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_fds
, unsigned n_socket_fds
, bool nonblock
) {
169 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
170 * O_NONBLOCK only applies to socket activation though. */
172 for (i
= 0; i
< n_fds
; i
++) {
174 if (i
< n_socket_fds
) {
175 r
= fd_nonblock(fds
[i
], nonblock
);
180 /* We unconditionally drop FD_CLOEXEC from the fds,
181 * since after all we want to pass these fds to our
184 r
= fd_cloexec(fds
[i
], false);
192 static const char *exec_context_tty_path(const ExecContext
*context
) {
195 if (context
->stdio_as_fds
)
198 if (context
->tty_path
)
199 return context
->tty_path
;
201 return "/dev/console";
204 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
209 path
= exec_context_tty_path(context
);
211 if (context
->tty_vhangup
) {
212 if (p
&& p
->stdin_fd
>= 0)
213 (void) terminal_vhangup_fd(p
->stdin_fd
);
215 (void) terminal_vhangup(path
);
218 if (context
->tty_reset
) {
219 if (p
&& p
->stdin_fd
>= 0)
220 (void) reset_terminal_fd(p
->stdin_fd
, true);
222 (void) reset_terminal(path
);
225 if (context
->tty_vt_disallocate
&& path
)
226 (void) vt_disallocate(path
);
229 static bool is_terminal_input(ExecInput i
) {
232 EXEC_INPUT_TTY_FORCE
,
233 EXEC_INPUT_TTY_FAIL
);
236 static bool is_terminal_output(ExecOutput o
) {
239 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
240 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
241 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
244 static bool exec_context_needs_term(const ExecContext
*c
) {
247 /* Return true if the execution context suggests we should set $TERM to something useful. */
249 if (is_terminal_input(c
->std_input
))
252 if (is_terminal_output(c
->std_output
))
255 if (is_terminal_output(c
->std_error
))
258 return !!c
->tty_path
;
261 static int open_null_as(int flags
, int nfd
) {
266 fd
= open("/dev/null", flags
|O_NOCTTY
);
271 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
279 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
280 union sockaddr_union sa
= {
281 .un
.sun_family
= AF_UNIX
,
282 .un
.sun_path
= "/run/systemd/journal/stdout",
284 uid_t olduid
= UID_INVALID
;
285 gid_t oldgid
= GID_INVALID
;
288 if (gid
!= GID_INVALID
) {
296 if (uid
!= UID_INVALID
) {
306 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
310 /* If we fail to restore the uid or gid, things will likely
311 fail later on. This should only happen if an LSM interferes. */
313 if (uid
!= UID_INVALID
)
314 (void) seteuid(olduid
);
317 if (gid
!= GID_INVALID
)
318 (void) setegid(oldgid
);
323 static int connect_logger_as(
325 const ExecContext
*context
,
335 assert(output
< _EXEC_OUTPUT_MAX
);
339 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
343 r
= connect_journal_socket(fd
, uid
, gid
);
347 if (shutdown(fd
, SHUT_RD
) < 0) {
352 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
362 context
->syslog_identifier
? context
->syslog_identifier
: ident
,
364 context
->syslog_priority
,
365 !!context
->syslog_level_prefix
,
366 output
== EXEC_OUTPUT_SYSLOG
|| output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
367 output
== EXEC_OUTPUT_KMSG
|| output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
,
368 is_terminal_output(output
));
373 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
378 static int open_terminal_as(const char *path
, mode_t mode
, int nfd
) {
384 fd
= open_terminal(path
, mode
| O_NOCTTY
);
389 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
397 static int fixup_input(ExecInput std_input
, int socket_fd
, bool apply_tty_stdin
) {
399 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
400 return EXEC_INPUT_NULL
;
402 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
403 return EXEC_INPUT_NULL
;
408 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
410 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
411 return EXEC_OUTPUT_INHERIT
;
416 static int setup_input(
417 const ExecContext
*context
,
418 const ExecParameters
*params
,
420 int named_iofds
[3]) {
427 if (params
->stdin_fd
>= 0) {
428 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
431 /* Try to make this the controlling tty, if it is a tty, and reset it */
432 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
433 (void) reset_terminal_fd(STDIN_FILENO
, true);
438 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
442 case EXEC_INPUT_NULL
:
443 return open_null_as(O_RDONLY
, STDIN_FILENO
);
446 case EXEC_INPUT_TTY_FORCE
:
447 case EXEC_INPUT_TTY_FAIL
: {
450 fd
= acquire_terminal(exec_context_tty_path(context
),
451 i
== EXEC_INPUT_TTY_FAIL
,
452 i
== EXEC_INPUT_TTY_FORCE
,
458 if (fd
!= STDIN_FILENO
) {
459 r
= dup2(fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
467 case EXEC_INPUT_SOCKET
:
468 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
470 case EXEC_INPUT_NAMED_FD
:
471 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
472 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
475 assert_not_reached("Unknown input type");
479 static int setup_output(
481 const ExecContext
*context
,
482 const ExecParameters
*params
,
489 dev_t
*journal_stream_dev
,
490 ino_t
*journal_stream_ino
) {
500 assert(journal_stream_dev
);
501 assert(journal_stream_ino
);
503 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
505 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
508 return STDOUT_FILENO
;
511 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
512 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
515 return STDERR_FILENO
;
518 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
519 o
= fixup_output(context
->std_output
, socket_fd
);
521 if (fileno
== STDERR_FILENO
) {
523 e
= fixup_output(context
->std_error
, socket_fd
);
525 /* This expects the input and output are already set up */
527 /* Don't change the stderr file descriptor if we inherit all
528 * the way and are not on a tty */
529 if (e
== EXEC_OUTPUT_INHERIT
&&
530 o
== EXEC_OUTPUT_INHERIT
&&
531 i
== EXEC_INPUT_NULL
&&
532 !is_terminal_input(context
->std_input
) &&
536 /* Duplicate from stdout if possible */
537 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
538 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
542 } else if (o
== EXEC_OUTPUT_INHERIT
) {
543 /* If input got downgraded, inherit the original value */
544 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
545 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
547 /* If the input is connected to anything that's not a /dev/null, inherit that... */
548 if (i
!= EXEC_INPUT_NULL
)
549 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
551 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
555 /* We need to open /dev/null here anew, to get the right access mode. */
556 return open_null_as(O_WRONLY
, fileno
);
561 case EXEC_OUTPUT_NULL
:
562 return open_null_as(O_WRONLY
, fileno
);
564 case EXEC_OUTPUT_TTY
:
565 if (is_terminal_input(i
))
566 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
568 /* We don't reset the terminal if this is just about output */
569 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
571 case EXEC_OUTPUT_SYSLOG
:
572 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
573 case EXEC_OUTPUT_KMSG
:
574 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
575 case EXEC_OUTPUT_JOURNAL
:
576 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
577 r
= connect_logger_as(unit
, context
, o
, ident
, fileno
, uid
, gid
);
579 log_unit_error_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
580 r
= open_null_as(O_WRONLY
, fileno
);
584 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
585 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
586 * services to detect whether they are connected to the journal or not. */
588 if (fstat(fileno
, &st
) >= 0) {
589 *journal_stream_dev
= st
.st_dev
;
590 *journal_stream_ino
= st
.st_ino
;
595 case EXEC_OUTPUT_SOCKET
:
596 assert(socket_fd
>= 0);
597 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
599 case EXEC_OUTPUT_NAMED_FD
:
600 (void) fd_nonblock(named_iofds
[fileno
], false);
601 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
604 assert_not_reached("Unknown error type");
608 static int chown_terminal(int fd
, uid_t uid
) {
613 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
617 /* This might fail. What matters are the results. */
618 (void) fchown(fd
, uid
, -1);
619 (void) fchmod(fd
, TTY_MODE
);
621 if (fstat(fd
, &st
) < 0)
624 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
630 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
631 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
634 assert(_saved_stdin
);
635 assert(_saved_stdout
);
637 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
641 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
642 if (saved_stdout
< 0)
645 fd
= acquire_terminal(vc
, false, false, false, DEFAULT_CONFIRM_USEC
);
649 r
= chown_terminal(fd
, getuid());
653 r
= reset_terminal_fd(fd
, true);
657 if (dup2(fd
, STDIN_FILENO
) < 0)
660 if (dup2(fd
, STDOUT_FILENO
) < 0)
667 *_saved_stdin
= saved_stdin
;
668 *_saved_stdout
= saved_stdout
;
670 saved_stdin
= saved_stdout
= -1;
675 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
678 if (err
== -ETIMEDOUT
)
679 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
682 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
686 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
687 _cleanup_close_
int fd
= -1;
691 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
695 write_confirm_error_fd(err
, fd
, u
);
698 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
702 assert(saved_stdout
);
706 if (*saved_stdin
>= 0)
707 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
710 if (*saved_stdout
>= 0)
711 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
714 *saved_stdin
= safe_close(*saved_stdin
);
715 *saved_stdout
= safe_close(*saved_stdout
);
721 CONFIRM_PRETEND_FAILURE
= -1,
722 CONFIRM_PRETEND_SUCCESS
= 0,
726 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
727 int saved_stdout
= -1, saved_stdin
= -1, r
;
728 _cleanup_free_
char *e
= NULL
;
731 /* For any internal errors, assume a positive response. */
732 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
734 write_confirm_error(r
, vc
, u
);
735 return CONFIRM_EXECUTE
;
738 /* confirm_spawn might have been disabled while we were sleeping. */
739 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
744 e
= ellipsize(cmdline
, 60, 100);
752 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
754 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
761 printf("Resuming normal execution.\n");
762 manager_disable_confirm_spawn();
766 unit_dump(u
, stdout
, " ");
767 continue; /* ask again */
769 printf("Failing execution.\n");
770 r
= CONFIRM_PRETEND_FAILURE
;
773 printf(" c - continue, proceed without asking anymore\n"
774 " D - dump, show the state of the unit\n"
775 " f - fail, don't execute the command and pretend it failed\n"
777 " i - info, show a short summary of the unit\n"
778 " j - jobs, show jobs that are in progress\n"
779 " s - skip, don't execute the command and pretend it succeeded\n"
780 " y - yes, execute the command\n");
781 continue; /* ask again */
783 printf(" Description: %s\n"
786 u
->id
, u
->description
, cmdline
);
787 continue; /* ask again */
789 manager_dump_jobs(u
->manager
, stdout
, " ");
790 continue; /* ask again */
792 /* 'n' was removed in favor of 'f'. */
793 printf("Didn't understand 'n', did you mean 'f'?\n");
794 continue; /* ask again */
796 printf("Skipping execution.\n");
797 r
= CONFIRM_PRETEND_SUCCESS
;
803 assert_not_reached("Unhandled choice");
809 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
813 static int get_fixed_user(const ExecContext
*c
, const char **user
,
814 uid_t
*uid
, gid_t
*gid
,
815 const char **home
, const char **shell
) {
824 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
825 * (i.e. are "/" or "/bin/nologin"). */
828 r
= get_user_creds_clean(&name
, uid
, gid
, home
, shell
);
836 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
846 r
= get_group_creds(&name
, gid
);
854 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
855 const char *group
, gid_t gid
,
856 gid_t
**supplementary_gids
, int *ngids
) {
860 bool keep_groups
= false;
861 gid_t
*groups
= NULL
;
862 _cleanup_free_ gid_t
*l_gids
= NULL
;
867 * If user is given, then lookup GID and supplementary groups list.
