empty */
952 if (context
->supplementary_groups
) {
953 r
= maybe_setgroups(ngids
, supplementary_gids
);
958 if (gid_is_valid(gid
)) {
959 /* Then set our gids */
960 if (setresgid(gid
, gid
, gid
) < 0)
967 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
970 if (!uid_is_valid(uid
))
973 /* Sets (but doesn't look up) the uid and make sure we keep the
974 * capabilities while doing so. */
976 if (context
->capability_ambient_set
!= 0) {
978 /* First step: If we need to keep capabilities but
979 * drop privileges we need to make sure we keep our
980 * caps, while we drop privileges. */
982 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
984 if (prctl(PR_GET_SECUREBITS
) != sb
)
985 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
990 /* Second step: actually set the uids */
991 if (setresuid(uid
, uid
, uid
) < 0)
994 /* At this point we should have all necessary capabilities but
995 are otherwise a normal user. However, the caps might got
996 corrupted due to the setresuid() so we need clean them up
997 later. This is done outside of this call. */
1004 static int null_conv(
1006 const struct pam_message
**msg
,
1007 struct pam_response
**resp
,
1008 void *appdata_ptr
) {
1010 /* We don't support conversations */
1012 return PAM_CONV_ERR
;
1017 static int setup_pam(
1024 int fds
[], unsigned n_fds
) {
1028 static const struct pam_conv conv
= {
1033 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1034 pam_handle_t
*handle
= NULL
;
1036 int pam_code
= PAM_SUCCESS
, r
;
1037 char **nv
, **e
= NULL
;
1038 bool close_session
= false;
1039 pid_t pam_pid
= 0, parent_pid
;
1046 /* We set up PAM in the parent process, then fork. The child
1047 * will then stay around until killed via PR_GET_PDEATHSIG or
1048 * systemd via the cgroup logic. It will then remove the PAM
1049 * session again. The parent process will exec() the actual
1050 * daemon. We do things this way to ensure that the main PID
1051 * of the daemon is the one we initially fork()ed. */
1053 r
= barrier_create(&barrier
);
1057 if (log_get_max_level() < LOG_DEBUG
)
1058 flags
|= PAM_SILENT
;
1060 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1061 if (pam_code
!= PAM_SUCCESS
) {
1067 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1068 if (pam_code
!= PAM_SUCCESS
)
1072 STRV_FOREACH(nv
, *env
) {
1073 pam_code
= pam_putenv(handle
, *nv
);
1074 if (pam_code
!= PAM_SUCCESS
)
1078 pam_code
= pam_acct_mgmt(handle
, flags
);
1079 if (pam_code
!= PAM_SUCCESS
)
1082 pam_code
= pam_open_session(handle
, flags
);
1083 if (pam_code
!= PAM_SUCCESS
)
1086 close_session
= true;
1088 e
= pam_getenvlist(handle
);
1090 pam_code
= PAM_BUF_ERR
;
1094 /* Block SIGTERM, so that we know that it won't get lost in
1097 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1099 parent_pid
= getpid_cached();
1108 int sig
, ret
= EXIT_PAM
;
1110 /* The child's job is to reset the PAM session on
1112 barrier_set_role(&barrier
, BARRIER_CHILD
);
1114 /* This string must fit in 10 chars (i.e. the length
1115 * of "/sbin/init"), to look pretty in /bin/ps */
1116 rename_process("(sd-pam)");
1118 /* Make sure we don't keep open the passed fds in this
1119 child. We assume that otherwise only those fds are
1120 open here that have been opened by PAM. */
1121 close_many(fds
, n_fds
);
1123 /* Drop privileges - we don't need any to pam_close_session
1124 * and this will make PR_SET_PDEATHSIG work in most cases.
1125 * If this fails, ignore the error - but expect sd-pam threads
1126 * to fail to exit normally */
1128 r
= maybe_setgroups(0, NULL
);
1130 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1131 if (setresgid(gid
, gid
, gid
) < 0)
1132 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1133 if (setresuid(uid
, uid
, uid
) < 0)
1134 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1136 (void) ignore_signals(SIGPIPE
, -1);
1138 /* Wait until our parent died. This will only work if
1139 * the above setresuid() succeeds, otherwise the kernel
1140 * will not allow unprivileged parents kill their privileged
1141 * children this way. We rely on the control groups kill logic
1142 * to do the rest for us. */
1143 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1146 /* Tell the parent that our setup is done. This is especially
1147 * important regarding dropping privileges. Otherwise, unit
1148 * setup might race against our setresuid(2) call.
1150 * If the parent aborted, we'll detect this below, hence ignore
1151 * return failure here. */
1152 (void) barrier_place(&barrier
);
1154 /* Check if our parent process might already have died? */
1155 if (getppid() == parent_pid
) {
1158 assert_se(sigemptyset(&ss
) >= 0);
1159 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1162 if (sigwait(&ss
, &sig
) < 0) {
1169 assert(sig
== SIGTERM
);
1174 /* If our parent died we'll end the session */
1175 if (getppid() != parent_pid
) {
1176 pam_code
= pam_close_session(handle
, flags
);
1177 if (pam_code
!= PAM_SUCCESS
)
1184 pam_end(handle
, pam_code
| flags
);
1188 barrier_set_role(&barrier
, BARRIER_PARENT
);
1190 /* If the child was forked off successfully it will do all the
1191 * cleanups, so forget about the handle here. */
1194 /* Unblock SIGTERM again in the parent */
1195 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1197 /* We close the log explicitly here, since the PAM modules
1198 * might have opened it, but we don't want this fd around. */
1201 /* Synchronously wait for the child to initialize. We don't care for
1202 * errors as we cannot recover. However, warn loudly if it happens. */
1203 if (!barrier_place_and_sync(&barrier
))
1204 log_error("PAM initialization failed");
1212 if (pam_code
!= PAM_SUCCESS
) {
1213 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1214 r
= -EPERM
; /* PAM errors do not map to errno */
1216 log_error_errno(r
, "PAM failed: %m");
1220 pam_code
= pam_close_session(handle
, flags
);
1222 pam_end(handle
, pam_code
| flags
);
1234 static void rename_process_from_path(const char *path
) {
1235 char process_name
[11];
1239 /* This resulting string must fit in 10 chars (i.e. the length
1240 * of "/sbin/init") to look pretty in /bin/ps */
1244 rename_process("(...)");
1250 /* The end of the process name is usually more
1251 * interesting, since the first bit might just be
1257 process_name
[0] = '(';
1258 memcpy(process_name
+1, p
, l
);
1259 process_name
[1+l
] = ')';
1260 process_name
[1+l
+1] = 0;
1262 rename_process(process_name
);
1265 static bool context_has_address_families(const ExecContext
*c
) {
1268 return c
->address_families_whitelist
||
1269 !set_isempty(c
->address_families
);
1272 static bool context_has_syscall_filters(const ExecContext
*c
) {
1275 return c
->syscall_whitelist
||
1276 !set_isempty(c
->syscall_filter
);
1279 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1282 if (c
->no_new_privileges
)
1285 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1288 /* We need NNP if we have any form of seccomp and are unprivileged */
1289 return context_has_address_families(c
) ||
1290 c
->memory_deny_write_execute
||
1291 c
->restrict_realtime
||
1292 exec_context_restrict_namespaces_set(c
) ||
1293 c
->protect_kernel_tunables
||
1294 c
->protect_kernel_modules
||
1295 c
->private_devices
||
1296 context_has_syscall_filters(c
) ||
1297 !set_isempty(c
->syscall_archs
);
1302 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1304 if (is_seccomp_available())
1308 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1313 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
) {
1314 uint32_t negative_action
, default_action
, action
;
1319 if (!context_has_syscall_filters(c
))
1322 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1325 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1327 if (c
->syscall_whitelist
) {
1328 default_action
= negative_action
;
1329 action
= SCMP_ACT_ALLOW
;
1331 default_action
= SCMP_ACT_ALLOW
;
1332 action
= negative_action
;
1335 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
);
1338 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1342 if (set_isempty(c
->syscall_archs
))
1345 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1348 return seccomp_restrict_archs(c
