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
, bool nonblock
) {
169 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
171 for (i
= 0; i
< n_fds
; i
++) {
173 r
= fd_nonblock(fds
[i
], nonblock
);
177 /* We unconditionally drop FD_CLOEXEC from the fds,
178 * since after all we want to pass these fds to our
181 r
= fd_cloexec(fds
[i
], false);
189 static const char *exec_context_tty_path(const ExecContext
*context
) {
192 if (context
->stdio_as_fds
)
195 if (context
->tty_path
)
196 return context
->tty_path
;
198 return "/dev/console";
201 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
206 path
= exec_context_tty_path(context
);
208 if (context
->tty_vhangup
) {
209 if (p
&& p
->stdin_fd
>= 0)
210 (void) terminal_vhangup_fd(p
->stdin_fd
);
212 (void) terminal_vhangup(path
);
215 if (context
->tty_reset
) {
216 if (p
&& p
->stdin_fd
>= 0)
217 (void) reset_terminal_fd(p
->stdin_fd
, true);
219 (void) reset_terminal(path
);
222 if (context
->tty_vt_disallocate
&& path
)
223 (void) vt_disallocate(path
);
226 static bool is_terminal_input(ExecInput i
) {
229 EXEC_INPUT_TTY_FORCE
,
230 EXEC_INPUT_TTY_FAIL
);
233 static bool is_terminal_output(ExecOutput o
) {
236 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
237 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
238 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
241 static bool exec_context_needs_term(const ExecContext
*c
) {
244 /* Return true if the execution context suggests we should set $TERM to something useful. */
246 if (is_terminal_input(c
->std_input
))
249 if (is_terminal_output(c
->std_output
))
252 if (is_terminal_output(c
->std_error
))
255 return !!c
->tty_path
;
258 static int open_null_as(int flags
, int nfd
) {
263 fd
= open("/dev/null", flags
|O_NOCTTY
);
268 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
276 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
277 union sockaddr_union sa
= {
278 .un
.sun_family
= AF_UNIX
,
279 .un
.sun_path
= "/run/systemd/journal/stdout",
281 uid_t olduid
= UID_INVALID
;
282 gid_t oldgid
= GID_INVALID
;
285 if (gid
!= GID_INVALID
) {
293 if (uid
!= UID_INVALID
) {
303 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
307 /* If we fail to restore the uid or gid, things will likely
308 fail later on. This should only happen if an LSM interferes. */
310 if (uid
!= UID_INVALID
)
311 (void) seteuid(olduid
);
314 if (gid
!= GID_INVALID
)
315 (void) setegid(oldgid
);
320 static int connect_logger_as(
322 const ExecContext
*context
,
332 assert(output
< _EXEC_OUTPUT_MAX
);
336 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
340 r
= connect_journal_socket(fd
, uid
, gid
);
344 if (shutdown(fd
, SHUT_RD
) < 0) {
349 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
359 context
->syslog_identifier
? context
->syslog_identifier
: ident
,
361 context
->syslog_priority
,
362 !!context
->syslog_level_prefix
,
363 output
== EXEC_OUTPUT_SYSLOG
|| output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
364 output
== EXEC_OUTPUT_KMSG
|| output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
,
365 is_terminal_output(output
));
370 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
375 static int open_terminal_as(const char *path
, mode_t mode
, int nfd
) {
381 fd
= open_terminal(path
, mode
| O_NOCTTY
);
386 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
394 static int fixup_input(ExecInput std_input
, int socket_fd
, bool apply_tty_stdin
) {
396 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
397 return EXEC_INPUT_NULL
;
399 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
400 return EXEC_INPUT_NULL
;
405 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
407 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
408 return EXEC_OUTPUT_INHERIT
;
413 static int setup_input(
414 const ExecContext
*context
,
415 const ExecParameters
*params
,
417 int named_iofds
[3]) {
424 if (params
->stdin_fd
>= 0) {
425 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
428 /* Try to make this the controlling tty, if it is a tty, and reset it */
429 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
430 (void) reset_terminal_fd(STDIN_FILENO
, true);
435 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
439 case EXEC_INPUT_NULL
:
440 return open_null_as(O_RDONLY
, STDIN_FILENO
);
443 case EXEC_INPUT_TTY_FORCE
:
444 case EXEC_INPUT_TTY_FAIL
: {
447 fd
= acquire_terminal(exec_context_tty_path(context
),
448 i
== EXEC_INPUT_TTY_FAIL
,
449 i
== EXEC_INPUT_TTY_FORCE
,
455 if (fd
!= STDIN_FILENO
) {
456 r
= dup2(fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
464 case EXEC_INPUT_SOCKET
:
465 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
467 case EXEC_INPUT_NAMED_FD
:
468 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
469 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
472 assert_not_reached("Unknown input type");
476 static int setup_output(
478 const ExecContext
*context
,
479 const ExecParameters
*params
,
486 dev_t
*journal_stream_dev
,
487 ino_t
*journal_stream_ino
) {
497 assert(journal_stream_dev
);
498 assert(journal_stream_ino
);
500 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
502 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
505 return STDOUT_FILENO
;
508 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
509 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
512 return STDERR_FILENO
;
515 i
= fixup_input(context
->std_input
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
516 o
= fixup_output(context
->std_output
, socket_fd
);
518 if (fileno
== STDERR_FILENO
) {
520 e
= fixup_output(context
->std_error
, socket_fd
);
522 /* This expects the input and output are already set up */
524 /* Don't change the stderr file descriptor if we inherit all
525 * the way and are not on a tty */
526 if (e
== EXEC_OUTPUT_INHERIT
&&
527 o
== EXEC_OUTPUT_INHERIT
&&
528 i
== EXEC_INPUT_NULL
&&
529 !is_terminal_input(context
->std_input
) &&
533 /* Duplicate from stdout if possible */
534 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
535 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
539 } else if (o
== EXEC_OUTPUT_INHERIT
) {
540 /* If input got downgraded, inherit the original value */
541 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
542 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
544 /* If the input is connected to anything that's not a /dev/null, inherit that... */
545 if (i
!= EXEC_INPUT_NULL
)
546 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
548 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
552 /* We need to open /dev/null here anew, to get the right access mode. */
553 return open_null_as(O_WRONLY
, fileno
);
558 case EXEC_OUTPUT_NULL
:
559 return open_null_as(O_WRONLY
, fileno
);
561 case EXEC_OUTPUT_TTY
:
562 if (is_terminal_input(i
))
563 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
565 /* We don't reset the terminal if this is just about output */
566 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
568 case EXEC_OUTPUT_SYSLOG
:
569 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
570 case EXEC_OUTPUT_KMSG
:
571 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
572 case EXEC_OUTPUT_JOURNAL
:
573 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
574 r
= connect_logger_as(unit
, context
, o
, ident
, fileno
, uid
, gid
);
576 log_unit_error_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
577 r
= open_null_as(O_WRONLY
, fileno
);
581 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
582 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
583 * services to detect whether they are connected to the journal or not. */
585 if (fstat(fileno
, &st
) >= 0) {
586 *journal_stream_dev
= st
.st_dev
;
587 *journal_stream_ino
= st
.st_ino
;
592 case EXEC_OUTPUT_SOCKET
:
593 assert(socket_fd
>= 0);
594 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
596 case EXEC_OUTPUT_NAMED_FD
:
597 (void) fd_nonblock(named_iofds
[fileno
], false);
598 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
601 assert_not_reached("Unknown error type");
605 static int chown_terminal(int fd
, uid_t uid
) {
610 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
614 /* This might fail. What matters are the results. */
615 (void) fchown(fd
, uid
, -1);
616 (void) fchmod(fd
, TTY_MODE
);
618 if (fstat(fd
, &st
) < 0)
621 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
627 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
628 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
631 assert(_saved_stdin
);
632 assert(_saved_stdout