868 * We avoid NSS lookups for gid=0. Also we have to initialize groups
869 * here and as early as possible so we keep the list of supplementary
870 * groups of the caller.
872 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
873 /* First step, initialize groups from /etc/groups */
874 if (initgroups(user
, gid
) < 0)
880 if (!c
->supplementary_groups
)
884 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
885 * be positive, otherwise fail.
888 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
889 if (ngroups_max
<= 0) {
893 return -EOPNOTSUPP
; /* For all other values */
896 l_gids
= new(gid_t
, ngroups_max
);
902 * Lookup the list of groups that the user belongs to, we
903 * avoid NSS lookups here too for gid=0.
906 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
911 STRV_FOREACH(i
, c
->supplementary_groups
) {
914 if (k
>= ngroups_max
)
918 r
= get_group_creds(&g
, l_gids
+k
);
926 * Sets ngids to zero to drop all supplementary groups, happens
927 * when we are under root and SupplementaryGroups= is empty.
934 /* Otherwise get the final list of supplementary groups */
935 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
939 *supplementary_gids
= groups
;
947 static int enforce_groups(const ExecContext
*context
, gid_t gid
,
948 gid_t
*supplementary_gids
, int ngids
) {
953 /* Handle SupplementaryGroups= even if it is empty */
954 if (context
->supplementary_groups
) {
955 r
= maybe_setgroups(ngids
, supplementary_gids
);
960 if (gid_is_valid(gid
)) {
961 /* Then set our gids */
962 if (setresgid(gid
, gid
, gid
) < 0)
969 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
972 if (!uid_is_valid(uid
))
975 /* Sets (but doesn't look up) the uid and make sure we keep the
976 * capabilities while doing so. */
978 if (context
->capability_ambient_set
!= 0) {
980 /* First step: If we need to keep capabilities but
981 * drop privileges we need to make sure we keep our
982 * caps, while we drop privileges. */
984 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
986 if (prctl(PR_GET_SECUREBITS
) != sb
)
987 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
992 /* Second step: actually set the uids */
993 if (setresuid(uid
, uid
, uid
) < 0)
996 /* At this point we should have all necessary capabilities but
997 are otherwise a normal user. However, the caps might got
998 corrupted due to the setresuid() so we need clean them up
999 later. This is done outside of this call. */
1006 static int null_conv(
1008 const struct pam_message
**msg
,
1009 struct pam_response
**resp
,
1010 void *appdata_ptr
) {
1012 /* We don't support conversations */
1014 return PAM_CONV_ERR
;
1019 static int setup_pam(
1026 int fds
[], unsigned n_fds
) {
1030 static const struct pam_conv conv
= {
1035 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1036 pam_handle_t
*handle
= NULL
;
1038 int pam_code
= PAM_SUCCESS
, r
;
1039 char **nv
, **e
= NULL
;
1040 bool close_session
= false;
1041 pid_t pam_pid
= 0, parent_pid
;
1048 /* We set up PAM in the parent process, then fork. The child
1049 * will then stay around until killed via PR_GET_PDEATHSIG or
1050 * systemd via the cgroup logic. It will then remove the PAM
1051 * session again. The parent process will exec() the actual
1052 * daemon. We do things this way to ensure that the main PID
1053 * of the daemon is the one we initially fork()ed. */
1055 r
= barrier_create(&barrier
);
1059 if (log_get_max_level() < LOG_DEBUG
)
1060 flags
|= PAM_SILENT
;
1062 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1063 if (pam_code
!= PAM_SUCCESS
) {
1069 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1070 if (pam_code
!= PAM_SUCCESS
)
1074 STRV_FOREACH(nv
, *env
) {
1075 pam_code
= pam_putenv(handle
, *nv
);
1076 if (pam_code
!= PAM_SUCCESS
)
1080 pam_code
= pam_acct_mgmt(handle
, flags
);
1081 if (pam_code
!= PAM_SUCCESS
)
1084 pam_code
= pam_open_session(handle
, flags
);
1085 if (pam_code
!= PAM_SUCCESS
)
1088 close_session
= true;
1090 e
= pam_getenvlist(handle
);
1092 pam_code
= PAM_BUF_ERR
;
1096 /* Block SIGTERM, so that we know that it won't get lost in
1099 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1101 parent_pid
= getpid();
1110 int sig
, ret
= EXIT_PAM
;
1112 /* The child's job is to reset the PAM session on
1114 barrier_set_role(&barrier
, BARRIER_CHILD
);
1116 /* This string must fit in 10 chars (i.e. the length
1117 * of "/sbin/init"), to look pretty in /bin/ps */
1118 rename_process("(sd-pam)");
1120 /* Make sure we don't keep open the passed fds in this
1121 child. We assume that otherwise only those fds are
1122 open here that have been opened by PAM. */
1123 close_many(fds
, n_fds
);
1125 /* Drop privileges - we don't need any to pam_close_session
1126 * and this will make PR_SET_PDEATHSIG work in most cases.
1127 * If this fails, ignore the error - but expect sd-pam threads
1128 * to fail to exit normally */
1130 r
= maybe_setgroups(0, NULL
);
1132 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1133 if (setresgid(gid
, gid
, gid
) < 0)
1134 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1135 if (setresuid(uid
, uid
, uid
) < 0)
1136 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1138 (void) ignore_signals(SIGPIPE
, -1);
1140 /* Wait until our parent died. This will only work if
1141 * the above setresuid() succeeds, otherwise the kernel
1142 * will not allow unprivileged parents kill their privileged
1143 * children this way. We rely on the control groups kill logic
1144 * to do the rest for us. */
1145 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1148 /* Tell the parent that our setup is done. This is especially
1149 * important regarding dropping privileges. Otherwise, unit
1150 * setup might race against our setresuid(2) call.