->syscall_archs
);
1351 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1355 if (!context_has_address_families(c
))
1358 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1361 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1364 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1368 if (!c
->memory_deny_write_execute
)
1371 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1374 return seccomp_memory_deny_write_execute();
1377 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1381 if (!c
->restrict_realtime
)
1384 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1387 return seccomp_restrict_realtime();
1390 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1394 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1395 * let's protect even those systems where this is left on in the kernel. */
1397 if (!c
->protect_kernel_tunables
)
1400 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1403 return seccomp_protect_sysctl();
1406 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1410 /* Turn off module syscalls on ProtectKernelModules=yes */
1412 if (!c
->protect_kernel_modules
)
1415 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1418 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
));
1421 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1425 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1427 if (!c
->private_devices
)
1430 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1433 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
));
1436 static int apply_restrict_namespaces(Unit
*u
, const ExecContext
*c
) {
1440 if (!exec_context_restrict_namespaces_set(c
))
1443 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1446 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1451 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1454 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1455 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1457 if (idle_pipe
[0] >= 0) {
1460 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1462 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1465 /* Signal systemd that we are bored and want to continue. */
1466 n
= write(idle_pipe
[3], "x", 1);
1468 /* Wait for systemd to react to the signal above. */
1469 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1472 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1476 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1479 static int build_environment(
1481 const ExecContext
*c
,
1482 const ExecParameters
*p
,
1485 const char *username
,
1487 dev_t journal_stream_dev
,
1488 ino_t journal_stream_ino
,
1491 _cleanup_strv_free_
char **our_env
= NULL
;
1499 our_env
= new0(char*, 14);
1504 _cleanup_free_
char *joined
= NULL
;
1506 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1508 our_env
[n_env
++] = x
;
1510 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1512 our_env
[n_env
++] = x
;
1514 joined
= strv_join(p
->fd_names
, ":");
1518 x
= strjoin("LISTEN_FDNAMES=", joined
);
1521 our_env
[n_env
++] = x
;
1524 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1525 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1527 our_env
[n_env
++] = x
;
1529 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1531 our_env
[n_env
++] = x
;
1534 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1535 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1536 * check the database directly. */
1537 if (unit_has_name(u
, SPECIAL_DBUS_SERVICE
)) {
1538 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1541 our_env
[n_env
++] = x
;
1545 x
= strappend("HOME=", home
);
1548 our_env
[n_env
++] = x
;
1552 x
= strappend("LOGNAME=", username
);
1555 our_env
[n_env
++] = x
;
1557 x
= strappend("USER=", username
);
1560 our_env
[n_env
++] = x
;
1564 x
= strappend("SHELL=", shell
);
1567 our_env
[n_env
++] = x
;
1570 if (!sd_id128_is_null(u
->invocation_id
)) {
1571 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1574 our_env
[n_env
++] = x
;
1577 if (exec_context_needs_term(c
)) {
1578 const char *tty_path
, *term
= NULL
;
1580 tty_path
= exec_context_tty_path(c
);
1582 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1583 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1584 * passes to PID 1 ends up all the way in the console login shown. */
1586 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1587 term
= getenv("TERM");
1589 term
= default_term_for_tty(tty_path
);
1591 x
= strappend("TERM=", term
);
1594 our_env
[n_env
++] = x
;
1597 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1598 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1601 our_env
[n_env
++] = x
;
1604 our_env
[n_env
++] = NULL
;
1605 assert(n_env
<= 12);
1613 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1614 _cleanup_strv_free_
char **pass_env
= NULL
;
1615 size_t n_env
= 0, n_bufsize
= 0;
1618 STRV_FOREACH(i
, c
->pass_environment
) {
1619 _cleanup_free_
char *x
= NULL
;
1625 x
= strjoin(*i
, "=", v
);
1628 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1630 pass_env
[n_env
++] = x
;
1631 pass_env
[n_env
] = NULL
;
1641 static bool exec_needs_mount_namespace(
1642 const ExecContext
*context
,
1643 const ExecParameters
*params
,
1644 ExecRuntime
*runtime
) {
1649 if (context
->root_image
)
1652 if (!strv_isempty(context
->read_write_paths
) ||
1653 !strv_isempty(context
->read_only_paths
) ||
1654 !strv_isempty(context
->inaccessible_paths
))
1657 if (context
->n_bind_mounts
> 0)
1660 if (context
->mount_flags
!= 0)
1663 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1666 if (context
->private_devices
||
1667 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1668 context
->protect_home
!= PROTECT_HOME_NO
||
1669 context
->protect_kernel_tunables
||
1670 context
->protect_kernel_modules
||
1671 context
->protect_control_groups
)
1674 if (context
->mount_apivfs
&& (context
->root_image
|| context
->root_directory
))
1680 static int setup_private_users(uid_t uid
, gid_t gid
) {
1681 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1682 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1683 _cleanup_close_
int unshare_ready_fd
= -1;
1684 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1690 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1691 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1692 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1693 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1694 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1695 * continues execution normally. */
1697 if (uid
!= 0 && uid_is_valid(uid
)) {
1698 r
= asprintf(&uid_map
,
1699 "0 0 1\n" /* Map root → root */
1700 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1705 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1710 if (gid
!= 0 && gid_is_valid(gid
)) {
1711 r
= asprintf(&gid_map
,
1712 "0 0 1\n" /* Map root → root */
1713 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1718 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1723 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1725 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1726 if (unshare_ready_fd
< 0)
1729 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1731 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1739 _cleanup_close_
int fd
= -1;
1743 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1744 * here, after the parent opened its own user namespace. */
1747 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1749 /* Wait until the parent unshared the user namespace */
1750 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1755 /* Disable the setgroups() system call in the child user namespace, for good. */
1756 a
= procfs_file_alloca(ppid
, "setgroups");
1757 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1759 if (errno
!= ENOENT
) {
1764 /* If the file is missing the kernel is too old, let's continue anyway. */
1766 if (write(fd
, "deny\n", 5) < 0) {
1771 fd
= safe_close(fd
);
1774 /* First write the GID map */
1775 a
= procfs_file_alloca(ppid
, "gid_map");
1776 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1781 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1785 fd
= safe_close(fd
);
1787 /* The write the UID map */
