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
) {
821 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
822 * (i.e. are "/" or "/bin/nologin"). */
825 r
= get_user_creds_clean(&name
, uid
, gid
, home
, shell
);
833 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
843 r
= get_group_creds(&name
, gid
);
851 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
852 const char *group
, gid_t gid
,
853 gid_t
**supplementary_gids
, int *ngids
) {
857 bool keep_groups
= false;
858 gid_t
*groups
= NULL
;
859 _cleanup_free_ gid_t
*l_gids
= NULL
;
864 * If user is given, then lookup GID and supplementary groups list.
865 * We avoid NSS lookups for gid=0. Also we have to initialize groups
866 * here and as early as possible so we keep the list of supplementary
867 * groups of the caller.
869 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
870 /* First step, initialize groups from /etc/groups */
871 if (initgroups(user
, gid
) < 0)
877 if (!c
->supplementary_groups
)
881 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
882 * be positive, otherwise fail.
885 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
886 if (ngroups_max
<= 0) {
890 return -EOPNOTSUPP
; /* For all other values */
893 l_gids
= new(gid_t
, ngroups_max
);
899 * Lookup the list of groups that the user belongs to, we
900 * avoid NSS lookups here too for gid=0.
903 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
908 STRV_FOREACH(i
, c
->supplementary_groups
) {
911 if (k
>= ngroups_max
)
915 r
= get_group_creds(&g
, l_gids
+k
);
923 * Sets ngids to zero to drop all supplementary groups, happens
924 * when we are under root and SupplementaryGroups= is empty.
931 /* Otherwise get the final list of supplementary groups */
932 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
936 *supplementary_gids
= groups
;
944 static int enforce_groups(const ExecContext
*context
, gid_t gid
,
945 gid_t
*supplementary_gids
, int ngids
) {
950 /* Handle SupplementaryGroups= even if it is empty */
951 if (context
->supplementary_groups
) {
952 r
= maybe_setgroups(ngids
, supplementary_gids
);
957 if (gid_is_valid(gid
)) {
958 /* Then set our gids */
959 if (setresgid(gid
, gid
, gid
) < 0)
966 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
969 if (!uid_is_valid(uid
))
972 /* Sets (but doesn't look up) the uid and make sure we keep the
973 * capabilities while doing so. */
975 if (context
->capability_ambient_set
!= 0) {
977 /* First step: If we need to keep capabilities but
978 * drop privileges we need to make sure we keep our
979 * caps, while we drop privileges. */
981 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
983 if (prctl(PR_GET_SECUREBITS
) != sb
)
984 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
989 /* Second step: actually set the uids */
990 if (setresuid(uid
, uid
, uid
) < 0)
993 /* At this point we should have all necessary capabilities but
994 are otherwise a normal user. However, the caps might got
995 corrupted due to the setresuid() so we need clean them up
996 later. This is done outside of this call. */
1003 static int null_conv(
1005 const struct pam_message
**msg
,
1006 struct pam_response
**resp
,
1007 void *appdata_ptr
) {
1009 /* We don't support conversations */
1011 return PAM_CONV_ERR
;
1016 static int setup_pam(
1023 int fds
[], unsigned n_fds
) {
1027 static const struct pam_conv conv
= {
1032 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1033 pam_handle_t
*handle
= NULL
;
1035 int pam_code
= PAM_SUCCESS
, r
;
1036 char **nv
, **e
= NULL
;
1037 bool close_session
= false;
1038 pid_t pam_pid
= 0, parent_pid
;
1045 /* We set up PAM in the parent process, then fork. The child
1046 * will then stay around until killed via PR_GET_PDEATHSIG or
1047 * systemd via the cgroup logic. It will then remove the PAM
1048 * session again. The parent process will exec() the actual
1049 * daemon. We do things this way to ensure that the main PID
1050 * of the daemon is the one we initially fork()ed. */
1052 r
= barrier_create(&barrier
);
1056 if (log_get_max_level() < LOG_DEBUG
)
1057 flags
|= PAM_SILENT
;
1059 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1060 if (pam_code
!= PAM_SUCCESS
) {
1066 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1067 if (pam_code
!= PAM_SUCCESS
)
1071 STRV_FOREACH(nv
, *env
) {
1072 pam_code
= pam_putenv(handle
, *nv
);
1073 if (pam_code
!= PAM_SUCCESS
)
1077 pam_code
= pam_acct_mgmt(handle
, flags
);
1078 if (pam_code
!= PAM_SUCCESS
)
1081 pam_code
= pam_open_session(handle
, flags
);
1082 if (pam_code
!= PAM_SUCCESS
)
1085 close_session
= true;
1087 e
= pam_getenvlist(handle
);
1089 pam_code
= PAM_BUF_ERR
;
1093 /* Block SIGTERM, so that we know that it won't get lost in
1096 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1098 parent_pid
= getpid();
1107 int sig
, ret
= EXIT_PAM
;
1109 /* The child's job is to reset the PAM session on
1111 barrier_set_role(&barrier
, BARRIER_CHILD
);
1113 /* This string must fit in 10 chars (i.e. the length
1114 * of "/sbin/init"), to look pretty in /bin/ps */
1115 rename_process("(sd-pam)");
1117 /* Make sure we don't keep open the passed fds in this
1118 child. We assume that otherwise only those fds are
1119 open here that have been opened by PAM. */
1120 close_many(fds
, n_fds
);
1122 /* Drop privileges - we don't need any to pam_close_session
1123 * and this will make PR_SET_PDEATHSIG work in most cases.