1152 * If the parent aborted, we'll detect this below, hence ignore
1153 * return failure here. */
1154 (void) barrier_place(&barrier
);
1156 /* Check if our parent process might already have died? */
1157 if (getppid() == parent_pid
) {
1160 assert_se(sigemptyset(&ss
) >= 0);
1161 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1164 if (sigwait(&ss
, &sig
) < 0) {
1171 assert(sig
== SIGTERM
);
1176 /* If our parent died we'll end the session */
1177 if (getppid() != parent_pid
) {
1178 pam_code
= pam_close_session(handle
, flags
);
1179 if (pam_code
!= PAM_SUCCESS
)
1186 pam_end(handle
, pam_code
| flags
);
1190 barrier_set_role(&barrier
, BARRIER_PARENT
);
1192 /* If the child was forked off successfully it will do all the
1193 * cleanups, so forget about the handle here. */
1196 /* Unblock SIGTERM again in the parent */
1197 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1199 /* We close the log explicitly here, since the PAM modules
1200 * might have opened it, but we don't want this fd around. */
1203 /* Synchronously wait for the child to initialize. We don't care for
1204 * errors as we cannot recover. However, warn loudly if it happens. */
1205 if (!barrier_place_and_sync(&barrier
))
1206 log_error("PAM initialization failed");
1214 if (pam_code
!= PAM_SUCCESS
) {
1215 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1216 r
= -EPERM
; /* PAM errors do not map to errno */
1218 log_error_errno(r
, "PAM failed: %m");
1222 pam_code
= pam_close_session(handle
, flags
);
1224 pam_end(handle
, pam_code
| flags
);
1236 static void rename_process_from_path(const char *path
) {
1237 char process_name
[11];
1241 /* This resulting string must fit in 10 chars (i.e. the length
1242 * of "/sbin/init") to look pretty in /bin/ps */
1246 rename_process("(...)");
1252 /* The end of the process name is usually more
1253 * interesting, since the first bit might just be
1259 process_name
[0] = '(';
1260 memcpy(process_name
+1, p
, l
);
1261 process_name
[1+l
] = ')';
1262 process_name
[1+l
+1] = 0;
1264 rename_process(process_name
);
1267 static bool context_has_address_families(const ExecContext
*c
) {
1270 return c
->address_families_whitelist
||
1271 !set_isempty(c
->address_families
);
1274 static bool context_has_syscall_filters(const ExecContext
*c
) {
1277 return c
->syscall_whitelist
||
1278 !set_isempty(c
->syscall_filter
);
1281 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1284 if (c
->no_new_privileges
)
1287 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1290 /* We need NNP if we have any form of seccomp and are unprivileged */
1291 return context_has_address_families(c
) ||
1292 c
->memory_deny_write_execute
||
1293 c
->restrict_realtime
||
1294 exec_context_restrict_namespaces_set(c
) ||
1295 c
->protect_kernel_tunables
||
1296 c
->protect_kernel_modules
||
1297 c
->private_devices
||
1298 context_has_syscall_filters(c
) ||
1299 !set_isempty(c
->syscall_archs
);
1304 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1306 if (is_seccomp_available())
1310 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1315 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
) {
1316 uint32_t negative_action
, default_action
, action
;
1321 if (!context_has_syscall_filters(c
))
1324 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1327 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1329 if (c
->syscall_whitelist
) {
1330 default_action
= negative_action
;
1331 action
= SCMP_ACT_ALLOW
;
1333 default_action
= SCMP_ACT_ALLOW
;
1334 action
= negative_action
;
1337 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
);
1340 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1344 if (set_isempty(c
->syscall_archs
))
1347 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1350 return seccomp_restrict_archs(c
->syscall_archs
);
1353 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1357 if (!context_has_address_families(c
))
1360 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1363 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1366 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1370 if (!c
->memory_deny_write_execute
)
1373 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1376 return seccomp_memory_deny_write_execute();
1379 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1383 if (!c
->restrict_realtime
)
1386 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1389 return seccomp_restrict_realtime();
1392 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1396 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1397 * let's protect even those systems where this is left on in the kernel. */
1399 if (!c
->protect_kernel_tunables
)
1402 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1405 return seccomp_protect_sysctl();
1408 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1412 /* Turn off module syscalls on ProtectKernelModules=yes */
1414 if (!c
->protect_kernel_modules
)
1417 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1420 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
));
1423 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1427 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1429 if (!c
->private_devices
)
1432 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1435 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
));
1438 static int apply_restrict_namespaces(Unit
*u
, const ExecContext
*c
) {
1442 if (!exec_context_restrict_namespaces_set(c
))
1445 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1448 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1453 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1456 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1457 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1459 if (idle_pipe
[0] >= 0) {
1462 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1464 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1467 /* Signal systemd that we are bored and want to continue. */
1468 n
= write(idle_pipe
[3], "x", 1);
1470 /* Wait for systemd to react to the signal above. */
1471 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1474 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1478 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1481 static int build_environment(
1483 const ExecContext
*c
,
1484 const ExecParameters
*p
,
1487 const char *username
,
1489 dev_t journal_stream_dev
,
1490 ino_t journal_stream_ino
,
1493 _cleanup_strv_free_
char **our_env
= NULL
;
1501 our_env
= new0(char*, 14);
1506 _cleanup_free_
char *joined
= NULL
;
1508 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid()) < 0)
1510 our_env
[n_env
++] = x
;
1512 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1514 our_env
[n_env
++] = x
;
1516 joined
= strv_join(p
->fd_names
, ":");
1520 x
= strjoin("LISTEN_FDNAMES=", joined
);
1523 our_env
[n_env
++] = x
;
1526 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1527 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid()) < 0)
1529 our_env
[n_env
++] = x
;
1531 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1533 our_env
[n_env
++] = x
;
1536 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1537 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1538 * check the database directly. */
1539 if (unit_has_name(u
, SPECIAL_DBUS_SERVICE
)) {
1540 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1543 our_env
[n_env
++] = x
;
1547 x
= strappend("HOME=", home
);
1550 our_env
[n_env
++] = x
;
1554 x
= strappend("LOGNAME=", username
);
1557 our_env
[n_env
++] = x
;
1559 x
= strappend("USER=", username
);
1562 our_env
[n_env
++] = x
;
1566 x
= strappend("SHELL=", shell
);
1569 our_env
[n_env
++] = x
;
1572 if (!sd_id128_is_null(u
->invocation_id
)) {
1573 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1576 our_env
[n_env
++] = x
;
1579 if (exec_context_needs_term(c
)) {
1580 const char *tty_path
, *term
= NULL
;
1582 tty_path
= exec_context_tty_path(c
);
1584 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1585 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1586 * passes to PID 1 ends up all the way in the console login shown. */
1588 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1589 term
= getenv("TERM");
1591 term
= default_term_for_tty(tty_path
);
1593 x
= strappend("TERM=", term
);
1596 our_env
[n_env
++] = x
;
1599 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1600 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1603 our_env
[n_env
++] = x
;
1606 our_env
[n_env
++] = NULL
;
1607 assert(n_env
<= 12);
1615 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1616 _cleanup_strv_free_
char **pass_env
= NULL
;
1617 size_t n_env
= 0, n_bufsize
= 0;
1620 STRV_FOREACH(i
, c
->pass_environment
) {
1621 _cleanup_free_
char *x
= NULL
;
1627 x
= strjoin(*i
, "=", v
);
1630 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1632 pass_env
[n_env
++] = x
;
1633 pass_env
[n_env
] = NULL
;
1643 static bool exec_needs_mount_namespace(
1644 const ExecContext
*context
,
1645 const ExecParameters
*params
,
1646 ExecRuntime
*runtime
) {
1651 if (context
->root_image
)
1654 if (!strv_isempty(context
->read_write_paths
) ||
1655 !strv_isempty(context
->read_only_paths
) ||
1656 !strv_isempty(context
->inaccessible_paths
))
1659 if (context
->n_bind_mounts
> 0)
1662 if (context
->mount_flags
!= 0)
1665 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1668 if (context
->private_devices
||
1669 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1670 context
->protect_home
!= PROTECT_HOME_NO
||
1671 context
->protect_kernel_tunables
||
1672 context
->protect_kernel_modules
||
1673 context
->protect_control_groups
)
1676 if (context
->mount_apivfs
&& (context
->root_image
|| context
->root_directory
))
1682 static int setup_private_users(uid_t uid
, gid_t gid
) {
1683 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1684 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1685 _cleanup_close_
int unshare_ready_fd
= -1;