);
634 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
638 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
639 if (saved_stdout
< 0)
642 fd
= acquire_terminal(vc
, false, false, false, DEFAULT_CONFIRM_USEC
);
646 r
= chown_terminal(fd
, getuid());
650 r
= reset_terminal_fd(fd
, true);
654 if (dup2(fd
, STDIN_FILENO
) < 0)
657 if (dup2(fd
, STDOUT_FILENO
) < 0)
664 *_saved_stdin
= saved_stdin
;
665 *_saved_stdout
= saved_stdout
;
667 saved_stdin
= saved_stdout
= -1;
672 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
675 if (err
== -ETIMEDOUT
)
676 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
679 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
683 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
684 _cleanup_close_
int fd
= -1;
688 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
692 write_confirm_error_fd(err
, fd
, u
);
695 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
699 assert(saved_stdout
);
703 if (*saved_stdin
>= 0)
704 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
707 if (*saved_stdout
>= 0)
708 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
711 *saved_stdin
= safe_close(*saved_stdin
);
712 *saved_stdout
= safe_close(*saved_stdout
);
718 CONFIRM_PRETEND_FAILURE
= -1,
719 CONFIRM_PRETEND_SUCCESS
= 0,
723 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
724 int saved_stdout
= -1, saved_stdin
= -1, r
;
725 _cleanup_free_
char *e
= NULL
;
728 /* For any internal errors, assume a positive response. */
729 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
731 write_confirm_error(r
, vc
, u
);
732 return CONFIRM_EXECUTE
;
735 /* confirm_spawn might have been disabled while we were sleeping. */
736 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
741 e
= ellipsize(cmdline
, 60, 100);
749 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
751 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
758 printf("Resuming normal execution.\n");
759 manager_disable_confirm_spawn();
763 unit_dump(u
, stdout
, " ");
764 continue; /* ask again */
766 printf("Failing execution.\n");
767 r
= CONFIRM_PRETEND_FAILURE
;
770 printf(" c - continue, proceed without asking anymore\n"
771 " D - dump, show the state of the unit\n"
772 " f - fail, don't execute the command and pretend it failed\n"
774 " i - info, show a short summary of the unit\n"
775 " j - jobs, show jobs that are in progress\n"
776 " s - skip, don't execute the command and pretend it succeeded\n"
777 " y - yes, execute the command\n");
778 continue; /* ask again */
780 printf(" Description: %s\n"
783 u
->id
, u
->description
, cmdline
);
784 continue; /* ask again */
786 manager_dump_jobs(u
->manager
, stdout
, " ");
787 continue; /* ask again */
789 /* 'n' was removed in favor of 'f'. */
790 printf("Didn't understand 'n', did you mean 'f'?\n");
791 continue; /* ask again */
793 printf("Skipping execution.\n");
794 r
= CONFIRM_PRETEND_SUCCESS
;
800 assert_not_reached("Unhandled choice");
806 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
810 static int get_fixed_user(const ExecContext
*c
, const char **user
,
811 uid_t
*uid
, gid_t
*gid
,
812 const char **home
, const char **shell
) {
818 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
819 * (i.e. are "/" or "/bin/nologin"). */
821 name
= c
->user
?: "root";
822 r
= get_user_creds_clean(&name
, uid
, gid
, home
, shell
);
830 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
840 r
= get_group_creds(&name
, gid
);
848 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
849 const char *group
, gid_t gid
,
850 gid_t
**supplementary_gids
, int *ngids
) {
854 bool keep_groups
= false;
855 gid_t
*groups
= NULL
;
856 _cleanup_free_ gid_t
*l_gids
= NULL
;
861 * If user is given, then lookup GID and supplementary groups list.
862 * We avoid NSS lookups for gid=0. Also we have to initialize groups
863 * here and as early as possible so we keep the list of supplementary
864 * groups of the caller.
866 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
867 /* First step, initialize groups from /etc/groups */
868 if (initgroups(user
, gid
) < 0)
874 if (!c
->supplementary_groups
)
878 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
879 * be positive, otherwise fail.
882 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
883 if (ngroups_max
<= 0) {
887 return -EOPNOTSUPP
; /* For all other values */
890 l_gids
= new(gid_t
, ngroups_max
);
896 * Lookup the list of groups that the user belongs to, we
897 * avoid NSS lookups here too for gid=0.
900 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
905 STRV_FOREACH(i
, c
->supplementary_groups
) {
908 if (k
>= ngroups_max
)
912 r
= get_group_creds(&g
, l_gids
+k
);
920 * Sets ngids to zero to drop all supplementary groups, happens
921 * when we are under root and SupplementaryGroups= is empty.
928 /* Otherwise get the final list of supplementary groups */
929 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
933 *supplementary_gids
= groups
;
941 static int enforce_groups(const ExecContext
*context
, gid_t gid
,
942 gid_t
*supplementary_gids
, int ngids
) {
947 /* Handle SupplementaryGroups= even if it is empty */
948 if (context
->supplementary_groups
) {
949 r
= maybe_setgroups(ngids
, supplementary_gids
);
954 if (gid_is_valid(gid
)) {
955 /* Then set our gids */
956 if (setresgid(gid
, gid
, gid
) < 0)
963 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
966 if (!uid_is_valid(uid
))
969 /* Sets (but doesn't look up) the uid and make sure we keep the
970 * capabilities while doing so. */
972 if (context
->capability_ambient_set
!= 0) {
974 /* First step: If we need to keep capabilities but
975 * drop privileges we need to make sure we keep our
976 * caps, while we drop privileges. */
978 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
980 if (prctl(PR_GET_SECUREBITS
) != sb
)
981 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
986 /* Second step: actually set the uids */
987 if (setresuid(uid
, uid
, uid
) < 0)
990 /* At this point we should have all necessary capabilities but
991 are otherwise a normal user. However, the caps might got
992 corrupted due to the setresuid() so we need clean them up
993 later. This is done outside of this call. */
1000 static int null_conv(
1002 const struct pam_message
**msg
,
1003 struct pam_response
**resp
,
1004 void *appdata_ptr
) {
1006 /* We don't support conversations */
1008 return PAM_CONV_ERR
;
1013 static int setup_pam(
1020 int fds
[], unsigned n_fds
) {
1024 static const struct pam_conv conv
= {
1029 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1030 pam_handle_t
*handle
= NULL
;
1032 int pam_code
= PAM_SUCCESS
, r
;
1033 char **nv
, **e
= NULL
;
1034 bool close_session
= false;
1035 pid_t pam_pid
= 0, parent_pid
;
1042 /* We set up PAM in the parent process, then fork. The child
1043 * will then stay around until killed via PR_GET_PDEATHSIG or
1044 * systemd via the cgroup logic. It will then remove the PAM
1045 * session again. The parent process will exec() the actual
1046 * daemon. We do things this way to ensure that the main PID
1047 * of the daemon is the one we initially fork()ed. */
1049 r
= barrier_create(&barrier
);
1053 if (log_get_max_level() < LOG_DEBUG
)
1054 flags
|= PAM_SILENT
;
1056 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1057 if (pam_code
!= PAM_SUCCESS
) {
1063 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1064 if (pam_code
!= PAM_SUCCESS
)
1068 STRV_FOREACH(nv
, *env
) {
1069 pam_code
= pam_putenv(handle
, *nv
);
1070 if (pam_code
!= PAM_SUCCESS
)
1074 pam_code
= pam_acct_mgmt(handle
, flags
);
1075 if (pam_code
!= PAM_SUCCESS
)
1078 pam_code
= pam_open_session(handle
, flags
);
1079 if (pam_code
!= PAM_SUCCESS
)
1082 close_session
= true;
1084 e
= pam_getenvlist(handle
);
1086 pam_code
= PAM_BUF_ERR
;
1090 /* Block SIGTERM, so that we know that it won't get lost in
1093 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1095 parent_pid
= getpid();
1104 int sig
, ret
= EXIT_PAM
;
1106 /* The child's job is to reset the PAM session on
1108 barrier_set_role(&barrier
, BARRIER_CHILD
);
1110 /* This string must fit in 10 chars (i.e. the length
1111 * of "/sbin/init"), to look pretty in /bin/ps */
1112 rename_process("(sd-pam)");
1114 /* Make sure we don't keep open the passed fds in this
1115 child. We assume that otherwise only those fds are
1116 open here that have been opened by PAM. */
1117 close_many(fds
, n_fds
);
1119 /* Drop privileges - we don't need any to pam_close_session
1120 * and this will make PR_SET_PDEATHSIG work in most cases.