1124 * If this fails, ignore the error - but expect sd-pam threads
1125 * to fail to exit normally */
1127 r
= maybe_setgroups(0, NULL
);
1129 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1130 if (setresgid(gid
, gid
, gid
) < 0)
1131 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1132 if (setresuid(uid
, uid
, uid
) < 0)
1133 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1135 (void) ignore_signals(SIGPIPE
, -1);
1137 /* Wait until our parent died. This will only work if
1138 * the above setresuid() succeeds, otherwise the kernel
1139 * will not allow unprivileged parents kill their privileged
1140 * children this way. We rely on the control groups kill logic
1141 * to do the rest for us. */
1142 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1145 /* Tell the parent that our setup is done. This is especially
1146 * important regarding dropping privileges. Otherwise, unit
1147 * setup might race against our setresuid(2) call.
1149 * If the parent aborted, we'll detect this below, hence ignore
1150 * return failure here. */
1151 (void) barrier_place(&barrier
);
1153 /* Check if our parent process might already have died? */
1154 if (getppid() == parent_pid
) {
1157 assert_se(sigemptyset(&ss
) >= 0);
1158 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1161 if (sigwait(&ss
, &sig
) < 0) {
1168 assert(sig
== SIGTERM
);
1173 /* If our parent died we'll end the session */
1174 if (getppid() != parent_pid
) {
1175 pam_code
= pam_close_session(handle
, flags
);
1176 if (pam_code
!= PAM_SUCCESS
)
1183 pam_end(handle
, pam_code
| flags
);
1187 barrier_set_role(&barrier
, BARRIER_PARENT
);
1189 /* If the child was forked off successfully it will do all the
1190 * cleanups, so forget about the handle here. */
1193 /* Unblock SIGTERM again in the parent */
1194 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1196 /* We close the log explicitly here, since the PAM modules
1197 * might have opened it, but we don't want this fd around. */
1200 /* Synchronously wait for the child to initialize. We don't care for
1201 * errors as we cannot recover. However, warn loudly if it happens. */
1202 if (!barrier_place_and_sync(&barrier
))
1203 log_error("PAM initialization failed");
1211 if (pam_code
!= PAM_SUCCESS
) {
1212 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1213 r
= -EPERM
; /* PAM errors do not map to errno */
1215 log_error_errno(r
, "PAM failed: %m");
1219 pam_code
= pam_close_session(handle
, flags
);
1221 pam_end(handle
, pam_code
| flags
);
1233 static void rename_process_from_path(const char *path
) {
1234 char process_name
[11];
1238 /* This resulting string must fit in 10 chars (i.e. the length
1239 * of "/sbin/init") to look pretty in /bin/ps */
1243 rename_process("(...)");
1249 /* The end of the process name is usually more
1250 * interesting, since the first bit might just be
1256 process_name
[0] = '(';
1257 memcpy(process_name
+1, p
, l
);
1258 process_name
[1+l
] = ')';
1259 process_name
[1+l
+1] = 0;
1261 rename_process(process_name
);
1264 static bool context_has_address_families(const ExecContext
*c
) {
1267 return c
->address_families_whitelist
||
1268 !set_isempty(c
->address_families
);
1271 static bool context_has_syscall_filters(const ExecContext
*c
) {
1274 return c
->syscall_whitelist
||
1275 !set_isempty(c
->syscall_filter
);
1278 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1281 if (c
->no_new_privileges
)
1284 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1287 /* We need NNP if we have any form of seccomp and are unprivileged */
1288 return context_has_address_families(c
) ||
1289 c
->memory_deny_write_execute
||
1290 c
->restrict_realtime
||
1291 exec_context_restrict_namespaces_set(c
) ||
1292 c
->protect_kernel_tunables
||
1293 c
->protect_kernel_modules
||
1294 c
->private_devices
||
1295 context_has_syscall_filters(c
) ||
1296 !set_isempty(c
->syscall_archs
);
1301 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1303 if (is_seccomp_available())
1307 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1312 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
) {
1313 uint32_t negative_action
, default_action
, action
;
1318 if (!context_has_syscall_filters(c
))
1321 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1324 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1326 if (c
->syscall_whitelist
) {
1327 default_action
= negative_action
;
1328 action
= SCMP_ACT_ALLOW
;
1330 default_action
= SCMP_ACT_ALLOW
;
1331 action
= negative_action
;
1334 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
);
1337 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1341 if (set_isempty(c
->syscall_archs
))
1344 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1347 return seccomp_restrict_archs(c
->syscall_archs
);
1350 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1354 if (!context_has_address_families(c
))
1357 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1360 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1363 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1367 if (!c
->memory_deny_write_execute
)
1370 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1373 return seccomp_memory_deny_write_execute();
1376 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1380 if (!c
->restrict_realtime
)
1383 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1386 return seccomp_restrict_realtime();
1389 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1393 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1394 * let's protect even those systems where this is left on in the kernel. */
1396 if (!c
->protect_kernel_tunables