1686 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1692 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1693 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1694 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1695 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1696 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1697 * continues execution normally. */
1699 if (uid
!= 0 && uid_is_valid(uid
)) {
1700 r
= asprintf(&uid_map
,
1701 "0 0 1\n" /* Map root → root */
1702 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1707 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1712 if (gid
!= 0 && gid_is_valid(gid
)) {
1713 r
= asprintf(&gid_map
,
1714 "0 0 1\n" /* Map root → root */
1715 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1720 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1725 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1727 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1728 if (unshare_ready_fd
< 0)
1731 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1733 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1741 _cleanup_close_
int fd
= -1;
1745 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1746 * here, after the parent opened its own user namespace. */
1749 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1751 /* Wait until the parent unshared the user namespace */
1752 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1757 /* Disable the setgroups() system call in the child user namespace, for good. */
1758 a
= procfs_file_alloca(ppid
, "setgroups");
1759 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1761 if (errno
!= ENOENT
) {
1766 /* If the file is missing the kernel is too old, let's continue anyway. */
1768 if (write(fd
, "deny\n", 5) < 0) {
1773 fd
= safe_close(fd
);
1776 /* First write the GID map */
1777 a
= procfs_file_alloca(ppid
, "gid_map");
1778 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1783 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1787 fd
= safe_close(fd
);
1789 /* The write the UID map */
1790 a
= procfs_file_alloca(ppid
, "uid_map");
1791 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1796 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1801 _exit(EXIT_SUCCESS
);
1804 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1805 _exit(EXIT_FAILURE
);
1808 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1810 if (unshare(CLONE_NEWUSER
) < 0)
1813 /* Let the child know that the namespace is ready now */
1814 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1817 /* Try to read an error code from the child */
1818 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1821 if (n
== sizeof(r
)) { /* an error code was sent to us */
1826 if (n
!= 0) /* on success we should have read 0 bytes */
1829 r
= wait_for_terminate(pid
, &si
);
1834 /* If something strange happened with the child, let's consider this fatal, too */
1835 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1841 static int setup_runtime_directory(
1842 const ExecContext
*context
,
1843 const ExecParameters
*params
,
1853 STRV_FOREACH(rt
, context
->runtime_directory
) {
1854 _cleanup_free_
char *p
;
1856 p
= strjoin(params
->runtime_prefix
, "/", *rt
);
1860 r
= mkdir_p_label(p
, context
->runtime_directory_mode
);
1864 r
= chmod_and_chown(p
, context
->runtime_directory_mode
, uid
, gid
);
1872 static int setup_smack(
1873 const ExecContext
*context
,
1874 const ExecCommand
*command
) {
1882 if (!mac_smack_use())
1885 if (context
->smack_process_label
) {
1886 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1890 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1892 _cleanup_free_
char *exec_label
= NULL
;
1894 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1895 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1898 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
1908 static int compile_read_write_paths(
1909 const ExecContext
*context
,
1910 const ExecParameters
*params
,
1913 _cleanup_strv_free_
char **l
= NULL
;
1916 /* Compile the list of writable paths. This is the combination of
1917 * the explicitly configured paths, plus all runtime directories. */
1919 if (strv_isempty(context
->read_write_paths
) &&
1920 strv_isempty(context
->runtime_directory
)) {
1921 *ret
= NULL
; /* NOP if neither is set */
1925 l
= strv_copy(context
->read_write_paths
);
1929 STRV_FOREACH(rt
, context
->runtime_directory
) {
1932 s
= strjoin(params
->runtime_prefix
, "/", *rt
);
1936 if (strv_consume(&l
, s
) < 0)
1946 static int apply_mount_namespace(
1948 ExecCommand
*command
,
1949 const ExecContext
*context
,
1950 const ExecParameters
*params
,
1951 ExecRuntime
*runtime
) {
1953 _cleanup_strv_free_
char **rw
= NULL
;
1954 char *tmp
= NULL
, *var
= NULL
;
1955 const char *root_dir
= NULL
, *root_image
= NULL
;
1956 NameSpaceInfo ns_info
= {
1957 .ignore_protect_paths
= false,
1958 .private_dev
= context
->private_devices
,
1959 .protect_control_groups
= context
->protect_control_groups
,
1960 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
1961 .protect_kernel_modules
= context
->protect_kernel_modules
,
1962 .mount_apivfs
= context
->mount_apivfs
,
1964 bool apply_restrictions
;
1969 /* The runtime struct only contains the parent of the private /tmp,
1970 * which is non-accessible to world users. Inside of it there's a /tmp
1971 * that is sticky, and that's the one we want to use here. */
1973 if (context
->private_tmp
&& runtime
) {
1974 if (runtime
->tmp_dir
)
1975 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
1976 if (runtime
->var_tmp_dir
)
1977 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
1980 r
= compile_read_write_paths(context
, params
, &rw
);
1984 if (params
->flags
& EXEC_APPLY_CHROOT
) {
1985 root_image
= context
->root_image
;
1988 root_dir
= context
->root_directory
;
1992 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
1993 * sandbox info, otherwise enforce it, don't ignore protected paths and
1994 * fail if we are enable to apply the sandbox inside the mount namespace.
1996 if (!context
->dynamic_user
&& root_dir
)
1997 ns_info
.ignore_protect_paths
= true;
1999 apply_restrictions
= (params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
;
2001 r
= setup_namespace(root_dir
, root_image
,
2003 apply_restrictions
? context
->read_only_paths
: NULL
,
2004 apply_restrictions
? context
->inaccessible_paths
: NULL
,
2005 context
->bind_mounts
,
2006 context
->n_bind_mounts
,
2009 apply_restrictions
? context
->protect_home
: PROTECT_HOME_NO
,
2010 apply_restrictions
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2011 context
->mount_flags
,
2012 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2014 /* If we couldn't set up the namespace this is probably due to a
2015 * missing capability. In this case, silently proceeed. */
2016 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2018 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2026 static int apply_working_directory(
2027 const ExecContext
*context
,
2028 const ExecParameters
*params
,
2030 const bool needs_mount_ns
,
2036 assert(exit_status
);
2038 if (context
->working_directory_home
) {
2041 *exit_status
= EXIT_CHDIR
;
2047 } else if (context
->working_directory
)
2048 wd
= context
->working_directory
;
2052 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2053 if (!needs_mount_ns
&& context
->root_directory
)
2054 if (chroot(context
->root_directory
) < 0) {
2055 *exit_status
= EXIT_CHROOT
;
2061 d
= prefix_roota(context
->root_directory
, wd
);
2063 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2064 *exit_status
= EXIT_CHDIR
;
2071 static int setup_keyring(Unit
*u
, const ExecParameters
*p
, uid_t uid
, gid_t gid
) {
2072 key_serial_t keyring
;
2077 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2078 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2079 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2080 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2081 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2082 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2084 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2087 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2088 if (keyring
== -1) {
2089 if (errno
== ENOSYS
)
2090 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2091 else if (IN_SET(errno
, EACCES
, EPERM
))
2092 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2093 else if (errno
== EDQUOT
)
2094 log_debug_errno(errno
, "Out of kernel keyrings to allocate, ignoring.");
2096 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2101 /* Populate they keyring with the invocation ID by default. */
2102 if (!sd_id128_is_null(u
->invocation_id
)) {
2105 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2107 log_debug_errno(errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2109 if (keyctl(KEYCTL_SETPERM
, key
,
2110 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2111 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2112 return log_error_errno(errno
, "Failed to restrict invocation ID permission: %m");
2116 /* And now, make the keyring owned by the service's user */
2117 if (uid_is_valid(uid
) || gid_is_valid(gid
))
2118 if (keyctl(KEYCTL_CHOWN
, keyring
, uid
, gid
, 0) < 0)
2119 return log_error_errno(errno
, "Failed to change ownership of session keyring: %m");
2124 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
2132 array
[(*n
)++] = pair
[0];
2134 array
[(*n
)++] = pair
[1];
2137 static int close_remaining_fds(
2138 const ExecParameters
*params
,
2139 ExecRuntime
*runtime
,
2140 DynamicCreds
*dcreds
,
2143 int *fds
, unsigned n_fds
) {
2145 unsigned n_dont_close
= 0;
2146 int dont_close
[n_fds
+ 12];
2150 if (params
->stdin_fd
>= 0)
2151 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2152 if (params
->stdout_fd
>= 0)
2153 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2154 if (params
->stderr_fd
>= 0)
2155 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2158 dont_close
[n_dont_close
++] = socket_fd
;
2160 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2161 n_dont_close
+= n_fds
;