1121 * If this fails, ignore the error - but expect sd-pam threads
1122 * to fail to exit normally */
1124 r
= maybe_setgroups(0, NULL
);
1126 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1127 if (setresgid(gid
, gid
, gid
) < 0)
1128 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1129 if (setresuid(uid
, uid
, uid
) < 0)
1130 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1132 (void) ignore_signals(SIGPIPE
, -1);
1134 /* Wait until our parent died. This will only work if
1135 * the above setresuid() succeeds, otherwise the kernel
1136 * will not allow unprivileged parents kill their privileged
1137 * children this way. We rely on the control groups kill logic
1138 * to do the rest for us. */
1139 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1142 /* Tell the parent that our setup is done. This is especially
1143 * important regarding dropping privileges. Otherwise, unit
1144 * setup might race against our setresuid(2) call. */
1145 barrier_place(&barrier
);
1147 /* Check if our parent process might already have
1149 if (getppid() == parent_pid
) {
1152 assert_se(sigemptyset(&ss
) >= 0);
1153 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1156 if (sigwait(&ss
, &sig
) < 0) {
1163 assert(sig
== SIGTERM
);
1168 /* If our parent died we'll end the session */
1169 if (getppid() != parent_pid
) {
1170 pam_code
= pam_close_session(handle
, flags
);
1171 if (pam_code
!= PAM_SUCCESS
)
1178 pam_end(handle
, pam_code
| flags
);
1182 barrier_set_role(&barrier
, BARRIER_PARENT
);
1184 /* If the child was forked off successfully it will do all the
1185 * cleanups, so forget about the handle here. */
1188 /* Unblock SIGTERM again in the parent */
1189 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1191 /* We close the log explicitly here, since the PAM modules
1192 * might have opened it, but we don't want this fd around. */
1195 /* Synchronously wait for the child to initialize. We don't care for
1196 * errors as we cannot recover. However, warn loudly if it happens. */
1197 if (!barrier_place_and_sync(&barrier
))
1198 log_error("PAM initialization failed");
1206 if (pam_code
!= PAM_SUCCESS
) {
1207 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1208 r
= -EPERM
; /* PAM errors do not map to errno */
1210 log_error_errno(r
, "PAM failed: %m");
1214 pam_code
= pam_close_session(handle
, flags
);
1216 pam_end(handle
, pam_code
| flags
);
1228 static void rename_process_from_path(const char *path
) {
1229 char process_name
[11];
1233 /* This resulting string must fit in 10 chars (i.e. the length
1234 * of "/sbin/init") to look pretty in /bin/ps */
1238 rename_process("(...)");
1244 /* The end of the process name is usually more
1245 * interesting, since the first bit might just be
1251 process_name
[0] = '(';
1252 memcpy(process_name
+1, p
, l
);
1253 process_name
[1+l
] = ')';
1254 process_name
[1+l
+1] = 0;
1256 rename_process(process_name
);
1259 static bool context_has_address_families(const ExecContext
*c
) {
1262 return c
->address_families_whitelist
||
1263 !set_isempty(c
->address_families
);
1266 static bool context_has_syscall_filters(const ExecContext
*c
) {
1269 return c
->syscall_whitelist
||
1270 !set_isempty(c
->syscall_filter
);
1273 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1276 if (c
->no_new_privileges
)
1279 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1282 /* We need NNP if we have any form of seccomp and are unprivileged */
1283 return context_has_address_families(c
) ||
1284 c
->memory_deny_write_execute
||
1285 c
->restrict_realtime
||
1286 exec_context_restrict_namespaces_set(c
) ||
1287 c
->protect_kernel_tunables
||
1288 c
->protect_kernel_modules
||
1289 c
->private_devices
||
1290 context_has_syscall_filters(c
) ||
1291 !set_isempty(c
->syscall_archs
);
1296 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1298 if (is_seccomp_available())
1302 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1307 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
) {
1308 uint32_t negative_action
, default_action
, action
;
1313 if (!context_has_syscall_filters(c
))
1316 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1319 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1321 if (c
->syscall_whitelist
) {
1322 default_action
= negative_action
;
1323 action
= SCMP_ACT_ALLOW
;
1325 default_action
= SCMP_ACT_ALLOW
;
1326 action
= negative_action
;
1329 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
);
1332 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1336 if (set_isempty(c
->syscall_archs
))
1339 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1342 return seccomp_restrict_archs(c
->syscall_archs
);
1345 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1349 if (!context_has_address_families(c
))
1352 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1355 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1358 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1362 if (!c
->memory_deny_write_execute
)
1365 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1368 return seccomp_memory_deny_write_execute();
1371 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1375 if (!c
->restrict_realtime
)
1378 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1381 return seccomp_restrict_realtime();
1384 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1388 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1389 * let's protect even those systems where this is left on in the kernel. */
1391 if (!c
->protect_kernel_tunables
)
1394 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1397 return seccomp_protect_sysctl();
1400 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1404 /* Turn off module syscalls on ProtectKernelModules=yes */
1406 if (!c
->protect_kernel_modules
)
1409 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1412 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
));
1415 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1419 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1421 if (!c
->private_devices
)
1424 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1427 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
));
1430 static int apply_restrict_namespaces(Unit
*u
, const ExecContext
*c
) {
1434 if (!exec_context_restrict_namespaces_set(c
))
1437 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1440 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1445 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1448 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1449 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1451 if (idle_pipe
[0] >= 0) {
1454 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1456 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1459 /* Signal systemd that we are bored and want to continue. */
1460 n
= write(idle_pipe
[3], "x", 1);
1462 /* Wait for systemd to react to the signal above. */
1463 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1466 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1470 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1473 static int build_environment(
1475 const ExecContext
*c
,
1476 const ExecParameters
*p
,
1479 const char *username
,
1481 dev_t journal_stream_dev
,
1482 ino_t journal_stream_ino
,
1485 _cleanup_strv_free_
char **our_env
= NULL
;
1493 our_env
= new0(char*, 14);
1498 _cleanup_free_
char *joined
= NULL
;
1500 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid()) < 0)
1502 our_env
[n_env
++] = x
;
1504 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1506 our_env
[n_env
++] = x
;
1508 joined
= strv_join(p
->fd_names
, ":");
1512 x
= strjoin("LISTEN_FDNAMES=", joined
);
1515 our_env
[n_env
++] = x
;
1518 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1519 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid()) < 0)
1521 our_env
[n_env
++] = x
;
1523 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1525 our_env
[n_env
++] = x
;
1528 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1529 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1530 * check the database directly. */
1531 if (unit_has_name(u
, SPECIAL_DBUS_SERVICE
)) {
1532 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1535 our_env
[n_env
++] = x
;
1539 x
= strappend("HOME=", home
);
1542 our_env
[n_env
++] = x
;
1546 x
= strappend("LOGNAME=", username
);
1549 our_env
[n_env
++] = x
;
1551 x
= strappend("USER=", username
);
1554 our_env
[n_env
++] = x
;
1558 x
= strappend("SHELL=", shell
);
1561 our_env
[n_env
++] = x
;
1564 if (!sd_id128_is_null(u
->invocation_id
)) {
1565 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1568 our_env
[n_env
++] = x
;
1571 if (exec_context_needs_term(c
)) {
1572 const char *tty_path
, *term
= NULL
;
1574 tty_path
= exec_context_tty_path(c
);
1576 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1577 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1578 * passes to PID 1 ends up all the way in the console login shown. */
1580 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1581 term
= getenv("TERM");
1583 term
= default_term_for_tty(tty_path
);
1585 x
= strappend("TERM=", term
);
1588 our_env
[n_env
++] = x
;
1591 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1592 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1595 our_env
[n_env
++] = x
;
1598 our_env
[n_env
++] = NULL
;
1599 assert(n_env
<= 12);
1607 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1608 _cleanup_strv_free_
char **pass_env
= NULL
;
1609 size_t n_env
= 0, n_bufsize
= 0;
1612 STRV_FOREACH(i
, c
->pass_environment
) {
1613 _cleanup_free_
char *x
= NULL
;
1619 x
= strjoin(*i
, "=", v
);
1622 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1624 pass_env
[n_env
++] = x
;
1625 pass_env
[n_env
] = NULL
;
1635 static bool exec_needs_mount_namespace(
1636 const ExecContext
*context
,
1637 const ExecParameters
*params
,
1638 ExecRuntime
*runtime
) {
1643 if (context
->root_image
)
1646 if (!strv_isempty(context
->read_write_paths
) ||
1647 !strv_isempty(context