)
1399 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1402 return seccomp_protect_sysctl();
1405 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1409 /* Turn off module syscalls on ProtectKernelModules=yes */
1411 if (!c
->protect_kernel_modules
)
1414 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1417 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
));
1420 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1424 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1426 if (!c
->private_devices
)
1429 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1432 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
));
1435 static int apply_restrict_namespaces(Unit
*u
, const ExecContext
*c
) {
1439 if (!exec_context_restrict_namespaces_set(c
))
1442 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1445 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1450 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1453 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1454 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1456 if (idle_pipe
[0] >= 0) {
1459 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1461 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1464 /* Signal systemd that we are bored and want to continue. */
1465 n
= write(idle_pipe
[3], "x", 1);
1467 /* Wait for systemd to react to the signal above. */
1468 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1471 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1475 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1478 static int build_environment(
1480 const ExecContext
*c
,
1481 const ExecParameters
*p
,
1484 const char *username
,
1486 dev_t journal_stream_dev
,
1487 ino_t journal_stream_ino
,
1490 _cleanup_strv_free_
char **our_env
= NULL
;
1498 our_env
= new0(char*, 14);
1503 _cleanup_free_
char *joined
= NULL
;
1505 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid()) < 0)
1507 our_env
[n_env
++] = x
;
1509 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1511 our_env
[n_env
++] = x
;
1513 joined
= strv_join(p
->fd_names
, ":");
1517 x
= strjoin("LISTEN_FDNAMES=", joined
);
1520 our_env
[n_env
++] = x
;
1523 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1524 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid()) < 0)
1526 our_env
[n_env
++] = x
;
1528 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1530 our_env
[n_env
++] = x
;
1533 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1534 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1535 * check the database directly. */
1536 if (unit_has_name(u
, SPECIAL_DBUS_SERVICE
)) {
1537 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1540 our_env
[n_env
++] = x
;
1544 x
= strappend("HOME=", home
);
1547 our_env
[n_env
++] = x
;
1551 x
= strappend("LOGNAME=", username
);
1554 our_env
[n_env
++] = x
;
1556 x
= strappend("USER=", username
);
1559 our_env
[n_env
++] = x
;
1563 x
= strappend("SHELL=", shell
);
1566 our_env
[n_env
++] = x
;
1569 if (!sd_id128_is_null(u
->invocation_id
)) {
1570 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1573 our_env
[n_env
++] = x
;
1576 if (exec_context_needs_term(c
)) {
1577 const char *tty_path
, *term
= NULL
;
1579 tty_path
= exec_context_tty_path(c
);
1581 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1582 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1583 * passes to PID 1 ends up all the way in the console login shown. */
1585 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1586 term
= getenv("TERM");
1588 term
= default_term_for_tty(tty_path
);
1590 x
= strappend("TERM=", term
);
1593 our_env
[n_env
++] = x
;
1596 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1597 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1600 our_env
[n_env
++] = x
;
1603 our_env
[n_env
++] = NULL
;
1604 assert(n_env
<= 12);
1612 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1613 _cleanup_strv_free_
char **pass_env
= NULL
;
1614 size_t n_env
= 0, n_bufsize
= 0;
1617 STRV_FOREACH(i
, c
->pass_environment
) {
1618 _cleanup_free_
char *x
= NULL
;
1624 x
= strjoin(*i
, "=", v
);
1627 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1629 pass_env
[n_env
++] = x
;
1630 pass_env
[n_env
] = NULL
;
1640 static bool exec_needs_mount_namespace(
1641 const ExecContext
*context
,
1642 const ExecParameters
*params
,
1643 ExecRuntime
*runtime
) {
1648 if (context
->root_image
)
1651 if (!strv_isempty(context
->read_write_paths
) ||
1652 !strv_isempty(context
->read_only_paths
) ||
1653 !strv_isempty(context
->inaccessible_paths
))
1656 if (context
->n_bind_mounts
> 0)
1659 if (context
->mount_flags
!= 0)
1662 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1665 if (context
->private_devices
||
1666 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1667 context
->protect_home
!= PROTECT_HOME_NO
||
1668 context
->protect_kernel_tunables
||
1669 context
->protect_kernel_modules
||
1670 context
->protect_control_groups
)
1673 if (context
->mount_apivfs
)
1679 static int setup_private_users(uid_t uid
, gid_t gid
) {
1680 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1681 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1682 _cleanup_close_
int unshare_ready_fd
= -1;
1683 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1689 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1690 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1691 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1692 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1693 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1694 * continues execution normally. */
1696 if (uid