2165 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2169 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2171 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2174 if (user_lookup_fd
>= 0)
2175 dont_close
[n_dont_close
++] = user_lookup_fd
;
2177 return close_all_fds(dont_close
, n_dont_close
);
2180 static int send_user_lookup(
2188 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2189 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2192 if (user_lookup_fd
< 0)
2195 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2198 if (writev(user_lookup_fd
,
2200 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2201 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2202 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2208 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2215 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2220 if (!c
->working_directory_home
)
2224 /* Hardcode /root as home directory for UID 0 */
2229 r
= get_home_dir(buf
);
2237 static int exec_child(
2239 ExecCommand
*command
,
2240 const ExecContext
*context
,
2241 const ExecParameters
*params
,
2242 ExecRuntime
*runtime
,
2243 DynamicCreds
*dcreds
,
2247 int *fds
, unsigned n_fds
,
2248 unsigned n_socket_fds
,
2252 char **error_message
) {
2254 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2255 _cleanup_free_
char *mac_selinux_context_net
= NULL
, *home_buffer
= NULL
;
2256 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2257 const char *username
= NULL
, *groupname
= NULL
;
2258 const char *home
= NULL
, *shell
= NULL
;
2259 dev_t journal_stream_dev
= 0;
2260 ino_t journal_stream_ino
= 0;
2261 bool needs_mount_namespace
;
2262 uid_t uid
= UID_INVALID
;
2263 gid_t gid
= GID_INVALID
;
2264 int i
, r
, ngids
= 0;
2270 assert(exit_status
);
2271 assert(error_message
);
2272 /* We don't always set error_message, hence it must be initialized */
2273 assert(*error_message
== NULL
);
2275 rename_process_from_path(command
->path
);
2277 /* We reset exactly these signals, since they are the
2278 * only ones we set to SIG_IGN in the main daemon. All
2279 * others we leave untouched because we set them to
2280 * SIG_DFL or a valid handler initially, both of which
2281 * will be demoted to SIG_DFL. */
2282 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2283 SIGNALS_IGNORE
, -1);
2285 if (context
->ignore_sigpipe
)
2286 (void) ignore_signals(SIGPIPE
, -1);
2288 r
= reset_signal_mask();
2290 *exit_status
= EXIT_SIGNAL_MASK
;
2291 *error_message
= strdup("Failed to reset signal mask");
2292 /* If strdup fails, here and below, we will just print the generic error message. */
2296 if (params
->idle_pipe
)
2297 do_idle_pipe_dance(params
->idle_pipe
);
2299 /* Close sockets very early to make sure we don't
2300 * block init reexecution because it cannot bind its
2305 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2307 *exit_status
= EXIT_FDS
;
2308 *error_message
= strdup("Failed to close remaining fds");
2312 if (!context
->same_pgrp
)
2314 *exit_status
= EXIT_SETSID
;
2318 exec_context_tty_reset(context
, params
);
2320 if (unit_shall_confirm_spawn(unit
)) {
2321 const char *vc
= params
->confirm_spawn
;
2322 _cleanup_free_
char *cmdline
= NULL
;
2324 cmdline
= exec_command_line(argv
);
2326 *exit_status
= EXIT_CONFIRM
;
2330 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2331 if (r
!= CONFIRM_EXECUTE
) {
2332 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2333 *exit_status
= EXIT_SUCCESS
;
2336 *exit_status
= EXIT_CONFIRM
;
2337 *error_message
= strdup("Execution cancelled");
2342 if (context
->dynamic_user
&& dcreds
) {
2344 /* Make sure we bypass our own NSS module for any NSS checks */
2345 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2346 *exit_status
= EXIT_USER
;
2347 *error_message
= strdup("Failed to update environment");
2351 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2353 *exit_status
= EXIT_USER
;
2354 *error_message
= strdup("Failed to update dynamic user credentials");
2358 if (!uid_is_valid(uid
)) {
2359 *exit_status
= EXIT_USER
;
2360 (void) asprintf(error_message
, "UID validation failed for \""UID_FMT
"\"", uid
);
2361 /* If asprintf fails, here and below, we will just print the generic error message. */
2365 if (!gid_is_valid(gid
)) {
2366 *exit_status
= EXIT_USER
;
2367 (void) asprintf(error_message
, "GID validation failed for \""GID_FMT
"\"", gid
);
2372 username
= dcreds
->user
->name
;
2375 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2377 *exit_status
= EXIT_USER
;
2378 *error_message
= strdup("Failed to determine user credentials");
2382 r
= get_fixed_group(context
, &groupname
, &gid
);
2384 *exit_status
= EXIT_GROUP
;
2385 *error_message
= strdup("Failed to determine group credentials");
2390 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2391 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2392 &supplementary_gids
, &ngids
);
2394 *exit_status
= EXIT_GROUP
;
2395 *error_message
= strdup("Failed to determine supplementary groups");
2399 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2401 *exit_status
= EXIT_USER
;
2402 *error_message
= strdup("Failed to send user credentials to PID1");
2406 user_lookup_fd
= safe_close(user_lookup_fd
);
2408 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2410 *exit_status
= EXIT_CHDIR
;
2411 *error_message
= strdup("Failed to determine $HOME for user");
2415 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2416 * must sure to drop O_NONBLOCK */
2418 (void) fd_nonblock(socket_fd
, false);
2420 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2422 *exit_status
= EXIT_STDIN
;
2423 *error_message
= strdup("Failed to set up stdin");
2427 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2429 *exit_status
= EXIT_STDOUT
;
2430 *error_message
= strdup("Failed to set up stdout");
2434 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2436 *exit_status
= EXIT_STDERR
;
2437 *error_message
= strdup("Failed to set up stderr");
2441 if (params
->cgroup_path
) {
2442 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2444 *exit_status
= EXIT_CGROUP
;
2445 (void) asprintf(error_message
, "Failed to attach to cgroup %s", params
->cgroup_path
);
2450 if (context
->oom_score_adjust_set
) {
2451 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2453 /* When we can't make this change due to EPERM, then
2454 * let's silently skip over it. User namespaces
2455 * prohibit write access to this file, and we
2456 * shouldn't trip up over that. */
2458 sprintf(t
, "%i", context
->oom_score_adjust
);
2459 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2460 if (r
== -EPERM
|| r
== -EACCES
) {
2462 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2465 *exit_status
= EXIT_OOM_ADJUST
;
2466 *error_message
= strdup("Failed to write /proc/self/oom_score_adj");
2471 if (context
->nice_set
)
2472 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2473 *exit_status
= EXIT_NICE
;
2477 if (context
->cpu_sched_set
) {
2478 struct sched_param param
= {
2479 .sched_priority
= context
->cpu_sched_priority
,
2482 r
= sched_setscheduler(0,
2483 context
->cpu_sched_policy
|
2484 (context
->cpu_sched_reset_on_fork
?
2485 SCHED_RESET_ON_FORK
: 0),
2488 *exit_status
= EXIT_SETSCHEDULER
;
2493 if (context
->cpuset
)
2494 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2495 *exit_status
= EXIT_CPUAFFINITY
;
2499 if (context
->ioprio_set
)
2500 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2501 *exit_status
= EXIT_IOPRIO
;
2505 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2506 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2507 *exit_status
= EXIT_TIMERSLACK
;
2511 if (context
->personality
!= PERSONALITY_INVALID
)
2512 if (personality(context
->personality
) < 0) {
2513 *exit_status
= EXIT_PERSONALITY
;
2517 if (context
->utmp_id
)
2518 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0),
2520 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2521 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2525 if (context
->user
) {
2526 r
= chown_terminal(STDIN_FILENO
, uid
);
2528 *exit_status
= EXIT_STDIN
;
2533 /* If delegation is enabled we'll pass ownership of the cgroup
2534 * (but only in systemd's own controller hierarchy!) to the
2535 * user of the new process. */
2536 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2537 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2539 *exit_status
= EXIT_CGROUP
;
2544 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2546 *exit_status
= EXIT_CGROUP
;
2551 if (!strv_isempty(context
->runtime_directory
) && params
->runtime_prefix
) {
2552 r
= setup_runtime_directory(context
, params
, uid
, gid
);
2554 *exit_status
= EXIT_RUNTIME_DIRECTORY
;
2559 r
= build_environment(
2571 *exit_status
= EXIT_MEMORY
;
2575 r
= build_pass_environment(context
, &pass_env
);
2577 *exit_status
= EXIT_MEMORY
;
2581 accum_env
= strv_env_merge(5,
2582 params
->environment
,
2585 context
->environment
,
2589 *exit_status
= EXIT_MEMORY
;
2592 accum_env
= strv_env_clean(accum_env
);
2594 (void) umask(context
->umask
);
2596 r
= setup_keyring(unit
, params
, uid
, gid
);
2598 *exit_status
= EXIT_KEYRING
;
2602 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2603 if (context
->pam_name
&& username
) {
2604 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2606 *exit_status
= EXIT_PAM
;
2612 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2613 r
= setup_netns(runtime
->netns_storage_socket
);
2615 *exit_status
= EXIT_NETWORK
;
2620 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2621 if (needs_mount_namespace
) {
2622 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
2624 *exit_status
= EXIT_NAMESPACE
;
2629 /* Apply just after mount namespace setup */
2630 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
2634 /* Drop groups as early as possbile */
2635 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2636 r
= enforce_groups(context
, gid
, supplementary_gids
, ngids
);
2638 *exit_status
= EXIT_GROUP
;
2644 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) &&
2645 mac_selinux_use() &&
2646 params
->selinux_context_net
&&
2648 !command
->privileged
) {