->read_only_paths
) ||
1648 !strv_isempty(context
->inaccessible_paths
))
1651 if (context
->n_bind_mounts
> 0)
1654 if (context
->mount_flags
!= 0)
1657 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1660 if (context
->private_devices
||
1661 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1662 context
->protect_home
!= PROTECT_HOME_NO
||
1663 context
->protect_kernel_tunables
||
1664 context
->protect_kernel_modules
||
1665 context
->protect_control_groups
)
1668 if (context
->mount_apivfs
)
1674 static int setup_private_users(uid_t uid
, gid_t gid
) {
1675 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1676 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1677 _cleanup_close_
int unshare_ready_fd
= -1;
1678 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1684 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1685 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1686 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1687 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1688 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1689 * continues execution normally. */
1691 if (uid
!= 0 && uid_is_valid(uid
)) {
1692 r
= asprintf(&uid_map
,
1693 "0 0 1\n" /* Map root → root */
1694 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1699 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1704 if (gid
!= 0 && gid_is_valid(gid
)) {
1705 r
= asprintf(&gid_map
,
1706 "0 0 1\n" /* Map root → root */
1707 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1712 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1717 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1719 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1720 if (unshare_ready_fd
< 0)
1723 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1725 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1733 _cleanup_close_
int fd
= -1;
1737 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1738 * here, after the parent opened its own user namespace. */
1741 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1743 /* Wait until the parent unshared the user namespace */
1744 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1749 /* Disable the setgroups() system call in the child user namespace, for good. */
1750 a
= procfs_file_alloca(ppid
, "setgroups");
1751 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1753 if (errno
!= ENOENT
) {
1758 /* If the file is missing the kernel is too old, let's continue anyway. */
1760 if (write(fd
, "deny\n", 5) < 0) {
1765 fd
= safe_close(fd
);
1768 /* First write the GID map */
1769 a
= procfs_file_alloca(ppid
, "gid_map");
1770 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1775 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1779 fd
= safe_close(fd
);
1781 /* The write the UID map */
1782 a
= procfs_file_alloca(ppid
, "uid_map");
1783 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1788 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1793 _exit(EXIT_SUCCESS
);
1796 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1797 _exit(EXIT_FAILURE
);
1800 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1802 if (unshare(CLONE_NEWUSER
) < 0)
1805 /* Let the child know that the namespace is ready now */
1806 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1809 /* Try to read an error code from the child */
1810 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1813 if (n
== sizeof(r
)) { /* an error code was sent to us */
1818 if (n
!= 0) /* on success we should have read 0 bytes */
1821 r
= wait_for_terminate(pid
, &si
);
1826 /* If something strange happened with the child, let's consider this fatal, too */
1827 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1833 static int setup_runtime_directory(
1834 const ExecContext
*context
,
1835 const ExecParameters
*params
,
1845 STRV_FOREACH(rt
, context
->runtime_directory
) {
1846 _cleanup_free_
char *p
;
1848 p
= strjoin(params
->runtime_prefix
, "/", *rt
);
1852 r
= mkdir_p_label(p
, context
->runtime_directory_mode
);
1856 r
= chmod_and_chown(p
, context
->runtime_directory_mode
, uid
, gid
);
1864 static int setup_smack(
1865 const ExecContext
*context
,
1866 const ExecCommand
*command
) {
1874 if (!mac_smack_use())
1877 if (context
->smack_process_label
) {
1878 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1882 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1884 _cleanup_free_
char *exec_label
= NULL
;
1886 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1887 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1890 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
1900 static int compile_read_write_paths(
1901 const ExecContext
*context
,
1902 const ExecParameters
*params
,
1905 _cleanup_strv_free_
char **l
= NULL
;
1908 /* Compile the list of writable paths. This is the combination of
1909 * the explicitly configured paths, plus all runtime directories. */
1911 if (strv_isempty(context
->read_write_paths
) &&
1912 strv_isempty(context
->runtime_directory
)) {
1913 *ret
= NULL
; /* NOP if neither is set */
1917 l
= strv_copy(context
->read_write_paths
);
1921 STRV_FOREACH(rt
, context
->runtime_directory
) {
1924 s
= strjoin(params
->runtime_prefix
, "/", *rt
);
1928 if (strv_consume(&l
, s
) < 0)
1938 static int apply_mount_namespace(Unit
*u
, const ExecContext
*context
,
1939 const ExecParameters
*params
,
1940 ExecRuntime
*runtime
) {
1942 _cleanup_strv_free_
char **rw
= NULL
;
1943 char *tmp
= NULL
, *var
= NULL
;
1944 const char *root_dir
= NULL
, *root_image
= NULL
;
1945 NameSpaceInfo ns_info
= {
1946 .ignore_protect_paths
= false,
1947 .private_dev
= context
->private_devices
,
1948 .protect_control_groups
= context
->protect_control_groups
,
1949 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
1950 .protect_kernel_modules
= context
->protect_kernel_modules
,
1951 .mount_apivfs
= context
->mount_apivfs
,
1956 /* The runtime struct only contains the parent of the private /tmp,
1957 * which is non-accessible to world users. Inside of it there's a /tmp
1958 * that is sticky, and that's the one we want to use here. */
1960 if (context
->private_tmp
&& runtime
) {
1961 if (runtime
->tmp_dir
)
1962 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
1963 if (runtime
->var_tmp_dir
)
1964 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
1967 r
= compile_read_write_paths(context
, params
, &rw
);
1971 if (params
->flags
& EXEC_APPLY_CHROOT
) {
1972 root_image
= context
->root_image
;
1975 root_dir
= context
->root_directory
;
1979 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
1980 * sandbox info, otherwise enforce it, don't ignore protected paths and
1981 * fail if we are enable to apply the sandbox inside the mount namespace.
1983 if (!context
->dynamic_user
&& root_dir
)
1984 ns_info
.ignore_protect_paths
= true;
1986 r
= setup_namespace(root_dir
, root_image
,
1988 context
->read_only_paths
,
1989 context
->inaccessible_paths
,
1990 context
->bind_mounts
,
1991 context
->n_bind_mounts
,
1994 context
->protect_home
,
1995 context
->protect_system
,
1996 context
->mount_flags
,
1997 DISSECT_IMAGE_DISCARD_ON_LOOP
);
1999 /* If we couldn't set up the namespace this is probably due to a
2000 * missing capability. In this case, silently proceeed. */
2001 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2003 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2011 static int apply_working_directory(
2012 const ExecContext
*context
,
2013 const ExecParameters
*params
,
2015 const bool needs_mount_ns
) {
2022 if (context
->working_directory_home
)
2024 else if (context
->working_directory
)
2025 wd
= context
->working_directory
;
2029 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2030 if (!needs_mount_ns
&& context
->root_directory
)
2031 if (chroot(context
->root_directory
) < 0)
2036 d
= strjoina(strempty(context
->root_directory
), "/", strempty(wd
));
2038 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
)
2044 static int setup_keyring(Unit
*u
, const ExecParameters
*p
, uid_t uid
, gid_t gid
) {
2045 key_serial_t keyring
;
2050 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2051 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2052 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2053 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2054 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2055 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2057 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2060 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2061 if (keyring
== -1) {
2062 if (errno
== ENOSYS
)
2063 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2064 else if (IN_SET(errno
, EACCES
, EPERM
))
2065 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2066 else if (errno
== EDQUOT
)
2067 log_debug_errno(errno
, "Out of kernel keyrings to allocate, ignoring.");
2069 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2074 /* Populate they keyring with the invocation ID by default. */
2075 if (!sd_id128_is_null(u
->invocation_id
)) {
2078 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2080 log_debug_errno(errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2082 if (keyctl(KEYCTL_SETPERM
, key
,
2083 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2084 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2085 return log_error_errno(errno
, "Failed to restrict invocation ID permission: %m");
2089 /* And now, make the keyring owned by the service's user */
2090 if (uid_is_valid(uid
) || gid_is_valid(gid
))
2091 if (keyctl(KEYCTL_CHOWN
, keyring
, uid
, gid
, 0) < 0)
2092 return log_error_errno(errno
, "Failed to change ownership of session keyring: %m");
2097 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
2105 array
[(*n
)++] = pair
[0];
2107 array
[(*n
)++] = pair
[1];
2110 static int close_remaining_fds(
2111 const ExecParameters
*params
,
2112 ExecRuntime
*runtime
,
2113 DynamicCreds
*dcreds
,
2116 int *fds
, unsigned n_fds
) {
2118 unsigned n_dont_close
= 0;
2119 int dont_close