!= 0 && uid_is_valid(uid
)) {
1697 r
= asprintf(&uid_map
,
1698 "0 0 1\n" /* Map root → root */
1699 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1704 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1709 if (gid
!= 0 && gid_is_valid(gid
)) {
1710 r
= asprintf(&gid_map
,
1711 "0 0 1\n" /* Map root → root */
1712 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1717 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1722 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1724 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1725 if (unshare_ready_fd
< 0)
1728 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1730 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1738 _cleanup_close_
int fd
= -1;
1742 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1743 * here, after the parent opened its own user namespace. */
1746 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1748 /* Wait until the parent unshared the user namespace */
1749 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1754 /* Disable the setgroups() system call in the child user namespace, for good. */
1755 a
= procfs_file_alloca(ppid
, "setgroups");
1756 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1758 if (errno
!= ENOENT
) {
1763 /* If the file is missing the kernel is too old, let's continue anyway. */
1765 if (write(fd
, "deny\n", 5) < 0) {
1770 fd
= safe_close(fd
);
1773 /* First write the GID map */
1774 a
= procfs_file_alloca(ppid
, "gid_map");
1775 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1780 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1784 fd
= safe_close(fd
);
1786 /* The write the UID map */
1787 a
= procfs_file_alloca(ppid
, "uid_map");
1788 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1793 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1798 _exit(EXIT_SUCCESS
);
1801 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1802 _exit(EXIT_FAILURE
);
1805 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1807 if (unshare(CLONE_NEWUSER
) < 0)
1810 /* Let the child know that the namespace is ready now */
1811 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1814 /* Try to read an error code from the child */
1815 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1818 if (n
== sizeof(r
)) { /* an error code was sent to us */
1823 if (n
!= 0) /* on success we should have read 0 bytes */
1826 r
= wait_for_terminate(pid
, &si
);
1831 /* If something strange happened with the child, let's consider this fatal, too */
1832 if (si
.si_code
!= CLD_EXITED
|| si
.si_status
!= 0)
1838 static int setup_runtime_directory(
1839 const ExecContext
*context
,
1840 const ExecParameters
*params
,
1850 STRV_FOREACH(rt
, context
->runtime_directory
) {
1851 _cleanup_free_
char *p
;
1853 p
= strjoin(params
->runtime_prefix
, "/", *rt
);
1857 r
= mkdir_p_label(p
, context
->runtime_directory_mode
);
1861 r
= chmod_and_chown(p
, context
->runtime_directory_mode
, uid
, gid
);
1869 static int setup_smack(
1870 const ExecContext
*context
,
1871 const ExecCommand
*command
) {
1879 if (!mac_smack_use())
1882 if (context
->smack_process_label
) {
1883 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
1887 #ifdef SMACK_DEFAULT_PROCESS_LABEL
1889 _cleanup_free_
char *exec_label
= NULL
;
1891 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
1892 if (r
< 0 && r
!= -ENODATA
&& r
!= -EOPNOTSUPP
)
1895 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
1905 static int compile_read_write_paths(
1906 const ExecContext
*context
,
1907 const ExecParameters
*params
,
1910 _cleanup_strv_free_
char **l
= NULL
;
1913 /* Compile the list of writable paths. This is the combination of
1914 * the explicitly configured paths, plus all runtime directories. */
1916 if (strv_isempty(context
->read_write_paths
) &&
1917 strv_isempty(context
->runtime_directory
)) {
1918 *ret
= NULL
; /* NOP if neither is set */
1922 l
= strv_copy(context
->read_write_paths
);
1926 STRV_FOREACH(rt
, context
->runtime_directory
) {
1929 s
= strjoin(params
->runtime_prefix
, "/", *rt
);
1933 if (strv_consume(&l
, s
) < 0)
1943 static int apply_mount_namespace(
1945 ExecCommand
*command
,
1946 const ExecContext
*context
,
1947 const ExecParameters
*params
,
1948 ExecRuntime
*runtime
) {
1950 _cleanup_strv_free_
char **rw
= NULL
;
1951 char *tmp
= NULL
, *var
= NULL
;
1952 const char *root_dir
= NULL
, *root_image
= NULL
;
1953 NameSpaceInfo ns_info
= {
1954 .ignore_protect_paths
= false,
1955 .private_dev
= context
->private_devices
,
1956 .protect_control_groups
= context
->protect_control_groups
,
1957 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
1958 .protect_kernel_modules
= context
->protect_kernel_modules
,
1959 .mount_apivfs
= context
->mount_apivfs
,
1961 bool apply_restrictions
;
1966 /* The runtime struct only contains the parent of the private /tmp,
1967 * which is non-accessible to world users. Inside of it there's a /tmp
1968 * that is sticky, and that's the one we want to use here. */
1970 if (context
->private_tmp
&& runtime
) {
1971 if (runtime
->tmp_dir
)
1972 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
1973 if (runtime
->var_tmp_dir
)
1974 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
1977 r
= compile_read_write_paths(context
, params
, &rw
);
1981 if (params
->flags
& EXEC_APPLY_CHROOT
) {
1982 root_image
= context
->root_image
;
1985 root_dir
= context
->root_directory
;
1989 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
1990 * sandbox info, otherwise enforce it, don't ignore protected paths and
1991 * fail if we are enable to apply the sandbox inside the mount namespace.