2650 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2652 *exit_status
= EXIT_SELINUX_CONTEXT
;
2658 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2659 r
= setup_private_users(uid
, gid
);
2661 *exit_status
= EXIT_USER
;
2666 /* We repeat the fd closing here, to make sure that
2667 * nothing is leaked from the PAM modules. Note that
2668 * we are more aggressive this time since socket_fd
2669 * and the netns fds we don't need anymore. The custom
2670 * endpoint fd was needed to upload the policy and can
2671 * now be closed as well. */
2672 r
= close_all_fds(fds
, n_fds
);
2674 r
= shift_fds(fds
, n_fds
);
2676 r
= flags_fds(fds
, n_fds
, n_socket_fds
, context
->non_blocking
);
2678 *exit_status
= EXIT_FDS
;
2682 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2684 int secure_bits
= context
->secure_bits
;
2686 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2688 if (!context
->rlimit
[i
])
2691 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2693 *exit_status
= EXIT_LIMITS
;
2698 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2699 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2700 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2701 *exit_status
= EXIT_LIMITS
;
2706 if (!cap_test_all(context
->capability_bounding_set
)) {
2707 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2709 *exit_status
= EXIT_CAPABILITIES
;
2710 *error_message
= strdup("Failed to drop capabilities");
2715 /* This is done before enforce_user, but ambient set
2716 * does not survive over setresuid() if keep_caps is not set. */
2717 if (context
->capability_ambient_set
!= 0) {
2718 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2720 *exit_status
= EXIT_CAPABILITIES
;
2721 *error_message
= strdup("Failed to apply ambient capabilities (before UID change)");
2726 if (context
->user
) {
2727 r
= enforce_user(context
, uid
);
2729 *exit_status
= EXIT_USER
;
2730 (void) asprintf(error_message
, "Failed to change UID to "UID_FMT
, uid
);
2733 if (context
->capability_ambient_set
!= 0) {
2735 /* Fix the ambient capabilities after user change. */
2736 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2738 *exit_status
= EXIT_CAPABILITIES
;
2739 *error_message
= strdup("Failed to apply ambient capabilities (after UID change)");
2743 /* If we were asked to change user and ambient capabilities
2744 * were requested, we had to add keep-caps to the securebits
2745 * so that we would maintain the inherited capability set
2746 * through the setresuid(). Make sure that the bit is added
2747 * also to the context secure_bits so that we don't try to
2748 * drop the bit away next. */
2750 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2754 /* Apply the MAC contexts late, but before seccomp syscall filtering, as those should really be last to
2755 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
2756 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
2757 * are restricted. */
2760 if (mac_selinux_use()) {
2761 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2764 r
= setexeccon(exec_context
);
2766 *exit_status
= EXIT_SELINUX_CONTEXT
;
2767 (void) asprintf(error_message
, "Failed to set SELinux context to %s", exec_context
);
2774 r
= setup_smack(context
, command
);
2776 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2777 *error_message
= strdup("Failed to set SMACK process label");
2781 #ifdef HAVE_APPARMOR
2782 if (context
->apparmor_profile
&& mac_apparmor_use()) {
2783 r
= aa_change_onexec(context
->apparmor_profile
);
2784 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2785 *exit_status
= EXIT_APPARMOR_PROFILE
;
2786 (void) asprintf(error_message
,
2787 "Failed to prepare AppArmor profile change to %s",
2788 context
->apparmor_profile
);
2794 /* PR_GET_SECUREBITS is not privileged, while
2795 * PR_SET_SECUREBITS is. So to suppress
2796 * potential EPERMs we'll try not to call
2797 * PR_SET_SECUREBITS unless necessary. */
2798 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2799 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2800 *exit_status
= EXIT_SECUREBITS
;
2801 *error_message
= strdup("Failed to set secure bits");
2805 if (context_has_no_new_privileges(context
))
2806 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2807 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2808 *error_message
= strdup("Failed to disable new privileges");
2813 r
= apply_address_families(unit
, context
);
2815 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2816 *error_message
= strdup("Failed to restrict address families");
2820 r
= apply_memory_deny_write_execute(unit
, context
);
2822 *exit_status
= EXIT_SECCOMP
;
2823 *error_message
= strdup("Failed to disable writing to executable memory");
2827 r
= apply_restrict_realtime(unit
, context
);
2829 *exit_status
= EXIT_SECCOMP
;
2830 *error_message
= strdup("Failed to apply realtime restrictions");
2834 r
= apply_restrict_namespaces(unit
, context
);
2836 *exit_status
= EXIT_SECCOMP
;
2837 *error_message
= strdup("Failed to apply namespace restrictions");
2841 r
= apply_protect_sysctl(unit
, context
);
2843 *exit_status
= EXIT_SECCOMP
;
2844 *error_message
= strdup("Failed to apply sysctl restrictions");
2848 r
= apply_protect_kernel_modules(unit
, context
);
2850 *exit_status
= EXIT_SECCOMP
;
2851 *error_message
= strdup("Failed to apply module loading restrictions");
2855 r
= apply_private_devices(unit
, context
);
2857 *exit_status
= EXIT_SECCOMP
;
2858 *error_message
= strdup("Failed to set up private devices");
2862 r
= apply_syscall_archs(unit
, context
);
2864 *exit_status
= EXIT_SECCOMP
;
2865 *error_message
= strdup("Failed to apply syscall architecture restrictions");
2869 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
2870 * by the filter as little as possible. */
2871 r
= apply_syscall_filter(unit
, context
);
2873 *exit_status
= EXIT_SECCOMP
;
2874 *error_message
= strdup("Failed to apply syscall filters");
2880 final_argv
= replace_env_argv(argv
, accum_env
);
2882 *exit_status
= EXIT_MEMORY
;
2883 *error_message
= strdup("Failed to prepare process arguments");
2887 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2888 _cleanup_free_
char *line
;
2890 line
= exec_command_line(final_argv
);
2893 log_struct(LOG_DEBUG
,
2894 "EXECUTABLE=%s", command
->path
,
2895 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2902 execve(command
->path
, final_argv
, accum_env
);
2903 *exit_status
= EXIT_EXEC
;
2907 int exec_spawn(Unit
*unit
,
2908 ExecCommand
*command
,
2909 const ExecContext
*context
,
2910 const ExecParameters
*params
,
2911 ExecRuntime
*runtime
,
2912 DynamicCreds
*dcreds
,
2915 _cleanup_strv_free_
char **files_env
= NULL
;
2917 unsigned n_fds
= 0, n_socket_fds
= 0;
2918 _cleanup_free_
char *line
= NULL
;
2920 int named_iofds
[3] = { -1, -1, -1 };
2929 assert(params
->fds
|| params
->n_fds
<= 0);
2931 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2932 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2933 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
2935 if (params
->n_fds
> 1) {
2936 log_unit_error(unit
, "Got more than one socket.");
2940 if (params
->n_fds
== 0) {
2941 log_unit_error(unit
, "Got no socket.");
2945 socket_fd
= params
->fds
[0];
2949 n_fds
= params
->n_fds
;
2950 n_socket_fds
= params
->n_socket_fds
;
2953 r
= exec_context_named_iofds(unit
, context
, params
, named_iofds
);
2955 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
2957 r
= exec_context_load_environment(unit
, context
, &files_env
);
2959 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
2961 argv
= params
->argv
?: command
->argv
;
2962 line
= exec_command_line(argv
);
2966 log_struct(LOG_DEBUG
,
2967 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
2968 "EXECUTABLE=%s", command
->path
,
2973 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
2977 _cleanup_free_
char *error_message
= NULL
;
2979 r
= exec_child(unit
,
2991 unit
->manager
->user_lookup_fds
[1],
2997 log_struct_errno(LOG_ERR
, r
,
2998 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3000 LOG_UNIT_MESSAGE(unit
, "%s: %m",
3002 "EXECUTABLE=%s", command
->path
,
3004 else if (r
== -ENOENT
&& command
->ignore
)
3005 log_struct_errno(LOG_INFO
, r
,
3006 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3008 LOG_UNIT_MESSAGE(unit
, "Skipped spawning %s: %m",
3010 "EXECUTABLE=%s", command
->path
,
3013 log_struct_errno(LOG_ERR
, r
,
3014 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3016 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3017 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3019 "EXECUTABLE=%s", command
->path
,
3026 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3028 /* We add the new process to the cgroup both in the child (so
3029 * that we can be sure that no user code is ever executed
3030 * outside of the cgroup) and in the parent (so that we can be
3031 * sure that when we kill the cgroup the process will be
3033 if (params
->cgroup_path
)
3034 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3036 exec_status_start(&command
->exec_status
, pid
);
3042 void exec_context_init(ExecContext
*c
) {
3046 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3047 c
->cpu_sched_policy
= SCHED_OTHER
;
3048 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3049 c
->syslog_level_prefix
= true;
3050 c
->ignore_sigpipe
= true;
3051 c
->timer_slack_nsec
= NSEC_INFINITY
;
3052 c
->personality
= PERSONALITY_INVALID
;
3053 c
->runtime_directory_mode
= 0755;
3054 c
->capability_bounding_set
= CAP_ALL
;
3055 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
3058 void exec_context_done(ExecContext
*c
) {
3063 c
->environment
= strv_free(c
->environment
);
3064 c
->environment_files
= strv_free(c
->environment_files
);
3065 c
->pass_environment
= strv_free(c
->pass_environment
);
3067 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
3068 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
3070 for (l
= 0; l
< 3; l
++)
3071 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3073 c
->working_directory
= mfree(c
->working_directory
);
3074 c
->root_directory
= mfree(c