[n_fds
+ 12];
2123 if (params
->stdin_fd
>= 0)
2124 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2125 if (params
->stdout_fd
>= 0)
2126 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2127 if (params
->stderr_fd
>= 0)
2128 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2131 dont_close
[n_dont_close
++] = socket_fd
;
2133 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2134 n_dont_close
+= n_fds
;
2138 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2142 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2144 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2147 if (user_lookup_fd
>= 0)
2148 dont_close
[n_dont_close
++] = user_lookup_fd
;
2150 return close_all_fds(dont_close
, n_dont_close
);
2153 static int send_user_lookup(
2161 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2162 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2165 if (user_lookup_fd
< 0)
2168 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2171 if (writev(user_lookup_fd
,
2173 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2174 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2175 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2181 static int exec_child(
2183 ExecCommand
*command
,
2184 const ExecContext
*context
,
2185 const ExecParameters
*params
,
2186 ExecRuntime
*runtime
,
2187 DynamicCreds
*dcreds
,
2191 int *fds
, unsigned n_fds
,
2195 char **error_message
) {
2197 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2198 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2199 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2200 const char *username
= NULL
, *groupname
= NULL
;
2201 const char *home
= NULL
, *shell
= NULL
;
2202 dev_t journal_stream_dev
= 0;
2203 ino_t journal_stream_ino
= 0;
2204 bool needs_mount_namespace
;
2205 uid_t uid
= UID_INVALID
;
2206 gid_t gid
= GID_INVALID
;
2207 int i
, r
, ngids
= 0;
2213 assert(exit_status
);
2214 assert(error_message
);
2215 /* We don't always set error_message, hence it must be initialized */
2216 assert(*error_message
== NULL
);
2218 rename_process_from_path(command
->path
);
2220 /* We reset exactly these signals, since they are the
2221 * only ones we set to SIG_IGN in the main daemon. All
2222 * others we leave untouched because we set them to
2223 * SIG_DFL or a valid handler initially, both of which
2224 * will be demoted to SIG_DFL. */
2225 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2226 SIGNALS_IGNORE
, -1);
2228 if (context
->ignore_sigpipe
)
2229 (void) ignore_signals(SIGPIPE
, -1);
2231 r
= reset_signal_mask();
2233 *exit_status
= EXIT_SIGNAL_MASK
;
2234 *error_message
= strdup("Failed to reset signal mask");
2235 /* If strdup fails, here and below, we will just print the generic error message. */
2239 if (params
->idle_pipe
)
2240 do_idle_pipe_dance(params
->idle_pipe
);
2242 /* Close sockets very early to make sure we don't
2243 * block init reexecution because it cannot bind its
2248 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2250 *exit_status
= EXIT_FDS
;
2251 *error_message
= strdup("Failed to close remaining fds");
2255 if (!context
->same_pgrp
)
2257 *exit_status
= EXIT_SETSID
;
2261 exec_context_tty_reset(context
, params
);
2263 if (unit_shall_confirm_spawn(unit
)) {
2264 const char *vc
= params
->confirm_spawn
;
2265 _cleanup_free_
char *cmdline
= NULL
;
2267 cmdline
= exec_command_line(argv
);
2269 *exit_status
= EXIT_CONFIRM
;
2273 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2274 if (r
!= CONFIRM_EXECUTE
) {
2275 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2276 *exit_status
= EXIT_SUCCESS
;
2279 *exit_status
= EXIT_CONFIRM
;
2280 *error_message
= strdup("Execution cancelled");
2285 if (context
->dynamic_user
&& dcreds
) {
2287 /* Make sure we bypass our own NSS module for any NSS checks */
2288 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2289 *exit_status
= EXIT_USER
;
2290 *error_message
= strdup("Failed to update environment");
2294 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2296 *exit_status
= EXIT_USER
;
2297 *error_message
= strdup("Failed to update dynamic user credentials");
2301 if (!uid_is_valid(uid
)) {
2302 *exit_status
= EXIT_USER
;
2303 (void) asprintf(error_message
, "UID validation failed for \""UID_FMT
"\"", uid
);
2304 /* If asprintf fails, here and below, we will just print the generic error message. */
2308 if (!gid_is_valid(gid
)) {
2309 *exit_status
= EXIT_USER
;
2310 (void) asprintf(error_message
, "GID validation failed for \""GID_FMT
"\"", gid
);
2315 username
= dcreds
->user
->name
;
2318 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2320 *exit_status
= EXIT_USER
;
2321 *error_message
= strdup("Failed to determine user credentials");
2325 r
= get_fixed_group(context
, &groupname
, &gid
);
2327 *exit_status
= EXIT_GROUP
;
2328 *error_message
= strdup("Failed to determine group credentials");
2333 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2334 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2335 &supplementary_gids
, &ngids
);
2337 *exit_status
= EXIT_GROUP
;
2338 *error_message
= strdup("Failed to determine supplementary groups");
2342 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2344 *exit_status
= EXIT_USER
;
2345 *error_message
= strdup("Failed to send user credentials to PID1");
2349 user_lookup_fd
= safe_close(user_lookup_fd
);
2351 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2352 * must sure to drop O_NONBLOCK */
2354 (void) fd_nonblock(socket_fd
, false);
2356 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2358 *exit_status
= EXIT_STDIN
;
2359 *error_message
= strdup("Failed to set up stdin");
2363 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2365 *exit_status
= EXIT_STDOUT
;
2366 *error_message
= strdup("Failed to set up stdout");
2370 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2372 *exit_status
= EXIT_STDERR
;
2373 *error_message
= strdup("Failed to set up stderr");
2377 if (params
->cgroup_path
) {
2378 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2380 *exit_status
= EXIT_CGROUP
;
2381 (void) asprintf(error_message
, "Failed to attach to cgroup %s", params
->cgroup_path
);
2386 if (context
->oom_score_adjust_set
) {
2387 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2389 /* When we can't make this change due to EPERM, then
2390 * let's silently skip over it. User namespaces
2391 * prohibit write access to this file, and we
2392 * shouldn't trip up over that. */
2394 sprintf(t
, "%i", context
->oom_score_adjust
);
2395 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2396 if (r
== -EPERM
|| r
== -EACCES
) {
2398 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2401 *exit_status
= EXIT_OOM_ADJUST
;
2402 *error_message
= strdup("Failed to write /proc/self/oom_score_adj");
2407 if (context
->nice_set
)
2408 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2409 *exit_status
= EXIT_NICE
;
2413 if (context
->cpu_sched_set
) {
2414 struct sched_param param
= {
2415 .sched_priority
= context
->cpu_sched_priority
,
2418 r
= sched_setscheduler(0,
2419 context
->cpu_sched_policy
|
2420 (context
->cpu_sched_reset_on_fork
?
2421 SCHED_RESET_ON_FORK
: 0),
2424 *exit_status
= EXIT_SETSCHEDULER
;
2429 if (context
->cpuset
)
2430 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2431 *exit_status
= EXIT_CPUAFFINITY
;
2435 if (context
->ioprio_set
)
2436 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2437 *exit_status
= EXIT_IOPRIO
;
2441 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2442 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2443 *exit_status
= EXIT_TIMERSLACK
;
2447 if (context
->personality
!= PERSONALITY_INVALID
)
2448 if (personality(context
->personality
) < 0) {
2449 *exit_status
= EXIT_PERSONALITY
;
2453 if (context
->utmp_id
)
2454 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0),
2456 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2457 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2461 if (context
->user
) {
2462 r
= chown_terminal(STDIN_FILENO
, uid
);
2464 *exit_status
= EXIT_STDIN
;
2469 /* If delegation is enabled we'll pass ownership of the cgroup
2470 * (but only in systemd's own controller hierarchy!) to the
2471 * user of the new process. */
2472 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2473 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2475 *exit_status
= EXIT_CGROUP
;
2480 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2482 *exit_status
= EXIT_CGROUP
;
2487 if (!strv_isempty(context
->runtime_directory
) && params
->runtime_prefix
) {
2488 r
= setup_runtime_directory(context
, params
, uid
, gid
);
2490 *exit_status
= EXIT_RUNTIME_DIRECTORY
;
2495 r
= build_environment(
2507 *exit_status
= EXIT_MEMORY
;
2511 r
= build_pass_environment(context
, &pass_env
);
2513 *exit_status
= EXIT_MEMORY
;
2517 accum_env
= strv_env_merge(5,
2518 params
->environment
,
2521 context
->environment
,
2525 *exit_status
= EXIT_MEMORY
;
2528 accum_env
= strv_env_clean(accum_env
);
2530 (void) umask(context
->umask
);
2532 r
= setup_keyring(unit
, params
, uid
, gid
);
2534 *exit_status
= EXIT_KEYRING
;
2538 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2539 if (context
->pam_name
&& username
) {
2540 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2542 *exit_status
= EXIT_PAM
;
2548 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2549 r
= setup_netns(runtime
->netns_storage_socket
);
2551 *exit_status
= EXIT_NETWORK
;
2556 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2557 if (needs_mount_namespace
) {
2558 r
= apply_mount_namespace(unit
, context
, params
, runtime
);
2560 *exit_status
= EXIT_NAMESPACE
;
2565 /* Apply just after mount namespace setup */
2566 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
);
2568 *exit_status
= EXIT_CHROOT
;
2572 /* Drop groups as early as possbile */
2573 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2574 r
= enforce_groups(context
, gid
, supplementary_gids
, ngids
);
2576 *exit_status
= EXIT_GROUP