1993 if (!context
->dynamic_user
&& root_dir
)
1994 ns_info
.ignore_protect_paths
= true;
1996 apply_restrictions
= (params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
;
1998 r
= setup_namespace(root_dir
, root_image
,
2000 apply_restrictions
? context
->read_only_paths
: NULL
,
2001 apply_restrictions
? context
->inaccessible_paths
: NULL
,
2002 context
->bind_mounts
,
2003 context
->n_bind_mounts
,
2006 apply_restrictions
? context
->protect_home
: PROTECT_HOME_NO
,
2007 apply_restrictions
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2008 context
->mount_flags
,
2009 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2011 /* If we couldn't set up the namespace this is probably due to a
2012 * missing capability. In this case, silently proceeed. */
2013 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2015 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2023 static int apply_working_directory(
2024 const ExecContext
*context
,
2025 const ExecParameters
*params
,
2027 const bool needs_mount_ns
,
2033 assert(exit_status
);
2035 if (context
->working_directory_home
) {
2038 *exit_status
= EXIT_CHDIR
;
2044 } else if (context
->working_directory
)
2045 wd
= context
->working_directory
;
2049 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2050 if (!needs_mount_ns
&& context
->root_directory
)
2051 if (chroot(context
->root_directory
) < 0) {
2052 *exit_status
= EXIT_CHROOT
;
2058 d
= prefix_roota(context
->root_directory
, wd
);
2060 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2061 *exit_status
= EXIT_CHDIR
;
2068 static int setup_keyring(Unit
*u
, const ExecParameters
*p
, uid_t uid
, gid_t gid
) {
2069 key_serial_t keyring
;
2074 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2075 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2076 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2077 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2078 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2079 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2081 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2084 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2085 if (keyring
== -1) {
2086 if (errno
== ENOSYS
)
2087 log_debug_errno(errno
, "Kernel keyring not supported, ignoring.");
2088 else if (IN_SET(errno
, EACCES
, EPERM
))
2089 log_debug_errno(errno
, "Kernel keyring access prohibited, ignoring.");
2090 else if (errno
== EDQUOT
)
2091 log_debug_errno(errno
, "Out of kernel keyrings to allocate, ignoring.");
2093 return log_error_errno(errno
, "Setting up kernel keyring failed: %m");
2098 /* Populate they keyring with the invocation ID by default. */
2099 if (!sd_id128_is_null(u
->invocation_id
)) {
2102 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2104 log_debug_errno(errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2106 if (keyctl(KEYCTL_SETPERM
, key
,
2107 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2108 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2109 return log_error_errno(errno
, "Failed to restrict invocation ID permission: %m");
2113 /* And now, make the keyring owned by the service's user */
2114 if (uid_is_valid(uid
) || gid_is_valid(gid
))
2115 if (keyctl(KEYCTL_CHOWN
, keyring
, uid
, gid
, 0) < 0)
2116 return log_error_errno(errno
, "Failed to change ownership of session keyring: %m");
2121 static void append_socket_pair(int *array
, unsigned *n
, int pair
[2]) {
2129 array
[(*n
)++] = pair
[0];
2131 array
[(*n
)++] = pair
[1];
2134 static int close_remaining_fds(
2135 const ExecParameters
*params
,
2136 ExecRuntime
*runtime
,
2137 DynamicCreds
*dcreds
,
2140 int *fds
, unsigned n_fds
) {
2142 unsigned n_dont_close
= 0;
2143 int dont_close
[n_fds
+ 12];
2147 if (params
->stdin_fd
>= 0)
2148 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2149 if (params
->stdout_fd
>= 0)
2150 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2151 if (params
->stderr_fd
>= 0)
2152 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2155 dont_close
[n_dont_close
++] = socket_fd
;
2157 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2158 n_dont_close
+= n_fds
;
2162 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2166 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2168 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2171 if (user_lookup_fd
>= 0)
2172 dont_close
[n_dont_close
++] = user_lookup_fd
;
2174 return close_all_fds(dont_close
, n_dont_close
);
2177 static int send_user_lookup(
2185 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2186 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2189 if (user_lookup_fd
< 0)
2192 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2195 if (writev(user_lookup_fd
,
2197 { .iov_base
= &uid
, .iov_len
= sizeof(uid
) },
2198 { .iov_base
= &gid
, .iov_len
= sizeof(gid
) },
2199 { .iov_base
= unit
->id
, .iov_len
= strlen(unit
->id
) }}, 3) < 0)
2205 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2212 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2217 if (!c
->working_directory_home
)
2221 /* Hardcode /root as home directory for UID 0 */
2226 r
= get_home_dir(buf
);
2234 static int exec_child(
2236 ExecCommand
*command
,
2237 const ExecContext
*context
,
2238 const ExecParameters
*params
,
2239 ExecRuntime
*runtime
,
2240 DynamicCreds
*dcreds
,
2244 int *fds
, unsigned n_fds
,
2248 char **error_message
) {
2250 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2251 _cleanup_free_
char *mac_selinux_context_net
= NULL
, *home_buffer
= NULL
;
2252 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2253 const char *username
= NULL
, *groupname
= NULL
;