1788 a
= procfs_file_alloca(ppid
, "uid_map");
1789 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1794 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1799 _exit(EXIT_SUCCESS
);
1802 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1803 _exit(EXIT_FAILURE
);
1806 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1808 if (unshare(CLONE_NEWUSER
) < 0)
1811 /* Let the child know that the namespace is ready now */
1812 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1815 /* Try to read an error code from the child */
1816 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1819 if (n
== sizeof(r
)) { /* an error code was sent to us */
1824 if (n
!= 0) /* on success we should have read 0 bytes */
1827 r
= wait_for_terminate(pid
, &si
);
1832 /* If something strange happened with the child, let's consider this fatal, too */
1833 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1839 static int setup_exec_directory(
1840 const ExecContext
*context
,
1841 const ExecParameters
*params
,
1844 bool manager_is_system
,
1845 ExecDirectoryType type
,
1848 static const int exit_status_table
[_EXEC_DIRECTORY_MAX
] = {
1849 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
1850 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
1851 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
1852 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
1853 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
1860 assert(type
>= 0 && type
< _EXEC_DIRECTORY_MAX
);
1861 assert(exit_status
);
1863 if (!params
->prefix
[type
])
1866 if (manager_is_system
) {
1867 if (!uid_is_valid(uid
))
1869 if (!gid_is_valid(gid
))
1873 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
1874 _cleanup_free_
char *p
;
1876 p
= strjoin(params
->prefix
[type
], "/", *rt
);
1882 r
= mkdir_parents_label(p
, 0755);
1886 r
= mkdir_p_label(p
, context
->directories
[type
].mode
);
1890 r
= chmod_and_chown(p
, context
->directories
[type
].mode
, uid
, gid
);
1898 *exit_status
= exit_status_table
[type
];
1903 static int setup_smack(
1904 const ExecContext
*context
,
1905 const ExecCommand
*command
) {
1912 if (context
->smack_process_label
) {
1913 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1917 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1919 _cleanup_free_
char *exec_label
= NULL
;
1921 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1922 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1925 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
1934 static int compile_read_write_paths(
1935 const ExecContext
*context
,
1936 const ExecParameters
*params
,
1939 _cleanup_strv_free_
char **l
= NULL
;
1941 ExecDirectoryType i
;
1943 /* Compile the list of writable paths. This is the combination of
1944 * the explicitly configured paths, plus all runtime directories. */
1946 if (strv_isempty(context
->read_write_paths
)) {
1947 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++)
1948 if (!strv_isempty(context
->directories
[i
].paths
))
1951 if (i
== _EXEC_DIRECTORY_MAX
) {
1952 *ret
= NULL
; /* NOP if neither is set */
1957 l
= strv_copy(context
->read_write_paths
);
1961 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++) {
1962 if (!params
->prefix
[i
])
1965 STRV_FOREACH(rt
, context
->directories
[i
].paths
) {
1968 s
= strjoin(params
->prefix
[i
], "/", *rt
);
1972 if (strv_consume(&l
, s
) < 0)
1983 static int apply_mount_namespace(
1985 ExecCommand
*command
,
1986 const ExecContext
*context
,
1987 const ExecParameters
*params
,
1988 ExecRuntime
*runtime
) {
1990 _cleanup_strv_free_
char **rw
= NULL
;
1991 char *tmp
= NULL
, *var
= NULL
;
1992 const char *root_dir
= NULL
, *root_image
= NULL
;
1993 NameSpaceInfo ns_info
= {
1994 .ignore_protect_paths
= false,
1995 .private_dev
= context
->private_devices
,
1996 .protect_control_groups
= context
->protect_control_groups
,
1997 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
1998 .protect_kernel_modules
= context
->protect_kernel_modules
,
1999 .mount_apivfs
= context
->mount_apivfs
,
2001 bool apply_restrictions
;
2006 /* The runtime struct only contains the parent of the private /tmp,
2007 * which is non-accessible to world users. Inside of it there's a /tmp
2008 * that is sticky, and that's the one we want to use here. */
2010 if (context
->private_tmp
&& runtime
) {
2011 if (runtime
->tmp_dir
)
2012 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2013 if (runtime
->var_tmp_dir
)
2014 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2017 r
= compile_read_write_paths(context
, params
, &rw
);
2021 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2022 root_image
= context
->root_image
;
2025 root_dir
= context
->root_directory
;
2029 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2030 * sandbox info, otherwise enforce it, don't ignore protected paths and
2031 * fail if we are enable to apply the sandbox inside the mount namespace.
2033 if (!context
->dynamic_user
&& root_dir
)
2034 ns_info
.ignore_protect_paths
= true;
2036 apply_restrictions
= (params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
;
2038 r
= setup_namespace(root_dir
, root_image
,
2040 apply_restrictions
? context
->read_only_paths
: NULL
,
2041 apply_restrictions
? context
->inaccessible_paths
: NULL
,
2042 context
->bind_mounts
,
2043 context
->n_bind_mounts
,
2046 apply_restrictions
? context
->protect_home
: PROTECT_HOME_NO
,
2047 apply_restrictions
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2048 context
->mount_flags
,
2049 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2051 /* If we couldn't set up the namespace this is probably due to a
2052 * missing capability. In this case, silently proceeed. */
2053 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2055 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2063 static int apply_working_directory(
2064 const ExecContext
*context
,
2065 const ExecParameters
*params
,
2067 const bool needs_mount_ns
,
2073 assert(exit_status
);
2075 if (context
->working_directory_home
) {
2078 *exit_status
= EXIT_CHDIR
;
2084 } else if (context
->working_directory
)
2085 wd
= context
->working_directory
;
2089 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2090 if (!needs_mount_ns
&& context
->root_directory
)
2091 if (chroot(context
->root_directory
) < 0) {
2092 *exit_status
= EXIT_CHROOT
;
2098 d
= prefix_roota(context
->root_directory
, wd
);
2100 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2101 *exit_status
= EXIT_CHDIR
;
2108 static int setup_keyring(Unit
*u
, const ExecParameters
*p
, uid_t uid
, gid_t gid
) {
2109 key_serial_t keyring
;
2114 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2115 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2116 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2117 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2118 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2119 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2121 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2124 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2125 if (keyring
== -1) {
2126 if (errno
== ENOSYS
)
2127 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2128 else if (IN_SET(errno
, EACCES
, EPERM
))
2129 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2130 else if (errno
== EDQUOT
)
2131 log_debug_errno(errno
, "Out of kernel keyrings to allocate, ignoring.");
2133 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2138 /* Populate they keyring with the invocation ID by default. */
2139 if (!sd_id128_is_null(u
->invocation_id
)) {
2142 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2144 log_debug_errno(errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2146 if (keyctl(KEYCTL_SETPERM
, key
,
2147 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2148 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2149 return log_error_errno(errno
, "Failed to restrict invocation ID permission: %m");
2153 /* And now, make the keyring owned by the service's user */
2154 if (uid_is_valid(uid
) || gid_is_valid(gid
))
2155 if (keyctl(KEYCTL_CHOWN
, keyring
, uid
, gid
, 0) < 0)
2156 return log_error_errno(errno