;
2582 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) &&
2583 mac_selinux_use() &&
2584 params
->selinux_context_net
&&
2586 !command
->privileged
) {
2588 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2590 *exit_status
= EXIT_SELINUX_CONTEXT
;
2596 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2597 r
= setup_private_users(uid
, gid
);
2599 *exit_status
= EXIT_USER
;
2604 /* We repeat the fd closing here, to make sure that
2605 * nothing is leaked from the PAM modules. Note that
2606 * we are more aggressive this time since socket_fd
2607 * and the netns fds we don't need anymore. The custom
2608 * endpoint fd was needed to upload the policy and can
2609 * now be closed as well. */
2610 r
= close_all_fds(fds
, n_fds
);
2612 r
= shift_fds(fds
, n_fds
);
2614 r
= flags_fds(fds
, n_fds
, context
->non_blocking
);
2616 *exit_status
= EXIT_FDS
;
2620 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2622 int secure_bits
= context
->secure_bits
;
2624 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2626 if (!context
->rlimit
[i
])
2629 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2631 *exit_status
= EXIT_LIMITS
;
2636 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2637 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2638 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2639 *exit_status
= EXIT_LIMITS
;
2644 if (!cap_test_all(context
->capability_bounding_set
)) {
2645 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2647 *exit_status
= EXIT_CAPABILITIES
;
2648 *error_message
= strdup("Failed to drop capabilities");
2653 /* This is done before enforce_user, but ambient set
2654 * does not survive over setresuid() if keep_caps is not set. */
2655 if (context
->capability_ambient_set
!= 0) {
2656 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2658 *exit_status
= EXIT_CAPABILITIES
;
2659 *error_message
= strdup("Failed to apply ambient capabilities (before UID change)");
2664 if (context
->user
) {
2665 r
= enforce_user(context
, uid
);
2667 *exit_status
= EXIT_USER
;
2668 (void) asprintf(error_message
, "Failed to change UID to "UID_FMT
, uid
);
2671 if (context
->capability_ambient_set
!= 0) {
2673 /* Fix the ambient capabilities after user change. */
2674 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2676 *exit_status
= EXIT_CAPABILITIES
;
2677 *error_message
= strdup("Failed to apply ambient capabilities (after UID change)");
2681 /* If we were asked to change user and ambient capabilities
2682 * were requested, we had to add keep-caps to the securebits
2683 * so that we would maintain the inherited capability set
2684 * through the setresuid(). Make sure that the bit is added
2685 * also to the context secure_bits so that we don't try to
2686 * drop the bit away next. */
2688 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2692 /* Apply the MAC contexts late, but before seccomp syscall filtering, as those should really be last to
2693 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
2694 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
2695 * are restricted. */
2698 if (mac_selinux_use()) {
2699 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2702 r
= setexeccon(exec_context
);
2704 *exit_status
= EXIT_SELINUX_CONTEXT
;
2705 (void) asprintf(error_message
, "Failed to set SELinux context to %s", exec_context
);
2712 r
= setup_smack(context
, command
);
2714 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2715 *error_message
= strdup("Failed to set SMACK process label");
2719 #ifdef HAVE_APPARMOR
2720 if (context
->apparmor_profile
&& mac_apparmor_use()) {
2721 r
= aa_change_onexec(context
->apparmor_profile
);
2722 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2723 *exit_status
= EXIT_APPARMOR_PROFILE
;
2724 (void) asprintf(error_message
,
2725 "Failed to prepare AppArmor profile change to %s",
2726 context
->apparmor_profile
);
2732 /* PR_GET_SECUREBITS is not privileged, while
2733 * PR_SET_SECUREBITS is. So to suppress
2734 * potential EPERMs we'll try not to call
2735 * PR_SET_SECUREBITS unless necessary. */
2736 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2737 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2738 *exit_status
= EXIT_SECUREBITS
;
2739 *error_message
= strdup("Failed to set secure bits");
2743 if (context_has_no_new_privileges(context
))
2744 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2745 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2746 *error_message
= strdup("Failed to disable new privileges");
2751 r
= apply_address_families(unit
, context
);
2753 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2754 *error_message
= strdup("Failed to restrict address families");
2758 r
= apply_memory_deny_write_execute(unit
, context
);
2760 *exit_status
= EXIT_SECCOMP
;
2761 *error_message
= strdup("Failed to disable writing to executable memory");
2765 r
= apply_restrict_realtime(unit
, context
);
2767 *exit_status
= EXIT_SECCOMP
;
2768 *error_message
= strdup("Failed to apply realtime restrictions");
2772 r
= apply_restrict_namespaces(unit
, context
);
2774 *exit_status
= EXIT_SECCOMP
;
2775 *error_message
= strdup("Failed to apply namespace restrictions");
2779 r
= apply_protect_sysctl(unit
, context
);
2781 *exit_status
= EXIT_SECCOMP
;
2782 *error_message
= strdup("Failed to apply sysctl restrictions");
2786 r
= apply_protect_kernel_modules(unit
, context
);
2788 *exit_status
= EXIT_SECCOMP
;
2789 *error_message
= strdup("Failed to apply module loading restrictions");
2793 r
= apply_private_devices(unit
, context
);
2795 *exit_status
= EXIT_SECCOMP
;
2796 *error_message
= strdup("Failed to set up private devices");
2800 r
= apply_syscall_archs(unit
, context
);
2802 *exit_status
= EXIT_SECCOMP
;
2803 *error_message
= strdup("Failed to apply syscall architecture restrictions");
2807 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
2808 * by the filter as little as possible. */
2809 r
= apply_syscall_filter(unit
, context
);
2811 *exit_status
= EXIT_SECCOMP
;
2812 *error_message
= strdup("Failed to apply syscall filters");
2818 final_argv
= replace_env_argv(argv
, accum_env
);
2820 *exit_status
= EXIT_MEMORY
;
2821 *error_message
= strdup("Failed to prepare process arguments");
2825 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2826 _cleanup_free_
char *line
;
2828 line
= exec_command_line(final_argv
);
2831 log_struct(LOG_DEBUG
,
2833 "EXECUTABLE=%s", command
->path
,
2834 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2840 execve(command
->path
, final_argv
, accum_env
);
2841 *exit_status
= EXIT_EXEC
;
2845 int exec_spawn(Unit
*unit
,
2846 ExecCommand
*command
,
2847 const ExecContext
*context
,
2848 const ExecParameters
*params
,
2849 ExecRuntime
*runtime
,
2850 DynamicCreds
*dcreds
,
2853 _cleanup_strv_free_
char **files_env
= NULL
;
2854 int *fds
= NULL
; unsigned n_fds
= 0;
2855 _cleanup_free_
char *line
= NULL
;
2857 int named_iofds
[3] = { -1, -1, -1 };
2866 assert(params
->fds
|| params
->n_fds
<= 0);
2868 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2869 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2870 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
2872 if (params
->n_fds
!= 1) {
2873 log_unit_error(unit
, "Got more than one socket.");
2877 socket_fd
= params
->fds
[0];
2881 n_fds
= params
->n_fds
;
2884 r
= exec_context_named_iofds(unit
, context
, params
, named_iofds
);
2886 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
2888 r
= exec_context_load_environment(unit
, context
, &files_env
);
2890 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
2892 argv
= params
->argv
?: command
->argv
;
2893 line
= exec_command_line(argv
);
2897 log_struct(LOG_DEBUG
,
2899 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
2900 "EXECUTABLE=%s", command
->path
,
2904 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
2908 _cleanup_free_
char *error_message
= NULL
;
2910 r
= exec_child(unit
,
2921 unit
->manager
->user_lookup_fds
[1],
2927 log_struct_errno(LOG_ERR
, r
,
2928 LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED
),
2930 LOG_UNIT_MESSAGE(unit
, "%s: %m",
2932 "EXECUTABLE=%s", command
->path
,
2935 log_struct_errno(LOG_ERR
, r
,
2936 LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED
),
2938 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
2939 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
2941 "EXECUTABLE=%s", command
->path
,
2948 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
2950 /* We add the new process to the cgroup both in the child (so
2951 * that we can be sure that no user code is ever executed
2952 * outside of the cgroup) and in the parent (so that we can be
2953 * sure that when we kill the cgroup the process will be
2955 if (params
->cgroup_path
)
2956 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
2958 exec_status_start(&command
->exec_status
, pid
);
2964 void exec_context_init(ExecContext
*c
) {
2968 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
2969 c
->cpu_sched_policy
= SCHED_OTHER
;
2970 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
2971 c
->syslog_level_prefix
= true;
2972 c
->ignore_sigpipe
= true;
2973 c
->timer_slack_nsec
= NSEC_INFINITY
;
2974 c
->personality
= PERSONALITY_INVALID
;
2975 c
->runtime_directory_mode
= 0755;
2976 c
->capability_bounding_set
= CAP_ALL
;
2977 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
2980 void exec_context_done(ExecContext
*c
) {
2985 c
->environment
= strv_free(c
->environment
);
2986 c
->environment_files
= strv_free(c
->environment_files
);
2987 c
->pass_environment
= strv_free(c
->pass_environment
);
2989 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
2990 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
2992 for (l
= 0; l
< 3; l
++)
2993 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
2995 c
->working_directory
= mfree(c
->working_directory
);
2996 c
->root_directory
= mfree(c
->root_directory
);
2997 c
->root_image
= mfree(c
->root_image
);
2998 c
->tty_path
= mfree(c