2254 const char *home
= NULL
, *shell
= NULL
;
2255 dev_t journal_stream_dev
= 0;
2256 ino_t journal_stream_ino
= 0;
2257 bool needs_mount_namespace
;
2258 uid_t uid
= UID_INVALID
;
2259 gid_t gid
= GID_INVALID
;
2260 int i
, r
, ngids
= 0;
2266 assert(exit_status
);
2267 assert(error_message
);
2268 /* We don't always set error_message, hence it must be initialized */
2269 assert(*error_message
== NULL
);
2271 rename_process_from_path(command
->path
);
2273 /* We reset exactly these signals, since they are the
2274 * only ones we set to SIG_IGN in the main daemon. All
2275 * others we leave untouched because we set them to
2276 * SIG_DFL or a valid handler initially, both of which
2277 * will be demoted to SIG_DFL. */
2278 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2279 SIGNALS_IGNORE
, -1);
2281 if (context
->ignore_sigpipe
)
2282 (void) ignore_signals(SIGPIPE
, -1);
2284 r
= reset_signal_mask();
2286 *exit_status
= EXIT_SIGNAL_MASK
;
2287 *error_message
= strdup("Failed to reset signal mask");
2288 /* If strdup fails, here and below, we will just print the generic error message. */
2292 if (params
->idle_pipe
)
2293 do_idle_pipe_dance(params
->idle_pipe
);
2295 /* Close sockets very early to make sure we don't
2296 * block init reexecution because it cannot bind its
2301 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2303 *exit_status
= EXIT_FDS
;
2304 *error_message
= strdup("Failed to close remaining fds");
2308 if (!context
->same_pgrp
)
2310 *exit_status
= EXIT_SETSID
;
2314 exec_context_tty_reset(context
, params
);
2316 if (unit_shall_confirm_spawn(unit
)) {
2317 const char *vc
= params
->confirm_spawn
;
2318 _cleanup_free_
char *cmdline
= NULL
;
2320 cmdline
= exec_command_line(argv
);
2322 *exit_status
= EXIT_CONFIRM
;
2326 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2327 if (r
!= CONFIRM_EXECUTE
) {
2328 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2329 *exit_status
= EXIT_SUCCESS
;
2332 *exit_status
= EXIT_CONFIRM
;
2333 *error_message
= strdup("Execution cancelled");
2338 if (context
->dynamic_user
&& dcreds
) {
2340 /* Make sure we bypass our own NSS module for any NSS checks */
2341 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2342 *exit_status
= EXIT_USER
;
2343 *error_message
= strdup("Failed to update environment");
2347 r
= dynamic_creds_realize(dcreds
, &uid
, &gid
);
2349 *exit_status
= EXIT_USER
;
2350 *error_message
= strdup("Failed to update dynamic user credentials");
2354 if (!uid_is_valid(uid
)) {
2355 *exit_status
= EXIT_USER
;
2356 (void) asprintf(error_message
, "UID validation failed for \""UID_FMT
"\"", uid
);
2357 /* If asprintf fails, here and below, we will just print the generic error message. */
2361 if (!gid_is_valid(gid
)) {
2362 *exit_status
= EXIT_USER
;
2363 (void) asprintf(error_message
, "GID validation failed for \""GID_FMT
"\"", gid
);
2368 username
= dcreds
->user
->name
;
2371 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2373 *exit_status
= EXIT_USER
;
2374 *error_message
= strdup("Failed to determine user credentials");
2378 r
= get_fixed_group(context
, &groupname
, &gid
);
2380 *exit_status
= EXIT_GROUP
;
2381 *error_message
= strdup("Failed to determine group credentials");
2386 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2387 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2388 &supplementary_gids
, &ngids
);
2390 *exit_status
= EXIT_GROUP
;
2391 *error_message
= strdup("Failed to determine supplementary groups");
2395 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2397 *exit_status
= EXIT_USER
;
2398 *error_message
= strdup("Failed to send user credentials to PID1");
2402 user_lookup_fd
= safe_close(user_lookup_fd
);
2404 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2406 *exit_status
= EXIT_CHDIR
;
2407 *error_message
= strdup("Failed to determine $HOME for user");
2411 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2412 * must sure to drop O_NONBLOCK */
2414 (void) fd_nonblock(socket_fd
, false);
2416 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2418 *exit_status
= EXIT_STDIN
;
2419 *error_message
= strdup("Failed to set up stdin");
2423 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2425 *exit_status
= EXIT_STDOUT
;
2426 *error_message
= strdup("Failed to set up stdout");
2430 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2432 *exit_status
= EXIT_STDERR
;
2433 *error_message
= strdup("Failed to set up stderr");
2437 if (params
->cgroup_path
) {
2438 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2440 *exit_status
= EXIT_CGROUP
;
2441 (void) asprintf(error_message
, "Failed to attach to cgroup %s", params
->cgroup_path
);
2446 if (context
->oom_score_adjust_set
) {
2447 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2449 /* When we can't make this change due to EPERM, then
2450 * let's silently skip over it. User namespaces
2451 * prohibit write access to this file, and we
2452 * shouldn't trip up over that. */
2454 sprintf(t
, "%i", context
->oom_score_adjust
);
2455 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2456 if (r
== -EPERM
|| r
== -EACCES
) {
2458 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2461 *exit_status
= EXIT_OOM_ADJUST
;
2462 *error_message
= strdup("Failed to write /proc/self/oom_score_adj");
2467 if (context
->nice_set
)
2468 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2469 *exit_status
= EXIT_NICE
;
2473 if (context
->cpu_sched_set
) {
2474 struct sched_param param
= {
2475 .sched_priority
= context
->cpu_sched_priority
,
2478 r
= sched_setscheduler(0,
2479 context
->cpu_sched_policy
|
2480 (context
->cpu_sched_reset_on_fork
?