, "Failed to change ownership of session keyring: %m");
2161 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
2169 array
[(*n
)++] = pair
[0];
2171 array
[(*n
)++] = pair
[1];
2174 static int close_remaining_fds(
2175 const ExecParameters
*params
,
2176 ExecRuntime
*runtime
,
2177 DynamicCreds
*dcreds
,
2180 int *fds
, unsigned n_fds
) {
2182 unsigned n_dont_close
= 0;
2183 int dont_close
[n_fds
+ 12];
2187 if (params
->stdin_fd
>= 0)
2188 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2189 if (params
->stdout_fd
>= 0)
2190 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2191 if (params
->stderr_fd
>= 0)
2192 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2195 dont_close
[n_dont_close
++] = socket_fd
;
2197 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2198 n_dont_close
+= n_fds
;
2202 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2206 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2208 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2211 if (user_lookup_fd
>= 0)
2212 dont_close
[n_dont_close
++] = user_lookup_fd
;
2214 return close_all_fds(dont_close
, n_dont_close
);
2217 static int send_user_lookup(
2225 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2226 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2229 if (user_lookup_fd
< 0)
2232 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2235 if (writev(user_lookup_fd
,
2237 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2238 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2239 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2245 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2252 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2257 if (!c
->working_directory_home
)
2261 /* Hardcode /root as home directory for UID 0 */
2266 r
= get_home_dir(buf
);
2274 static int exec_child(
2276 ExecCommand
*command
,
2277 const ExecContext
*context
,
2278 const ExecParameters
*params
,
2279 ExecRuntime
*runtime
,
2280 DynamicCreds
*dcreds
,
2285 unsigned n_storage_fds
,
2286 unsigned n_socket_fds
,
2290 char **error_message
) {
2292 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2293 _cleanup_free_
char *mac_selinux_context_net
= NULL
, *home_buffer
= NULL
;
2294 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2295 const char *username
= NULL
, *groupname
= NULL
;
2296 const char *home
= NULL
, *shell
= NULL
;
2297 dev_t journal_stream_dev
= 0;
2298 ino_t journal_stream_ino
= 0;
2299 bool needs_exec_restrictions
, needs_mount_namespace
;
2301 bool needs_selinux
= false;
2304 bool needs_smack
= false;
2306 #ifdef HAVE_APPARMOR
2307 bool needs_apparmor
= false;
2309 uid_t uid
= UID_INVALID
;
2310 gid_t gid
= GID_INVALID
;
2311 int i
, r
, ngids
= 0;
2313 ExecDirectoryType dt
;
2319 assert(exit_status
);
2320 assert(error_message
);
2321 /* We don't always set error_message, hence it must be initialized */
2322 assert(*error_message
== NULL
);
2324 rename_process_from_path(command
->path
);
2326 /* We reset exactly these signals, since they are the
2327 * only ones we set to SIG_IGN in the main daemon. All
2328 * others we leave untouched because we set them to
2329 * SIG_DFL or a valid handler initially, both of which
2330 * will be demoted to SIG_DFL. */
2331 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2332 SIGNALS_IGNORE
, -1);
2334 if (context
->ignore_sigpipe
)
2335 (void) ignore_signals(SIGPIPE
, -1);
2337 r
= reset_signal_mask();
2339 *exit_status
= EXIT_SIGNAL_MASK
;
2340 *error_message
= strdup("Failed to reset signal mask");
2341 /* If strdup fails, here and below, we will just print the generic error message. */
2345 if (params
->idle_pipe
)
2346 do_idle_pipe_dance(params
->idle_pipe
);
2348 /* Close sockets very early to make sure we don't
2349 * block init reexecution because it cannot bind its
2354 n_fds
= n_storage_fds
+ n_socket_fds
;
2355 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2357 *exit_status
= EXIT_FDS
;
2358 *error_message
= strdup("Failed to close remaining fds");
2362 if (!context
->same_pgrp
)
2364 *exit_status
= EXIT_SETSID
;
2368 exec_context_tty_reset(context
, params
);
2370 if (unit_shall_confirm_spawn(unit
)) {
2371 const char *vc
= params
->confirm_spawn
;
2372 _cleanup_free_
char *cmdline
= NULL
;
2374 cmdline
= exec_command_line(argv
);
2376 *exit_status
= EXIT_CONFIRM
;
2380 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2381 if (r
!= CONFIRM_EXECUTE
) {
2382 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2383 *exit_status
= EXIT_SUCCESS
;
2386 *exit_status
= EXIT_CONFIRM
;
2387 *error_message
= strdup("Execution cancelled");
2392 if (context
->dynamic_user
&& dcreds
) {
2394 /* Make sure we bypass our own NSS module for any NSS checks */
2395 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2396 *exit_status
= EXIT_USER
;
2397 *error_message
= strdup("Failed to update environment");
2401 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2403 *exit_status
= EXIT_USER
;
2404 *error_message
= strdup("Failed to update dynamic user credentials");
2408 if (!uid_is_valid(uid
)) {
2409 *exit_status
= EXIT_USER
;
2410 (void) asprintf(error_message
, "UID validation failed for \""UID_FMT
"\"", uid
);
2411 /* If asprintf fails, here and below, we will just print the generic error message. */
2415 if (!gid_is_valid(gid
)) {
2416 *exit_status
= EXIT_USER
;
2417 (void) asprintf(error_message
, "GID validation failed for \""GID_FMT
"\"", gid
);
2422 username
= dcreds
->user
->name
;
2425 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2427 *exit_status
= EXIT_USER
;
2428 *error_message
= strdup("Failed to determine user credentials");
2432 r
= get_fixed_group(context
, &groupname
, &gid
);
2434 *exit_status
= EXIT_GROUP
;
2435 *error_message
= strdup("Failed to determine group credentials");
2440 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2441 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2442 &supplementary_gids
, &ngids
);
2444 *exit_status
= EXIT_GROUP
;
2445 *error_message
= strdup("Failed to determine supplementary groups");
2449 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2451 *exit_status
= EXIT_USER
;
2452 *error_message
= strdup("Failed to send user credentials to PID1");
2456 user_lookup_fd
= safe_close(user_lookup_fd
);
2458 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2460 *exit_status
= EXIT_CHDIR
;
2461 *error_message
= strdup("Failed to determine $HOME for user");
2465 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2466 * must sure to drop O_NONBLOCK */
2468 (void) fd_nonblock(socket_fd
, false);
2470 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2472 *exit_status
= EXIT_STDIN
;
2473 *error_message
= strdup("Failed to set up stdin");
2477 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2479 *exit_status
= EXIT_STDOUT
;
2480 *error_message
= strdup("Failed to set up stdout");
2484 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2486 *exit_status
= EXIT_STDERR
;
2487 *error_message
= strdup("Failed to set up stderr");
2491 if (params
->cgroup_path
) {
2492 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2494 *exit_status
= EXIT_CGROUP
;
2495 (void) asprintf(error_message
, "Failed to attach to cgroup %s", params
->cgroup_path
);
2500 if (context
->oom_score_adjust_set
) {
2501 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2503 /* When we can't make this change due to EPERM, then
2504 * let's silently skip over it. User namespaces
2505 * prohibit write access to this file, and we
2506 * shouldn't trip up over that. */
2508 sprintf(t
, "%i", context
->oom_score_adjust
);
2509 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2510 if (r
== -EPERM
|| r
== -EACCES
) {
2512 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2515 *exit_status
= EXIT_OOM_ADJUST
;
2516 *error_message
= strdup("Failed to write /proc/self/oom_score_adj");
2521 if (context
->nice_set
)
2522 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2523 *exit_status
= EXIT_NICE
;
2527 if (context
->cpu_sched_set
) {
2528 struct sched_param param
= {
2529 .sched_priority
= context
->cpu_sched_priority
,
2532 r
= sched_setscheduler(0,
2533 context
->cpu_sched_policy
|
2534 (context
->cpu_sched_reset_on_fork
?