->root_directory
);
3075 c
->root_image
= mfree(c
->root_image
);
3076 c
->tty_path
= mfree(c
->tty_path
);
3077 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3078 c
->user
= mfree(c
->user
);
3079 c
->group
= mfree(c
->group
);
3081 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3083 c
->pam_name
= mfree(c
->pam_name
);
3085 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3086 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3087 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3089 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3092 CPU_FREE(c
->cpuset
);
3094 c
->utmp_id
= mfree(c
->utmp_id
);
3095 c
->selinux_context
= mfree(c
->selinux_context
);
3096 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3098 c
->syscall_filter
= set_free(c
->syscall_filter
);
3099 c
->syscall_archs
= set_free(c
->syscall_archs
);
3100 c
->address_families
= set_free(c
->address_families
);
3102 c
->runtime_directory
= strv_free(c
->runtime_directory
);
3105 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
3110 if (!runtime_prefix
)
3113 STRV_FOREACH(i
, c
->runtime_directory
) {
3114 _cleanup_free_
char *p
;
3116 p
= strjoin(runtime_prefix
, "/", *i
);
3120 /* We execute this synchronously, since we need to be
3121 * sure this is gone when we start the service
3123 (void) rm_rf(p
, REMOVE_ROOT
);
3129 void exec_command_done(ExecCommand
*c
) {
3132 c
->path
= mfree(c
->path
);
3134 c
->argv
= strv_free(c
->argv
);
3137 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3140 for (i
= 0; i
< n
; i
++)
3141 exec_command_done(c
+i
);
3144 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3148 LIST_REMOVE(command
, c
, i
);
3149 exec_command_done(i
);
3156 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3159 for (i
= 0; i
< n
; i
++)
3160 c
[i
] = exec_command_free_list(c
[i
]);
3163 typedef struct InvalidEnvInfo
{
3168 static void invalid_env(const char *p
, void *userdata
) {
3169 InvalidEnvInfo
*info
= userdata
;
3171 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3174 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3179 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3181 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3183 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3185 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3187 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3189 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3195 int exec_context_named_iofds(Unit
*unit
, const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3196 unsigned i
, targets
;
3197 const char* stdio_fdname
[3];
3202 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3203 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3204 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3206 for (i
= 0; i
< 3; i
++)
3207 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3209 for (i
= 0; i
< p
->n_fds
&& targets
> 0; i
++)
3210 if (named_iofds
[STDIN_FILENO
] < 0 &&
3211 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3212 stdio_fdname
[STDIN_FILENO
] &&
3213 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3215 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3218 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3219 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3220 stdio_fdname
[STDOUT_FILENO
] &&
3221 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3223 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3226 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3227 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3228 stdio_fdname
[STDERR_FILENO
] &&
3229 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3231 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3235 return targets
== 0 ? 0 : -ENOENT
;
3238 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
3239 char **i
, **r
= NULL
;
3244 STRV_FOREACH(i
, c
->environment_files
) {
3247 bool ignore
= false;
3249 _cleanup_globfree_ glob_t pglob
= {};
3258 if (!path_is_absolute(fn
)) {
3266 /* Filename supports globbing, take all matching files */
3267 k
= safe_glob(fn
, 0, &pglob
);
3276 /* When we don't match anything, -ENOENT should be returned */
3277 assert(pglob
.gl_pathc
> 0);
3279 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
3280 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3288 /* Log invalid environment variables with filename */
3290 InvalidEnvInfo info
= {
3292 .path
= pglob
.gl_pathv
[n
]
3295 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3303 m
= strv_env_merge(2, r
, p
);
3319 static bool tty_may_match_dev_console(const char *tty
) {
3320 _cleanup_free_
char *active
= NULL
;
3326 if (startswith(tty
, "/dev/"))
3329 /* trivial identity? */
3330 if (streq(tty
, "console"))
3333 console
= resolve_dev_console(&active
);
3334 /* if we could not resolve, assume it may */
3338 /* "tty0" means the active VC, so it may be the same sometimes */
3339 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3342 bool exec_context_may_touch_console(ExecContext
*ec
) {
3344 return (ec
->tty_reset
||
3346 ec
->tty_vt_disallocate
||
3347 is_terminal_input(ec
->std_input
) ||
3348 is_terminal_output(ec
->std_output
) ||
3349 is_terminal_output(ec
->std_error
)) &&
3350 tty_may_match_dev_console(exec_context_tty_path(ec
));
3353 static void strv_fprintf(FILE *f
, char **l
) {
3359 fprintf(f
, " %s", *g
);
3362 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3370 prefix
= strempty(prefix
);
3374 "%sWorkingDirectory: %s\n"
3375 "%sRootDirectory: %s\n"
3376 "%sNonBlocking: %s\n"
3377 "%sPrivateTmp: %s\n"
3378 "%sPrivateDevices: %s\n"
3379 "%sProtectKernelTunables: %s\n"
3380 "%sProtectKernelModules: %s\n"
3381 "%sProtectControlGroups: %s\n"
3382 "%sPrivateNetwork: %s\n"
3383 "%sPrivateUsers: %s\n"
3384 "%sProtectHome: %s\n"
3385 "%sProtectSystem: %s\n"
3386 "%sMountAPIVFS: %s\n"
3387 "%sIgnoreSIGPIPE: %s\n"
3388 "%sMemoryDenyWriteExecute: %s\n"
3389 "%sRestrictRealtime: %s\n",
3391 prefix
, c
->working_directory
? c
->working_directory
: "/",
3392 prefix
, c
->root_directory
? c
->root_directory
: "/",
3393 prefix
, yes_no(c
->non_blocking
),
3394 prefix
, yes_no(c
->private_tmp
),
3395 prefix
, yes_no(c
->private_devices
),
3396 prefix
, yes_no(c
->protect_kernel_tunables
),
3397 prefix
, yes_no(c
->protect_kernel_modules
),
3398 prefix
, yes_no(c
->protect_control_groups
),
3399 prefix
, yes_no(c
->private_network
),
3400 prefix
, yes_no(c
->private_users
),
3401 prefix
, protect_home_to_string(c
->protect_home
),
3402 prefix
, protect_system_to_string(c
->protect_system
),
3403 prefix
, yes_no(c
->mount_apivfs
),
3404 prefix
, yes_no(c
->ignore_sigpipe
),
3405 prefix
, yes_no(c
->memory_deny_write_execute
),
3406 prefix
, yes_no(c
->restrict_realtime
));
3409 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3411 STRV_FOREACH(e
, c
->environment
)
3412 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3414 STRV_FOREACH(e
, c
->environment_files
)
3415 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3417 STRV_FOREACH(e
, c
->pass_environment
)
3418 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3420 fprintf(f
, "%sRuntimeDirectoryMode: %04o\n", prefix
, c
->runtime_directory_mode
);
3422 STRV_FOREACH(d
, c
->runtime_directory
)
3423 fprintf(f
, "%sRuntimeDirectory: %s\n", prefix
, *d
);
3430 if (c
->oom_score_adjust_set
)
3432 "%sOOMScoreAdjust: %i\n",
3433 prefix
, c
->oom_score_adjust
);
3435 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3437 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3438 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3439 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3440 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3443 if (c
->ioprio_set
) {
3444 _cleanup_free_
char *class_str
= NULL
;
3446 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3448 "%sIOSchedulingClass: %s\n"
3449 "%sIOPriority: %i\n",
3450 prefix
, strna(class_str
),
3451 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
3454 if (c
->cpu_sched_set
) {
3455 _cleanup_free_
char *policy_str
= NULL
;
3457 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3459 "%sCPUSchedulingPolicy: %s\n"
3460 "%sCPUSchedulingPriority: %i\n"
3461 "%sCPUSchedulingResetOnFork: %s\n",
3462 prefix
, strna(policy_str
),
3463 prefix
, c
->cpu_sched_priority
,
3464 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3468 fprintf(f
, "%sCPUAffinity:", prefix
);
3469 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3470 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3471 fprintf(f
, " %u", i
);
3475 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3476 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3479 "%sStandardInput: %s\n"
3480 "%sStandardOutput: %s\n"
3481 "%sStandardError: %s\n",
3482 prefix
, exec_input_to_string(c
->std_input
),
3483 prefix
, exec_output_to_string(c
->std_output
),
3484 prefix
, exec_output_to_string(c
->std_error
));
3490 "%sTTYVHangup: %s\n"
3491 "%sTTYVTDisallocate: %s\n",
3492 prefix
, c
->tty_path
,
3493 prefix
, yes_no(c
->tty_reset
),
3494 prefix
, yes_no(c
->tty_vhangup
),
3495 prefix
, yes_no(c
->tty_vt_disallocate
));
3497 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3498 c
->std_output
== EXEC_OUTPUT_KMSG
||
3499 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3500 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3501 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3502 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3503 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3504 c
->std_error
== EXEC_OUTPUT_KMSG
||
3505 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3506 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3507 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3508 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3510 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3512 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3513 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3516 "%sSyslogFacility: %s\n"
3517 "%sSyslogLevel: %s\n",
3518 prefix
, strna(fac_str
),
3519 prefix
, strna(lvl_str