->tty_path
);
2999 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3000 c
->user
= mfree(c
->user
);
3001 c
->group
= mfree(c
->group
);
3003 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3005 c
->pam_name
= mfree(c
->pam_name
);
3007 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3008 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3009 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3011 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3014 CPU_FREE(c
->cpuset
);
3016 c
->utmp_id
= mfree(c
->utmp_id
);
3017 c
->selinux_context
= mfree(c
->selinux_context
);
3018 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3020 c
->syscall_filter
= set_free(c
->syscall_filter
);
3021 c
->syscall_archs
= set_free(c
->syscall_archs
);
3022 c
->address_families
= set_free(c
->address_families
);
3024 c
->runtime_directory
= strv_free(c
->runtime_directory
);
3027 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
3032 if (!runtime_prefix
)
3035 STRV_FOREACH(i
, c
->runtime_directory
) {
3036 _cleanup_free_
char *p
;
3038 p
= strjoin(runtime_prefix
, "/", *i
);
3042 /* We execute this synchronously, since we need to be
3043 * sure this is gone when we start the service
3045 (void) rm_rf(p
, REMOVE_ROOT
);
3051 void exec_command_done(ExecCommand
*c
) {
3054 c
->path
= mfree(c
->path
);
3056 c
->argv
= strv_free(c
->argv
);
3059 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3062 for (i
= 0; i
< n
; i
++)
3063 exec_command_done(c
+i
);
3066 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3070 LIST_REMOVE(command
, c
, i
);
3071 exec_command_done(i
);
3078 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3081 for (i
= 0; i
< n
; i
++)
3082 c
[i
] = exec_command_free_list(c
[i
]);
3085 typedef struct InvalidEnvInfo
{
3090 static void invalid_env(const char *p
, void *userdata
) {
3091 InvalidEnvInfo
*info
= userdata
;
3093 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3096 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3101 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3103 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3105 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3107 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3109 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3111 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3117 int exec_context_named_iofds(Unit
*unit
, const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3118 unsigned i
, targets
;
3119 const char* stdio_fdname
[3];
3124 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3125 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3126 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3128 for (i
= 0; i
< 3; i
++)
3129 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3131 for (i
= 0; i
< p
->n_fds
&& targets
> 0; i
++)
3132 if (named_iofds
[STDIN_FILENO
] < 0 &&
3133 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3134 stdio_fdname
[STDIN_FILENO
] &&
3135 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3137 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3140 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3141 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3142 stdio_fdname
[STDOUT_FILENO
] &&
3143 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3145 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3148 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3149 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3150 stdio_fdname
[STDERR_FILENO
] &&
3151 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3153 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3157 return targets
== 0 ? 0 : -ENOENT
;
3160 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
3161 char **i
, **r
= NULL
;
3166 STRV_FOREACH(i
, c
->environment_files
) {
3169 bool ignore
= false;
3171 _cleanup_globfree_ glob_t pglob
= {};
3181 if (!path_is_absolute(fn
)) {
3189 /* Filename supports globbing, take all matching files */
3191 if (glob(fn
, 0, NULL
, &pglob
) != 0) {
3196 return errno
> 0 ? -errno
: -EINVAL
;
3198 count
= pglob
.gl_pathc
;
3206 for (n
= 0; n
< count
; n
++) {
3207 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3215 /* Log invalid environment variables with filename */
3217 InvalidEnvInfo info
= {
3219 .path
= pglob
.gl_pathv
[n
]
3222 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3230 m
= strv_env_merge(2, r
, p
);
3246 static bool tty_may_match_dev_console(const char *tty
) {
3247 _cleanup_free_
char *active
= NULL
;
3253 if (startswith(tty
, "/dev/"))
3256 /* trivial identity? */
3257 if (streq(tty
, "console"))
3260 console
= resolve_dev_console(&active
);
3261 /* if we could not resolve, assume it may */
3265 /* "tty0" means the active VC, so it may be the same sometimes */
3266 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3269 bool exec_context_may_touch_console(ExecContext
*ec
) {
3271 return (ec
->tty_reset
||
3273 ec
->tty_vt_disallocate
||
3274 is_terminal_input(ec
->std_input
) ||
3275 is_terminal_output(ec
->std_output
) ||
3276 is_terminal_output(ec
->std_error
)) &&
3277 tty_may_match_dev_console(exec_context_tty_path(ec
));
3280 static void strv_fprintf(FILE *f
, char **l
) {
3286 fprintf(f
, " %s", *g
);
3289 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3297 prefix
= strempty(prefix
);
3301 "%sWorkingDirectory: %s\n"
3302 "%sRootDirectory: %s\n"
3303 "%sNonBlocking: %s\n"
3304 "%sPrivateTmp: %s\n"
3305 "%sPrivateDevices: %s\n"
3306 "%sProtectKernelTunables: %s\n"
3307 "%sProtectKernelModules: %s\n"
3308 "%sProtectControlGroups: %s\n"
3309 "%sPrivateNetwork: %s\n"
3310 "%sPrivateUsers: %s\n"
3311 "%sProtectHome: %s\n"
3312 "%sProtectSystem: %s\n"
3313 "%sMountAPIVFS: %s\n"
3314 "%sIgnoreSIGPIPE: %s\n"
3315 "%sMemoryDenyWriteExecute: %s\n"
3316 "%sRestrictRealtime: %s\n",
3318 prefix
, c
->working_directory
? c
->working_directory
: "/",
3319 prefix
, c
->root_directory
? c
->root_directory
: "/",
3320 prefix
, yes_no(c
->non_blocking
),
3321 prefix
, yes_no(c
->private_tmp
),
3322 prefix
, yes_no(c
->private_devices
),
3323 prefix
, yes_no(c
->protect_kernel_tunables
),
3324 prefix
, yes_no(c
->protect_kernel_modules
),
3325 prefix
, yes_no(c
->protect_control_groups
),
3326 prefix
, yes_no(c
->private_network
),
3327 prefix
, yes_no(c
->private_users
),
3328 prefix
, protect_home_to_string(c
->protect_home
),
3329 prefix
, protect_system_to_string(c
->protect_system
),
3330 prefix
, yes_no(c
->mount_apivfs
),
3331 prefix
, yes_no(c
->ignore_sigpipe
),
3332 prefix
, yes_no(c
->memory_deny_write_execute
),
3333 prefix
, yes_no(c
->restrict_realtime
));
3336 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3338 STRV_FOREACH(e
, c
->environment
)
3339 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3341 STRV_FOREACH(e
, c
->environment_files
)
3342 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3344 STRV_FOREACH(e
, c
->pass_environment
)
3345 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3347 fprintf(f
, "%sRuntimeDirectoryMode: %04o\n", prefix
, c
->runtime_directory_mode
);
3349 STRV_FOREACH(d
, c
->runtime_directory
)
3350 fprintf(f
, "%sRuntimeDirectory: %s\n", prefix
, *d
);
3357 if (c
->oom_score_adjust_set
)
3359 "%sOOMScoreAdjust: %i\n",
3360 prefix
, c
->oom_score_adjust
);
3362 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3364 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3365 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3366 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3367 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3370 if (c
->ioprio_set
) {
3371 _cleanup_free_
char *class_str
= NULL
;
3373 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3375 "%sIOSchedulingClass: %s\n"
3376 "%sIOPriority: %i\n",
3377 prefix
, strna(class_str
),
3378 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
3381 if (c
->cpu_sched_set
) {
3382 _cleanup_free_
char *policy_str
= NULL
;
3384 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3386 "%sCPUSchedulingPolicy: %s\n"
3387 "%sCPUSchedulingPriority: %i\n"
3388 "%sCPUSchedulingResetOnFork: %s\n",
3389 prefix
, strna(policy_str
),
3390 prefix
, c
->cpu_sched_priority
,
3391 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3395 fprintf(f
, "%sCPUAffinity:", prefix
);
3396 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3397 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3398 fprintf(f
, " %u", i
);
3402 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3403 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3406 "%sStandardInput: %s\n"
3407 "%sStandardOutput: %s\n"
3408 "%sStandardError: %s\n",
3409 prefix
, exec_input_to_string(c
->std_input
),
3410 prefix
, exec_output_to_string(c
->std_output
),
3411 prefix
, exec_output_to_string(c
->std_error
));
3417 "%sTTYVHangup: %s\n"
3418 "%sTTYVTDisallocate: %s\n",
3419 prefix
, c
->tty_path
,
3420 prefix
, yes_no(c
->tty_reset
),
3421 prefix
, yes_no(c
->tty_vhangup
),
3422 prefix
, yes_no(c
->tty_vt_disallocate
));
3424 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3425 c
->std_output
== EXEC_OUTPUT_KMSG
||
3426 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3427 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3428 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3429 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3430 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3431 c
->std_error
== EXEC_OUTPUT_KMSG
||
3432 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3433 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3434 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3435 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3437 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3439 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3440 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3443 "%sSyslogFacility: %s\n"
3444 "%sSyslogLevel: %s\n",
3445 prefix
, strna(fac_str
),
3446 prefix
, strna(lvl_str
));
3450 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