2535 SCHED_RESET_ON_FORK
: 0),
2538 *exit_status
= EXIT_SETSCHEDULER
;
2543 if (context
->cpuset
)
2544 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2545 *exit_status
= EXIT_CPUAFFINITY
;
2549 if (context
->ioprio_set
)
2550 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2551 *exit_status
= EXIT_IOPRIO
;
2555 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2556 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2557 *exit_status
= EXIT_TIMERSLACK
;
2561 if (context
->personality
!= PERSONALITY_INVALID
)
2562 if (personality(context
->personality
) < 0) {
2563 *exit_status
= EXIT_PERSONALITY
;
2567 if (context
->utmp_id
)
2568 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
2570 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2571 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2575 if (context
->user
) {
2576 r
= chown_terminal(STDIN_FILENO
, uid
);
2578 *exit_status
= EXIT_STDIN
;
2583 /* If delegation is enabled we'll pass ownership of the cgroup
2584 * (but only in systemd's own controller hierarchy!) to the
2585 * user of the new process. */
2586 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2587 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2589 *exit_status
= EXIT_CGROUP
;
2594 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2596 *exit_status
= EXIT_CGROUP
;
2601 for (dt
= 0; dt
< _EXEC_DIRECTORY_MAX
; dt
++) {
2602 r
= setup_exec_directory(context
, params
, uid
, gid
, MANAGER_IS_SYSTEM(unit
->manager
), dt
, exit_status
);
2607 r
= build_environment(
2619 *exit_status
= EXIT_MEMORY
;
2623 r
= build_pass_environment(context
, &pass_env
);
2625 *exit_status
= EXIT_MEMORY
;
2629 accum_env
= strv_env_merge(5,
2630 params
->environment
,
2633 context
->environment
,
2637 *exit_status
= EXIT_MEMORY
;
2640 accum_env
= strv_env_clean(accum_env
);
2642 (void) umask(context
->umask
);
2644 r
= setup_keyring(unit
, params
, uid
, gid
);
2646 *exit_status
= EXIT_KEYRING
;
2650 needs_exec_restrictions
= (params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
;
2652 if (needs_exec_restrictions
) {
2653 if (context
->pam_name
&& username
) {
2654 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2656 *exit_status
= EXIT_PAM
;
2661 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
2662 * present. The actual MAC context application will happen later, as late as possible, to avoid
2663 * impacting our own code paths. */
2666 needs_selinux
= mac_selinux_use();
2670 needs_smack
= mac_smack_use();
2673 #ifdef HAVE_APPARMOR
2674 needs_apparmor
= context
->apparmor_profile
&& mac_apparmor_use();
2679 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2680 r
= setup_netns(runtime
->netns_storage_socket
);
2682 *exit_status
= EXIT_NETWORK
;
2687 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2688 if (needs_mount_namespace
) {
2689 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
2691 *exit_status
= EXIT_NAMESPACE
;
2696 /* Apply just after mount namespace setup */
2697 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
2701 /* Drop groups as early as possbile */
2702 if (needs_exec_restrictions
) {
2703 r
= enforce_groups(context
, gid
, supplementary_gids
, ngids
);
2705 *exit_status
= EXIT_GROUP
;
2711 if (needs_exec_restrictions
&& needs_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
2712 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2714 *exit_status
= EXIT_SELINUX_CONTEXT
;
2720 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2721 r
= setup_private_users(uid
, gid
);
2723 *exit_status
= EXIT_USER
;
2728 /* We repeat the fd closing here, to make sure that
2729 * nothing is leaked from the PAM modules. Note that
2730 * we are more aggressive this time since socket_fd
2731 * and the netns fds we don't need anymore. The custom
2732 * endpoint fd was needed to upload the policy and can
2733 * now be closed as well. */
2734 r
= close_all_fds(fds
, n_fds
);
2736 r
= shift_fds(fds
, n_fds
);
2738 r
= flags_fds(fds
, n_storage_fds
, n_socket_fds
, context
->non_blocking
);
2740 *exit_status
= EXIT_FDS
;
2744 if (needs_exec_restrictions
) {
2746 int secure_bits
= context
->secure_bits
;
2748 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2750 if (!context
->rlimit
[i
])
2753 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2755 *exit_status
= EXIT_LIMITS
;
2760 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2761 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2762 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2763 *exit_status
= EXIT_LIMITS
;
2768 if (!cap_test_all(context
->capability_bounding_set
)) {
2769 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2771 *exit_status
= EXIT_CAPABILITIES
;
2772 *error_message
= strdup("Failed to drop capabilities");
2777 /* This is done before enforce_user, but ambient set
2778 * does not survive over setresuid() if keep_caps is not set. */
2779 if (context
->capability_ambient_set
!= 0) {
2780 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2782 *exit_status
= EXIT_CAPABILITIES
;
2783 *error_message
= strdup("Failed to apply ambient capabilities (before UID change)");
2788 if (context
->user
) {
2789 r
= enforce_user(context
, uid
);
2791 *exit_status
= EXIT_USER
;
2792 (void) asprintf(error_message
, "Failed to change UID to "UID_FMT
, uid
);
2795 if (context
->capability_ambient_set
!= 0) {
2797 /* Fix the ambient capabilities after user change. */
2798 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2800 *exit_status
= EXIT_CAPABILITIES
;
2801 *error_message
= strdup("Failed to apply ambient capabilities (after UID change)");
2805 /* If we were asked to change user and ambient capabilities
2806 * were requested, we had to add keep-caps to the securebits
2807 * so that we would maintain the inherited capability set
2808 * through the setresuid(). Make sure that the bit is added
2809 * also to the context secure_bits so that we don't try to
2810 * drop the bit away next. */
2812 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2816 /* Apply the MAC contexts late, but before seccomp syscall filtering, as those should really be last to
2817 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
2818 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
2819 * are restricted. */
2822 if (needs_selinux
) {
2823 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2826 r
= setexeccon(exec_context
);
2828 *exit_status
= EXIT_SELINUX_CONTEXT
;
2829 (void) asprintf(error_message
, "Failed to set SELinux context to %s", exec_context
);
2838 r
= setup_smack(context
, command
);
2840 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2841 *error_message
= strdup("Failed to set SMACK process label");
2847 #ifdef HAVE_APPARMOR
2848 if (needs_apparmor
) {
2849 r
= aa_change_onexec(context
->apparmor_profile
);
2850 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2851 *exit_status
= EXIT_APPARMOR_PROFILE
;
2852 (void) asprintf(error_message
,
2853 "Failed to prepare AppArmor profile change to %s",
2854 context
->apparmor_profile
);
2860 /* PR_GET_SECUREBITS is not privileged, while
2861 * PR_SET_SECUREBITS is. So to suppress
2862 * potential EPERMs we'll try not to call
2863 * PR_SET_SECUREBITS unless necessary. */
2864 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2865 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2866 *exit_status
= EXIT_SECUREBITS
;
2867 *error_message
= strdup("Failed to set secure bits");
2871 if (context_has_no_new_privileges(context
))
2872 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2873 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2874 *error_message
= strdup("Failed to disable new privileges");
2879 r
= apply_address_families(unit
, context
);
2881 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2882 *error_message
= strdup("Failed to restrict address families");
2886 r
= apply_memory_deny_write_execute(unit
, context
);
2888 *exit_status
= EXIT_SECCOMP
;
2889 *error_message
= strdup("Failed to disable writing to executable memory");
2893 r
= apply_restrict_realtime(unit
, context
);
2895 *exit_status
= EXIT_SECCOMP
;
2896 *error_message
= strdup("Failed to apply realtime restrictions");
2900 r
= apply_restrict_namespaces(unit
, context
);
2902 *exit_status
= EXIT_SECCOMP
;
2903 *error_message
= strdup("Failed to apply namespace restrictions");
2907 r
= apply_protect_sysctl(unit
, context
);
2909 *exit_status
= EXIT_SECCOMP
;
2910 *error_message
= strdup("Failed to apply sysctl restrictions");
2914 r
= apply_protect_kernel_modules(unit
, context
);
2916 *exit_status
= EXIT_SECCOMP
;
2917 *error_message
= strdup("Failed to apply module loading restrictions");
2921 r
= apply_private_devices(unit