));
3523 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
3525 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
3526 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
3527 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
3528 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
3529 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
3530 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
3532 if (c
->capability_bounding_set
!= CAP_ALL
) {
3534 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
3536 for (l
= 0; l
<= cap_last_cap(); l
++)
3537 if (c
->capability_bounding_set
& (UINT64_C(1) << l
))
3538 fprintf(f
, " %s", strna(capability_to_name(l
)));
3543 if (c
->capability_ambient_set
!= 0) {
3545 fprintf(f
, "%sAmbientCapabilities:", prefix
);
3547 for (l
= 0; l
<= cap_last_cap(); l
++)
3548 if (c
->capability_ambient_set
& (UINT64_C(1) << l
))
3549 fprintf(f
, " %s", strna(capability_to_name(l
)));
3555 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3557 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3559 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3561 if (strv_length(c
->supplementary_groups
) > 0) {
3562 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3563 strv_fprintf(f
, c
->supplementary_groups
);
3568 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3570 if (strv_length(c
->read_write_paths
) > 0) {
3571 fprintf(f
, "%sReadWritePaths:", prefix
);
3572 strv_fprintf(f
, c
->read_write_paths
);
3576 if (strv_length(c
->read_only_paths
) > 0) {
3577 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3578 strv_fprintf(f
, c
->read_only_paths
);
3582 if (strv_length(c
->inaccessible_paths
) > 0) {
3583 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3584 strv_fprintf(f
, c
->inaccessible_paths
);
3588 if (c
->n_bind_mounts
> 0)
3589 for (i
= 0; i
< c
->n_bind_mounts
; i
++) {
3590 fprintf(f
, "%s%s: %s:%s:%s\n", prefix
,
3591 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
3592 c
->bind_mounts
[i
].source
,
3593 c
->bind_mounts
[i
].destination
,
3594 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
3599 "%sUtmpIdentifier: %s\n",
3600 prefix
, c
->utmp_id
);
3602 if (c
->selinux_context
)
3604 "%sSELinuxContext: %s%s\n",
3605 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3607 if (c
->personality
!= PERSONALITY_INVALID
)
3609 "%sPersonality: %s\n",
3610 prefix
, strna(personality_to_string(c
->personality
)));
3612 if (c
->syscall_filter
) {
3620 "%sSystemCallFilter: ",
3623 if (!c
->syscall_whitelist
)
3627 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3628 _cleanup_free_
char *name
= NULL
;
3635 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3636 fputs(strna(name
), f
);
3643 if (c
->syscall_archs
) {
3650 "%sSystemCallArchitectures:",
3654 SET_FOREACH(id
, c
->syscall_archs
, j
)
3655 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3660 if (exec_context_restrict_namespaces_set(c
)) {
3661 _cleanup_free_
char *s
= NULL
;
3663 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
3665 fprintf(f
, "%sRestrictNamespaces: %s\n",
3669 if (c
->syscall_errno
> 0)
3671 "%sSystemCallErrorNumber: %s\n",
3672 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3674 if (c
->apparmor_profile
)
3676 "%sAppArmorProfile: %s%s\n",
3677 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3680 bool exec_context_maintains_privileges(ExecContext
*c
) {
3683 /* Returns true if the process forked off would run under
3684 * an unchanged UID or as root. */
3689 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3695 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3700 dual_timestamp_get(&s
->start_timestamp
);
3703 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3706 if (s
->pid
&& s
->pid
!= pid
)
3710 dual_timestamp_get(&s
->exit_timestamp
);
3716 if (context
->utmp_id
)
3717 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3719 exec_context_tty_reset(context
, NULL
);
3723 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3724 char buf
[FORMAT_TIMESTAMP_MAX
];
3732 prefix
= strempty(prefix
);
3735 "%sPID: "PID_FMT
"\n",
3738 if (dual_timestamp_is_set(&s
->start_timestamp
))
3740 "%sStart Timestamp: %s\n",
3741 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3743 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3745 "%sExit Timestamp: %s\n"
3747 "%sExit Status: %i\n",
3748 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3749 prefix
, sigchld_code_to_string(s
->code
),
3753 char *exec_command_line(char **argv
) {
3761 STRV_FOREACH(a
, argv
)
3769 STRV_FOREACH(a
, argv
) {
3776 if (strpbrk(*a
, WHITESPACE
)) {
3787 /* FIXME: this doesn't really handle arguments that have
3788 * spaces and ticks in them */
3793 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3794 _cleanup_free_
char *cmd
= NULL
;
3795 const char *prefix2
;
3800 prefix
= strempty(prefix
);
3801 prefix2
= strjoina(prefix
, "\t");
3803 cmd
= exec_command_line(c
->argv
);
3805 "%sCommand Line: %s\n",
3806 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3808 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3811 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3814 prefix
= strempty(prefix
);
3816 LIST_FOREACH(command
, c
, c
)
3817 exec_command_dump(c
, f
, prefix
);
3820 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3827 /* It's kind of important, that we keep the order here */
3828 LIST_FIND_TAIL(command
, *l
, end
);
3829 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3834 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3842 l
= strv_new_ap(path
, ap
);
3863 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3864 _cleanup_strv_free_
char **l
= NULL
;
3872 l
= strv_new_ap(path
, ap
);
3878 r
= strv_extend_strv(&c
->argv
, l
, false);
3886 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3891 *rt
= new0(ExecRuntime
, 1);
3896 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3901 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
3911 if (!c
->private_network
&& !c
->private_tmp
)
3914 r
= exec_runtime_allocate(rt
);
3918 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
3919 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
3923 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
3924 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
3932 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
3934 assert(r
->n_ref
> 0);
3940 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
3945 assert(r
->n_ref
> 0);
3952 free(r
->var_tmp_dir
);
3953 safe_close_pair(r
->netns_storage_socket
);
3957 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
3966 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
3968 if (rt
->var_tmp_dir
)
3969 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
3971 if (rt
->netns_storage_socket
[0] >= 0) {
3974 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
3978 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
3981 if (rt
->netns_storage_socket
[1] >= 0) {
3984 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
3988 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
3994 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
4001 if (streq(key
, "tmp-dir")) {
4004 r
= exec_runtime_allocate(rt
);
4008 copy
= strdup(value
);
4012 free((*rt
)->tmp_dir
);
4013 (*rt
)->tmp_dir
= copy
;
4015 } else if (streq(key
, "var-tmp-dir")) {
4018 r
= exec_runtime_allocate(rt
);
4022 copy
= strdup(value
);
4026 free((*rt
)->var_tmp_dir
);
4027 (*rt
)->var_tmp_dir
= copy
;
4029 } else if (streq(key
, "netns-socket-0")) {
4032 r
= exec_runtime_allocate(rt
);
4036 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4037 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4039 safe_close((*rt
)->netns_storage_socket
[0]);
4040 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4042 } else if (streq(key
, "netns-socket-1")) {
4045 r
= exec_runtime_allocate(rt
);
4049 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4050 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4052 safe_close((*rt
)->netns_storage_socket
[1]);
4053 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
4061 static void *remove_tmpdir_thread(void *p
) {
4062 _cleanup_free_
char *path
= p
;
4064 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4068 void exec_runtime_destroy(ExecRuntime
*rt
) {
4074 /* If there are multiple users of this, let's leave the stuff around */
4079 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4081 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4083 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4090 if (rt
->var_tmp_dir
) {
4091 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4093 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4095 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4096 free(rt
->var_tmp_dir
);
4099 rt
->var_tmp_dir
= NULL
;
4102 safe_close_pair(rt
->netns_storage_socket
);
4105 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4106 [EXEC_INPUT_NULL
] = "null",
4107 [EXEC_INPUT_TTY
] = "tty",
4108 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4109 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4110 [EXEC_INPUT_SOCKET
] = "socket",
4111 [EXEC_INPUT_NAMED_FD
] = "fd",
4114 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4116 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4117 [EXEC_OUTPUT_INHERIT
] = "inherit",
4118 [EXEC_OUTPUT_NULL
] = "null",
4119 [EXEC_OUTPUT_TTY
] = "tty",
4120 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4121 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4122 [EXEC_OUTPUT_KMSG
] = "kmsg",
4123 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4124 [EXEC_OUTPUT_JOURNAL
] = "journal",
4125 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4126 [EXEC_OUTPUT_SOCKET
] = "socket",
4127 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4130 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4132 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4133 [EXEC_UTMP_INIT
] = "init",
4134 [EXEC_UTMP_LOGIN
] = "login",
4135 [EXEC_UTMP_USER
] = "user",
4138 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);