2481 SCHED_RESET_ON_FORK
: 0),
2484 *exit_status
= EXIT_SETSCHEDULER
;
2489 if (context
->cpuset
)
2490 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2491 *exit_status
= EXIT_CPUAFFINITY
;
2495 if (context
->ioprio_set
)
2496 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2497 *exit_status
= EXIT_IOPRIO
;
2501 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2502 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2503 *exit_status
= EXIT_TIMERSLACK
;
2507 if (context
->personality
!= PERSONALITY_INVALID
)
2508 if (personality(context
->personality
) < 0) {
2509 *exit_status
= EXIT_PERSONALITY
;
2513 if (context
->utmp_id
)
2514 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0),
2516 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
2517 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
2521 if (context
->user
) {
2522 r
= chown_terminal(STDIN_FILENO
, uid
);
2524 *exit_status
= EXIT_STDIN
;
2529 /* If delegation is enabled we'll pass ownership of the cgroup
2530 * (but only in systemd's own controller hierarchy!) to the
2531 * user of the new process. */
2532 if (params
->cgroup_path
&& context
->user
&& params
->cgroup_delegate
) {
2533 r
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
2535 *exit_status
= EXIT_CGROUP
;
2540 r
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
2542 *exit_status
= EXIT_CGROUP
;
2547 if (!strv_isempty(context
->runtime_directory
) && params
->runtime_prefix
) {
2548 r
= setup_runtime_directory(context
, params
, uid
, gid
);
2550 *exit_status
= EXIT_RUNTIME_DIRECTORY
;
2555 r
= build_environment(
2567 *exit_status
= EXIT_MEMORY
;
2571 r
= build_pass_environment(context
, &pass_env
);
2573 *exit_status
= EXIT_MEMORY
;
2577 accum_env
= strv_env_merge(5,
2578 params
->environment
,
2581 context
->environment
,
2585 *exit_status
= EXIT_MEMORY
;
2588 accum_env
= strv_env_clean(accum_env
);
2590 (void) umask(context
->umask
);
2592 r
= setup_keyring(unit
, params
, uid
, gid
);
2594 *exit_status
= EXIT_KEYRING
;
2598 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2599 if (context
->pam_name
&& username
) {
2600 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
2602 *exit_status
= EXIT_PAM
;
2608 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
2609 r
= setup_netns(runtime
->netns_storage_socket
);
2611 *exit_status
= EXIT_NETWORK
;
2616 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
2617 if (needs_mount_namespace
) {
2618 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
2620 *exit_status
= EXIT_NAMESPACE
;
2625 /* Apply just after mount namespace setup */
2626 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
2630 /* Drop groups as early as possbile */
2631 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2632 r
= enforce_groups(context
, gid
, supplementary_gids
, ngids
);
2634 *exit_status
= EXIT_GROUP
;
2640 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) &&
2641 mac_selinux_use() &&
2642 params
->selinux_context_net
&&
2644 !command
->privileged
) {
2646 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
2648 *exit_status
= EXIT_SELINUX_CONTEXT
;
2654 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && context
->private_users
) {
2655 r
= setup_private_users(uid
, gid
);
2657 *exit_status
= EXIT_USER
;
2662 /* We repeat the fd closing here, to make sure that
2663 * nothing is leaked from the PAM modules. Note that
2664 * we are more aggressive this time since socket_fd
2665 * and the netns fds we don't need anymore. The custom
2666 * endpoint fd was needed to upload the policy and can
2667 * now be closed as well. */
2668 r
= close_all_fds(fds
, n_fds
);
2670 r
= shift_fds(fds
, n_fds
);
2672 r
= flags_fds(fds
, n_fds
, context
->non_blocking
);
2674 *exit_status
= EXIT_FDS
;
2678 if ((params
->flags
& EXEC_APPLY_PERMISSIONS
) && !command
->privileged
) {
2680 int secure_bits
= context
->secure_bits
;
2682 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2684 if (!context
->rlimit
[i
])
2687 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
2689 *exit_status
= EXIT_LIMITS
;
2694 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
2695 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
2696 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
2697 *exit_status
= EXIT_LIMITS
;
2702 if (!cap_test_all(context
->capability_bounding_set
)) {
2703 r
= capability_bounding_set_drop(context
->capability_bounding_set
, false);
2705 *exit_status
= EXIT_CAPABILITIES
;
2706 *error_message
= strdup("Failed to drop capabilities");
2711 /* This is done before enforce_user, but ambient set
2712 * does not survive over setresuid() if keep_caps is not set. */
2713 if (context
->capability_ambient_set
!= 0) {
2714 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
2716 *exit_status
= EXIT_CAPABILITIES
;
2717 *error_message
= strdup("Failed to apply ambient capabilities (before UID change)");
2722 if (context
->user
) {
2723 r
= enforce_user(context
, uid
);
2725 *exit_status
= EXIT_USER
;
2726 (void) asprintf(error_message
, "Failed to change UID to "UID_FMT
, uid
);
2729 if (context