, context
);
2923 *exit_status
= EXIT_SECCOMP
;
2924 *error_message
= strdup("Failed to set up private devices");
2928 r
= apply_syscall_archs(unit
, context
);
2930 *exit_status
= EXIT_SECCOMP
;
2931 *error_message
= strdup("Failed to apply syscall architecture restrictions");
2935 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
2936 * by the filter as little as possible. */
2937 r
= apply_syscall_filter(unit
, context
);
2939 *exit_status
= EXIT_SECCOMP
;
2940 *error_message
= strdup("Failed to apply syscall filters");
2946 final_argv
= replace_env_argv(argv
, accum_env
);
2948 *exit_status
= EXIT_MEMORY
;
2949 *error_message
= strdup("Failed to prepare process arguments");
2953 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2954 _cleanup_free_
char *line
;
2956 line
= exec_command_line(final_argv
);
2959 log_struct(LOG_DEBUG
,
2960 "EXECUTABLE=%s", command
->path
,
2961 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2968 execve(command
->path
, final_argv
, accum_env
);
2969 *exit_status
= EXIT_EXEC
;
2973 int exec_spawn(Unit
*unit
,
2974 ExecCommand
*command
,
2975 const ExecContext
*context
,
2976 const ExecParameters
*params
,
2977 ExecRuntime
*runtime
,
2978 DynamicCreds
*dcreds
,
2981 _cleanup_strv_free_
char **files_env
= NULL
;
2983 unsigned n_storage_fds
= 0, n_socket_fds
= 0;
2984 _cleanup_free_
char *line
= NULL
;
2986 int named_iofds
[3] = { -1, -1, -1 };
2995 assert(params
->fds
|| (params
->n_storage_fds
+ params
->n_socket_fds
<= 0));
2997 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2998 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2999 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3001 if (params
->n_socket_fds
> 1) {
3002 log_unit_error(unit
, "Got more than one socket.");
3006 if (params
->n_socket_fds
== 0) {
3007 log_unit_error(unit
, "Got no socket.");
3011 socket_fd
= params
->fds
[0];
3015 n_storage_fds
= params
->n_storage_fds
;
3016 n_socket_fds
= params
->n_socket_fds
;
3019 r
= exec_context_named_iofds(unit
, context
, params
, named_iofds
);
3021 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3023 r
= exec_context_load_environment(unit
, context
, &files_env
);
3025 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3027 argv
= params
->argv
?: command
->argv
;
3028 line
= exec_command_line(argv
);
3032 log_struct(LOG_DEBUG
,
3033 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3034 "EXECUTABLE=%s", command
->path
,
3039 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3043 _cleanup_free_
char *error_message
= NULL
;
3045 r
= exec_child(unit
,
3058 unit
->manager
->user_lookup_fds
[1],
3064 log_struct_errno(LOG_ERR
, r
,
3065 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3067 LOG_UNIT_MESSAGE(unit
, "%s: %m",
3069 "EXECUTABLE=%s", command
->path
,
3071 else if (r
== -ENOENT
&& command
->ignore
)
3072 log_struct_errno(LOG_INFO
, r
,
3073 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3075 LOG_UNIT_MESSAGE(unit
, "Skipped spawning %s: %m",
3077 "EXECUTABLE=%s", command
->path
,
3080 log_struct_errno(LOG_ERR
, r
,
3081 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3083 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3084 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3086 "EXECUTABLE=%s", command
->path
,
3093 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3095 /* We add the new process to the cgroup both in the child (so
3096 * that we can be sure that no user code is ever executed
3097 * outside of the cgroup) and in the parent (so that we can be
3098 * sure that when we kill the cgroup the process will be
3100 if (params
->cgroup_path
)
3101 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3103 exec_status_start(&command
->exec_status
, pid
);
3109 void exec_context_init(ExecContext
*c
) {
3110 ExecDirectoryType i
;
3115 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3116 c
->cpu_sched_policy
= SCHED_OTHER
;
3117 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3118 c
->syslog_level_prefix
= true;
3119 c
->ignore_sigpipe
= true;
3120 c
->timer_slack_nsec
= NSEC_INFINITY
;
3121 c
->personality
= PERSONALITY_INVALID
;
3122 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++)
3123 c
->directories
[i
].mode
= 0755;
3124 c
->capability_bounding_set
= CAP_ALL
;
3125 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
3128 void exec_context_done(ExecContext
*c
) {
3130 ExecDirectoryType i
;
3134 c
->environment
= strv_free(c
->environment
);
3135 c
->environment_files
= strv_free(c
->environment_files
);
3136 c
->pass_environment
= strv_free(c
->pass_environment
);
3138 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
3139 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
3141 for (l
= 0; l
< 3; l
++)
3142 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3144 c
->working_directory
= mfree(c
->working_directory
);
3145 c
->root_directory
= mfree(c
->root_directory
);
3146 c
->root_image
= mfree(c
->root_image
);
3147 c
->tty_path
= mfree(c
->tty_path
);
3148 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3149 c
->user
= mfree(c
->user
);
3150 c
->group
= mfree(c
->group
);
3152 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3154 c
->pam_name
= mfree(c
->pam_name
);
3156 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3157 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3158 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3160 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3163 CPU_FREE(c
->cpuset
);
3165 c
->utmp_id
= mfree(c
->utmp_id
);
3166 c
->selinux_context
= mfree(c
->selinux_context
);
3167 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3168 c
->smack_process_label
= mfree(c
->smack_process_label
);
3170 c
->syscall_filter
= set_free(c
->syscall_filter
);
3171 c
->syscall_archs
= set_free(c
->syscall_archs
);
3172 c
->address_families
= set_free(c
->address_families
);
3174 for (i
= 0; i
< _EXEC_DIRECTORY_MAX
; i
++)
3175 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3178 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
3183 if (!runtime_prefix
)
3186 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3187 _cleanup_free_
char *p
;
3189 p
= strjoin(runtime_prefix
, "/", *i
);
3193 /* We execute this synchronously, since we need to be
3194 * sure this is gone when we start the service
3196 (void) rm_rf(p
, REMOVE_ROOT
);
3202 void exec_command_done(ExecCommand
*c
) {
3205 c
->path
= mfree(c
->path
);
3207 c
->argv
= strv_free(c
->argv
);
3210 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3213 for (i
= 0; i
< n
; i
++)
3214 exec_command_done(c
+i
);
3217 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3221 LIST_REMOVE(command
, c
, i
);
3222 exec_command_done(i
);
3229 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3232 for (i
= 0; i
< n
; i
++)
3233 c
[i
] = exec_command_free_list(c
[i
]);
3236 typedef struct InvalidEnvInfo
{
3241 static void invalid_env(const char *p
, void *userdata
) {
3242 InvalidEnvInfo
*info
= userdata
;
3244 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3247 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3252 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3254 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3256 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3258 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3260 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3262 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3268 int exec_context_named_iofds(Unit
*unit
, const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3269 unsigned i
, targets
;
3270 const char* stdio_fdname
[3];
3276 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3277 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3278 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3280 for (i
= 0; i
< 3; i
++)
3281 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3283 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3285 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3286 if (named_iofds
[STDIN_FILENO
] < 0 &&
3287 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3288 stdio_fdname
[STDIN_FILENO
] &&
3289 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3291 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3294 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3295 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3296 stdio_fdname
[STDOUT_FILENO
] &&
3297 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3299 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3302 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3303 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3304 stdio_fdname
[STDERR_FILENO
] &&