3452 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
3453 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
3454 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
3455 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
3456 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
3457 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
3459 if (c
->capability_bounding_set
!= CAP_ALL
) {
3461 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
3463 for (l
= 0; l
<= cap_last_cap(); l
++)
3464 if (c
->capability_bounding_set
& (UINT64_C(1) << l
))
3465 fprintf(f
, " %s", strna(capability_to_name(l
)));
3470 if (c
->capability_ambient_set
!= 0) {
3472 fprintf(f
, "%sAmbientCapabilities:", prefix
);
3474 for (l
= 0; l
<= cap_last_cap(); l
++)
3475 if (c
->capability_ambient_set
& (UINT64_C(1) << l
))
3476 fprintf(f
, " %s", strna(capability_to_name(l
)));
3482 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3484 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3486 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3488 if (strv_length(c
->supplementary_groups
) > 0) {
3489 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3490 strv_fprintf(f
, c
->supplementary_groups
);
3495 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3497 if (strv_length(c
->read_write_paths
) > 0) {
3498 fprintf(f
, "%sReadWritePaths:", prefix
);
3499 strv_fprintf(f
, c
->read_write_paths
);
3503 if (strv_length(c
->read_only_paths
) > 0) {
3504 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3505 strv_fprintf(f
, c
->read_only_paths
);
3509 if (strv_length(c
->inaccessible_paths
) > 0) {
3510 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3511 strv_fprintf(f
, c
->inaccessible_paths
);
3515 if (c
->n_bind_mounts
> 0)
3516 for (i
= 0; i
< c
->n_bind_mounts
; i
++) {
3517 fprintf(f
, "%s%s: %s:%s:%s\n", prefix
,
3518 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
3519 c
->bind_mounts
[i
].source
,
3520 c
->bind_mounts
[i
].destination
,
3521 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
3526 "%sUtmpIdentifier: %s\n",
3527 prefix
, c
->utmp_id
);
3529 if (c
->selinux_context
)
3531 "%sSELinuxContext: %s%s\n",
3532 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3534 if (c
->personality
!= PERSONALITY_INVALID
)
3536 "%sPersonality: %s\n",
3537 prefix
, strna(personality_to_string(c
->personality
)));
3539 if (c
->syscall_filter
) {
3547 "%sSystemCallFilter: ",
3550 if (!c
->syscall_whitelist
)
3554 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3555 _cleanup_free_
char *name
= NULL
;
3562 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3563 fputs(strna(name
), f
);
3570 if (c
->syscall_archs
) {
3577 "%sSystemCallArchitectures:",
3581 SET_FOREACH(id
, c
->syscall_archs
, j
)
3582 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3587 if (exec_context_restrict_namespaces_set(c
)) {
3588 _cleanup_free_
char *s
= NULL
;
3590 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
3592 fprintf(f
, "%sRestrictNamespaces: %s\n",
3596 if (c
->syscall_errno
> 0)
3598 "%sSystemCallErrorNumber: %s\n",
3599 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3601 if (c
->apparmor_profile
)
3603 "%sAppArmorProfile: %s%s\n",
3604 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3607 bool exec_context_maintains_privileges(ExecContext
*c
) {
3610 /* Returns true if the process forked off would run under
3611 * an unchanged UID or as root. */
3616 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3622 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3627 dual_timestamp_get(&s
->start_timestamp
);
3630 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3633 if (s
->pid
&& s
->pid
!= pid
)
3637 dual_timestamp_get(&s
->exit_timestamp
);
3643 if (context
->utmp_id
)
3644 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3646 exec_context_tty_reset(context
, NULL
);
3650 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3651 char buf
[FORMAT_TIMESTAMP_MAX
];
3659 prefix
= strempty(prefix
);
3662 "%sPID: "PID_FMT
"\n",
3665 if (dual_timestamp_is_set(&s
->start_timestamp
))
3667 "%sStart Timestamp: %s\n",
3668 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3670 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3672 "%sExit Timestamp: %s\n"
3674 "%sExit Status: %i\n",
3675 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3676 prefix
, sigchld_code_to_string(s
->code
),
3680 char *exec_command_line(char **argv
) {
3688 STRV_FOREACH(a
, argv
)
3696 STRV_FOREACH(a
, argv
) {
3703 if (strpbrk(*a
, WHITESPACE
)) {
3714 /* FIXME: this doesn't really handle arguments that have
3715 * spaces and ticks in them */
3720 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3721 _cleanup_free_
char *cmd
= NULL
;
3722 const char *prefix2
;
3727 prefix
= strempty(prefix
);
3728 prefix2
= strjoina(prefix
, "\t");
3730 cmd
= exec_command_line(c
->argv
);
3732 "%sCommand Line: %s\n",
3733 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3735 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3738 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3741 prefix
= strempty(prefix
);
3743 LIST_FOREACH(command
, c
, c
)
3744 exec_command_dump(c
, f
, prefix
);
3747 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3754 /* It's kind of important, that we keep the order here */
3755 LIST_FIND_TAIL(command
, *l
, end
);
3756 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3761 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3769 l
= strv_new_ap(path
, ap
);
3790 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3791 _cleanup_strv_free_
char **l
= NULL
;
3799 l
= strv_new_ap(path
, ap
);
3805 r
= strv_extend_strv(&c
->argv
, l
, false);
3813 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3818 *rt
= new0(ExecRuntime
, 1);
3823 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3828 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
3838 if (!c
->private_network
&& !c
->private_tmp
)
3841 r
= exec_runtime_allocate(rt
);
3845 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
3846 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
3850 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
3851 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
3859 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
3861 assert(r
->n_ref
> 0);
3867 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
3872 assert(r
->n_ref
> 0);
3879 free(r
->var_tmp_dir
);
3880 safe_close_pair(r
->netns_storage_socket
);
3884 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
3893 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
3895 if (rt
->var_tmp_dir
)
3896 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
3898 if (rt
->netns_storage_socket
[0] >= 0) {
3901 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
3905 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
3908 if (rt
->netns_storage_socket
[1] >= 0) {
3911 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
3915 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
3921 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
3928 if (streq(key
, "tmp-dir")) {
3931 r
= exec_runtime_allocate(rt
);
3935 copy
= strdup(value
);
3939 free((*rt
)->tmp_dir
);
3940 (*rt
)->tmp_dir
= copy
;
3942 } else if (streq(key
, "var-tmp-dir")) {
3945 r
= exec_runtime_allocate(rt
);
3949 copy
= strdup(value
);
3953 free((*rt
)->var_tmp_dir
);
3954 (*rt
)->var_tmp_dir
= copy
;
3956 } else if (streq(key
, "netns-socket-0")) {
3959 r
= exec_runtime_allocate(rt
);
3963 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
3964 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
3966 safe_close((*rt
)->netns_storage_socket
[0]);
3967 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
3969 } else if (streq(key
, "netns-socket-1")) {
3972 r
= exec_runtime_allocate(rt
);
3976 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
3977 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
3979 safe_close((*rt
)->netns_storage_socket
[1]);
3980 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
3988 static void *remove_tmpdir_thread(void *p
) {
3989 _cleanup_free_
char *path
= p
;
3991 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
3995 void exec_runtime_destroy(ExecRuntime
*rt
) {
4001 /* If there are multiple users of this, let's leave the stuff around */
4006 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4008 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4010 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4017 if (rt
->var_tmp_dir
) {
4018 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4020 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4022 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4023 free(rt
->var_tmp_dir
);
4026 rt
->var_tmp_dir
= NULL
;
4029 safe_close_pair(rt
->netns_storage_socket
);
4032 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4033 [EXEC_INPUT_NULL
] = "null",
4034 [EXEC_INPUT_TTY
] = "tty",
4035 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4036 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4037 [EXEC_INPUT_SOCKET
] = "socket",
4038 [EXEC_INPUT_NAMED_FD
] = "fd",
4041 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4043 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4044 [EXEC_OUTPUT_INHERIT
] = "inherit",
4045 [EXEC_OUTPUT_NULL
] = "null",
4046 [EXEC_OUTPUT_TTY
] = "tty",
4047 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4048 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4049 [EXEC_OUTPUT_KMSG
] = "kmsg",
4050 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4051 [EXEC_OUTPUT_JOURNAL
] = "journal",
4052 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4053 [EXEC_OUTPUT_SOCKET
] = "socket",
4054 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4057 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4059 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4060 [EXEC_UTMP_INIT
] = "init",
4061 [EXEC_UTMP_LOGIN
] = "login",
4062 [EXEC_UTMP_USER
] = "user",
4065 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);