3305 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3307 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3311 return targets
== 0 ? 0 : -ENOENT
;
3314 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
3315 char **i
, **r
= NULL
;
3320 STRV_FOREACH(i
, c
->environment_files
) {
3324 bool ignore
= false;
3326 _cleanup_globfree_ glob_t pglob
= {};
3335 if (!path_is_absolute(fn
)) {
3343 /* Filename supports globbing, take all matching files */
3344 k
= safe_glob(fn
, 0, &pglob
);
3353 /* When we don't match anything, -ENOENT should be returned */
3354 assert(pglob
.gl_pathc
> 0);
3356 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
3357 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3365 /* Log invalid environment variables with filename */
3367 InvalidEnvInfo info
= {
3369 .path
= pglob
.gl_pathv
[n
]
3372 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3380 m
= strv_env_merge(2, r
, p
);
3396 static bool tty_may_match_dev_console(const char *tty
) {
3397 _cleanup_free_
char *active
= NULL
;
3403 if (startswith(tty
, "/dev/"))
3406 /* trivial identity? */
3407 if (streq(tty
, "console"))
3410 console
= resolve_dev_console(&active
);
3411 /* if we could not resolve, assume it may */
3415 /* "tty0" means the active VC, so it may be the same sometimes */
3416 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3419 bool exec_context_may_touch_console(ExecContext
*ec
) {
3421 return (ec
->tty_reset
||
3423 ec
->tty_vt_disallocate
||
3424 is_terminal_input(ec
->std_input
) ||
3425 is_terminal_output(ec
->std_output
) ||
3426 is_terminal_output(ec
->std_error
)) &&
3427 tty_may_match_dev_console(exec_context_tty_path(ec
));
3430 static void strv_fprintf(FILE *f
, char **l
) {
3436 fprintf(f
, " %s", *g
);
3439 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3442 ExecDirectoryType dt
;
3448 prefix
= strempty(prefix
);
3452 "%sWorkingDirectory: %s\n"
3453 "%sRootDirectory: %s\n"
3454 "%sNonBlocking: %s\n"
3455 "%sPrivateTmp: %s\n"
3456 "%sPrivateDevices: %s\n"
3457 "%sProtectKernelTunables: %s\n"
3458 "%sProtectKernelModules: %s\n"
3459 "%sProtectControlGroups: %s\n"
3460 "%sPrivateNetwork: %s\n"
3461 "%sPrivateUsers: %s\n"
3462 "%sProtectHome: %s\n"
3463 "%sProtectSystem: %s\n"
3464 "%sMountAPIVFS: %s\n"
3465 "%sIgnoreSIGPIPE: %s\n"
3466 "%sMemoryDenyWriteExecute: %s\n"
3467 "%sRestrictRealtime: %s\n",
3469 prefix
, c
->working_directory
? c
->working_directory
: "/",
3470 prefix
, c
->root_directory
? c
->root_directory
: "/",
3471 prefix
, yes_no(c
->non_blocking
),
3472 prefix
, yes_no(c
->private_tmp
),
3473 prefix
, yes_no(c
->private_devices
),
3474 prefix
, yes_no(c
->protect_kernel_tunables
),
3475 prefix
, yes_no(c
->protect_kernel_modules
),
3476 prefix
, yes_no(c
->protect_control_groups
),
3477 prefix
, yes_no(c
->private_network
),
3478 prefix
, yes_no(c
->private_users
),
3479 prefix
, protect_home_to_string(c
->protect_home
),
3480 prefix
, protect_system_to_string(c
->protect_system
),
3481 prefix
, yes_no(c
->mount_apivfs
),
3482 prefix
, yes_no(c
->ignore_sigpipe
),
3483 prefix
, yes_no(c
->memory_deny_write_execute
),
3484 prefix
, yes_no(c
->restrict_realtime
));
3487 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3489 STRV_FOREACH(e
, c
->environment
)
3490 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3492 STRV_FOREACH(e
, c
->environment_files
)
3493 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3495 STRV_FOREACH(e
, c
->pass_environment
)
3496 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3498 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
3500 for (dt
= 0; dt
< _EXEC_DIRECTORY_MAX
; dt
++) {
3501 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
3503 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
3504 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
3512 if (c
->oom_score_adjust_set
)
3514 "%sOOMScoreAdjust: %i\n",
3515 prefix
, c
->oom_score_adjust
);
3517 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3519 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3520 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3521 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3522 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3525 if (c
->ioprio_set
) {
3526 _cleanup_free_
char *class_str
= NULL
;
3528 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3530 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
3532 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
3535 if (c
->cpu_sched_set
) {
3536 _cleanup_free_
char *policy_str
= NULL
;
3538 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3540 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
3543 "%sCPUSchedulingPriority: %i\n"
3544 "%sCPUSchedulingResetOnFork: %s\n",
3545 prefix
, c
->cpu_sched_priority
,
3546 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3550 fprintf(f
, "%sCPUAffinity:", prefix
);
3551 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3552 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3553 fprintf(f
, " %u", i
);
3557 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3558 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3561 "%sStandardInput: %s\n"
3562 "%sStandardOutput: %s\n"
3563 "%sStandardError: %s\n",
3564 prefix
, exec_input_to_string(c
->std_input
),
3565 prefix
, exec_output_to_string(c
->std_output
),
3566 prefix
, exec_output_to_string(c
->std_error
));
3572 "%sTTYVHangup: %s\n"
3573 "%sTTYVTDisallocate: %s\n",
3574 prefix
, c
->tty_path
,
3575 prefix
, yes_no(c
->tty_reset
),
3576 prefix
, yes_no(c
->tty_vhangup
),
3577 prefix
, yes_no(c
->tty_vt_disallocate
));
3579 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3580 c
->std_output
== EXEC_OUTPUT_KMSG
||
3581 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3582 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3583 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3584 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3585 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3586 c
->std_error
== EXEC_OUTPUT_KMSG
||
3587 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3588 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3589 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3590 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3592 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3594 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3596 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
3598 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3600 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
3603 if (c
->secure_bits
) {
3604 _cleanup_free_
char *str
= NULL
;
3606 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
3608 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
3611 if (c
->capability_bounding_set
!= CAP_ALL
) {
3612 _cleanup_free_
char *str
= NULL
;
3614 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
3616 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
3619 if (c
->capability_ambient_set
!= 0) {
3620 _cleanup_free_
char *str
= NULL
;
3622 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
3624 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
3628 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3630 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3632 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3634 if (strv_length(c
->supplementary_groups
) > 0) {
3635 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3636 strv_fprintf(f
, c
->supplementary_groups
);
3641 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3643 if (strv_length(c
->read_write_paths
) > 0) {
3644 fprintf(f
, "%sReadWritePaths:", prefix
);
3645 strv_fprintf(f
, c
->read_write_paths
);
3649 if (strv_length(c
->read_only_paths
) > 0) {
3650 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3651 strv_fprintf(f
, c
->read_only_paths
);
3655 if (strv_length(c
->inaccessible_paths
) > 0) {
3656 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3657 strv_fprintf(f
, c
->inaccessible_paths
);
3661 if (c
->n_bind_mounts
> 0)
3662 for (i
= 0; i
< c
->n_bind_mounts
; i
++) {
3663 fprintf(f
, "%s%s: %s:%s:%s\n", prefix
,
3664 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
3665 c
->bind_mounts
[i
].source
,
3666 c
->bind_mounts
[i
].destination
,
3667 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
3672 "%sUtmpIdentifier: %s\n",
3673 prefix
, c
->utmp_id
);
3675 if (c
->selinux_context
)
3677 "%sSELinuxContext: %s%s\n",
3678 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3680 if (c
->apparmor_profile
)
3682 "%sAppArmorProfile: %s%s\n",
3683 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3685 if (c
->smack_process_label
)
3687 "%sSmackProcessLabel: %s%s\n",
3688 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
3690 if (c
->personality
!= PERSONALITY_INVALID
)
3692 "%sPersonality: %s\n",
3693 prefix
, strna(personality_to_string(c
->personality
)));
3695 if (c
->syscall_filter
) {
3703 "%sSystemCallFilter: ",
3706 if (!c
->syscall_whitelist
)
3710 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3711 _cleanup_free_
char *name
= NULL
;
3718 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3719 fputs(strna(name
), f
);
3726 if (c
->syscall_archs
) {
3733 "%sSystemCallArchitectures:",
3737 SET_FOREACH(id
, c
->syscall_archs
, j
)
3738 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3743 if (exec_context_restrict_namespaces_set(c
)) {
3744 _cleanup_free_
char *s
= NULL
;
3746 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
3748 fprintf(f
, "%sRestrictNamespaces: %s\n",
3752 if (c
->syscall_errno
> 0)
3754 "%sSystemCallErrorNumber: %s\n",
3755 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3757 if (c
->apparmor_profile
)
3759 "%sAppArmorProfile: %s%s\n",
3760 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3763 bool exec_context_maintains_privileges(ExecContext
*c
) {
3766 /* Returns true if the process forked off would run under
3767 * an unchanged UID or as root. */
3772 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3778 int exec_context_get_effective_ioprio(ExecContext
*c
) {
3786 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
3788 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
3793 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3798 dual_timestamp_get(&s
->start_timestamp
);
3801 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3804 if (s
->pid
&& s
->pid
!= pid
)
3808 dual_timestamp_get(&s
->exit_timestamp
);
3814 if (context
->utmp_id
)
3815 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3817 exec_context_tty_reset(context
, NULL
);
3821 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3822 char buf
[FORMAT_TIMESTAMP_MAX
];
3830 prefix
= strempty(prefix
);
3833 "%sPID: "PID_FMT
"\n",
3836 if (dual_timestamp_is_set(&s
->start_timestamp
))
3838 "%sStart Timestamp: %s\n",
3839 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3841 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3843 "%sExit Timestamp: %s\n"
3845 "%sExit Status: %i\n",
3846 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3847 prefix
, sigchld_code_to_string(s
->code
),
3851 char *exec_command_line(char **argv
) {
3859 STRV_FOREACH(a
, argv
)
3867 STRV_FOREACH(a
, argv
) {
3874 if (strpbrk(*a
, WHITESPACE
)) {
3885 /* FIXME: this doesn't really handle arguments that have
3886 * spaces and ticks in them */
3891 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3892 _cleanup_free_
char *cmd
= NULL
;
3893 const char *prefix2
;
3898 prefix
= strempty(prefix
);
3899 prefix2
= strjoina(prefix
, "\t");
3901 cmd
= exec_command_line(c
->argv
);
3903 "%sCommand Line: %s\n",
3904 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3906 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3909 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3912 prefix
= strempty(prefix
);
3914 LIST_FOREACH(command
, c
, c
)
3915 exec_command_dump(c
, f
, prefix
);
3918 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3925 /* It's kind of important, that we keep the order here */
3926 LIST_FIND_TAIL(command
, *l
, end
);
3927 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3932 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3940 l
= strv_new_ap(path
, ap
);
3961 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3962 _cleanup_strv_free_
char **l
= NULL
;
3970 l
= strv_new_ap(path
, ap
);
3976 r
= strv_extend_strv(&c
->argv
, l
, false);
3984 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3989 *rt
= new0(ExecRuntime
, 1);
3994 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3999 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
4009 if (!c
->private_network
&& !c
->private_tmp
)
4012 r
= exec_runtime_allocate(rt
);
4016 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
4017 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
4021 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
4022 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
4030 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
4032 assert(r
->n_ref
> 0);
4038 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
4043 assert(r
->n_ref
> 0);
4050 free(r
->var_tmp_dir
);
4051 safe_close_pair(r
->netns_storage_socket
);
4055 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
4064 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
4066 if (rt
->var_tmp_dir
)
4067 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
4069 if (rt
->netns_storage_socket
[0] >= 0) {
4072 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4076 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
4079 if (rt
->netns_storage_socket
[1] >= 0) {
4082 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
4086 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
4092 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
4099 if (streq(key
, "tmp-dir")) {
4102 r
= exec_runtime_allocate(rt
);
4106 copy
= strdup(value
);
4110 free((*rt
)->tmp_dir
);
4111 (*rt
)->tmp_dir
= copy
;
4113 } else if (streq(key
, "var-tmp-dir")) {
4116 r
= exec_runtime_allocate(rt
);
4120 copy
= strdup(value
);
4124 free((*rt
)->var_tmp_dir
);
4125 (*rt
)->var_tmp_dir
= copy
;
4127 } else if (streq(key
, "netns-socket-0")) {
4130 r
= exec_runtime_allocate(rt
);
4134 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4135 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4137 safe_close((*rt
)->netns_storage_socket
[0]);
4138 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4140 } else if (streq(key
, "netns-socket-1")) {
4143 r
= exec_runtime_allocate(rt
);
4147 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4148 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4150 safe_close((*rt
)->netns_storage_socket
[1]);
4151 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
4159 static void *remove_tmpdir_thread(void *p
) {
4160 _cleanup_free_
char *path
= p
;
4162 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4166 void exec_runtime_destroy(ExecRuntime
*rt
) {
4172 /* If there are multiple users of this, let's leave the stuff around */
4177 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4179 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4181 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4188 if (rt
->var_tmp_dir
) {
4189 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4191 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4193 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4194 free(rt
->var_tmp_dir
);
4197 rt
->var_tmp_dir
= NULL
;
4200 safe_close_pair(rt
->netns_storage_socket
);
4203 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4204 [EXEC_INPUT_NULL
] = "null",
4205 [EXEC_INPUT_TTY
] = "tty",
4206 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4207 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4208 [EXEC_INPUT_SOCKET
] = "socket",
4209 [EXEC_INPUT_NAMED_FD
] = "fd",
4212 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4214 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4215 [EXEC_OUTPUT_INHERIT
] = "inherit",
4216 [EXEC_OUTPUT_NULL
] = "null",
4217 [EXEC_OUTPUT_TTY
] = "tty",
4218 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4219 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4220 [EXEC_OUTPUT_KMSG
] = "kmsg",
4221 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4222 [EXEC_OUTPUT_JOURNAL
] = "journal",
4223 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4224 [EXEC_OUTPUT_SOCKET
] = "socket",
4225 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4228 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4230 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4231 [EXEC_UTMP_INIT
] = "init",
4232 [EXEC_UTMP_LOGIN
] = "login",
4233 [EXEC_UTMP_USER
] = "user",
4236 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
4238 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
4239 [EXEC_PRESERVE_NO
] = "no",
4240 [EXEC_PRESERVE_YES
] = "yes",
4241 [EXEC_PRESERVE_RESTART
] = "restart",
4244 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
4246 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_MAX
] = {
4247 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
4248 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
4249 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
4250 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
4251 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
4254 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);