->capability_ambient_set
!= 0) {
2731 /* Fix the ambient capabilities after user change. */
2732 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
2734 *exit_status
= EXIT_CAPABILITIES
;
2735 *error_message
= strdup("Failed to apply ambient capabilities (after UID change)");
2739 /* If we were asked to change user and ambient capabilities
2740 * were requested, we had to add keep-caps to the securebits
2741 * so that we would maintain the inherited capability set
2742 * through the setresuid(). Make sure that the bit is added
2743 * also to the context secure_bits so that we don't try to
2744 * drop the bit away next. */
2746 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
2750 /* Apply the MAC contexts late, but before seccomp syscall filtering, as those should really be last to
2751 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
2752 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
2753 * are restricted. */
2756 if (mac_selinux_use()) {
2757 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
2760 r
= setexeccon(exec_context
);
2762 *exit_status
= EXIT_SELINUX_CONTEXT
;
2763 (void) asprintf(error_message
, "Failed to set SELinux context to %s", exec_context
);
2770 r
= setup_smack(context
, command
);
2772 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
2773 *error_message
= strdup("Failed to set SMACK process label");
2777 #ifdef HAVE_APPARMOR
2778 if (context
->apparmor_profile
&& mac_apparmor_use()) {
2779 r
= aa_change_onexec(context
->apparmor_profile
);
2780 if (r
< 0 && !context
->apparmor_profile_ignore
) {
2781 *exit_status
= EXIT_APPARMOR_PROFILE
;
2782 (void) asprintf(error_message
,
2783 "Failed to prepare AppArmor profile change to %s",
2784 context
->apparmor_profile
);
2790 /* PR_GET_SECUREBITS is not privileged, while
2791 * PR_SET_SECUREBITS is. So to suppress
2792 * potential EPERMs we'll try not to call
2793 * PR_SET_SECUREBITS unless necessary. */
2794 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
2795 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
2796 *exit_status
= EXIT_SECUREBITS
;
2797 *error_message
= strdup("Failed to set secure bits");
2801 if (context_has_no_new_privileges(context
))
2802 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
2803 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
2804 *error_message
= strdup("Failed to disable new privileges");
2809 r
= apply_address_families(unit
, context
);
2811 *exit_status
= EXIT_ADDRESS_FAMILIES
;
2812 *error_message
= strdup("Failed to restrict address families");
2816 r
= apply_memory_deny_write_execute(unit
, context
);
2818 *exit_status
= EXIT_SECCOMP
;
2819 *error_message
= strdup("Failed to disable writing to executable memory");
2823 r
= apply_restrict_realtime(unit
, context
);
2825 *exit_status
= EXIT_SECCOMP
;
2826 *error_message
= strdup("Failed to apply realtime restrictions");
2830 r
= apply_restrict_namespaces(unit
, context
);
2832 *exit_status
= EXIT_SECCOMP
;
2833 *error_message
= strdup("Failed to apply namespace restrictions");
2837 r
= apply_protect_sysctl(unit
, context
);
2839 *exit_status
= EXIT_SECCOMP
;
2840 *error_message
= strdup("Failed to apply sysctl restrictions");
2844 r
= apply_protect_kernel_modules(unit
, context
);
2846 *exit_status
= EXIT_SECCOMP
;
2847 *error_message
= strdup("Failed to apply module loading restrictions");
2851 r
= apply_private_devices(unit
, context
);
2853 *exit_status
= EXIT_SECCOMP
;
2854 *error_message
= strdup("Failed to set up private devices");
2858 r
= apply_syscall_archs(unit
, context
);
2860 *exit_status
= EXIT_SECCOMP
;
2861 *error_message
= strdup("Failed to apply syscall architecture restrictions");
2865 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
2866 * by the filter as little as possible. */
2867 r
= apply_syscall_filter(unit
, context
);
2869 *exit_status
= EXIT_SECCOMP
;
2870 *error_message
= strdup("Failed to apply syscall filters");
2876 final_argv
= replace_env_argv(argv
, accum_env
);
2878 *exit_status
= EXIT_MEMORY
;
2879 *error_message
= strdup("Failed to prepare process arguments");
2883 if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
2884 _cleanup_free_
char *line
;
2886 line
= exec_command_line(final_argv
);
2889 log_struct(LOG_DEBUG
,
2890 "EXECUTABLE=%s", command
->path
,
2891 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
2898 execve(command
->path
, final_argv
, accum_env
);
2899 *exit_status
= EXIT_EXEC
;
2903 int exec_spawn(Unit
*unit
,
2904 ExecCommand
*command
,
2905 const ExecContext
*context
,
2906 const ExecParameters
*params
,
2907 ExecRuntime
*runtime
,
2908 DynamicCreds
*dcreds
,
2911 _cleanup_strv_free_
char **files_env
= NULL
;
2912 int *fds
= NULL
; unsigned n_fds
= 0;
2913 _cleanup_free_
char *line
= NULL
;
2915 int named_iofds
[3] = { -1, -1, -1 };
2924 assert(params
->fds
|| params
->n_fds
<= 0);
2926 if (context
->std_input
== EXEC_INPUT_SOCKET
||
2927 context
->std_output
== EXEC_OUTPUT_SOCKET
||
2928 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
2930 if (params
->n_fds
!= 1) {
2931 log_unit_error(unit
, "Got more than one socket.");
2935 socket_fd
= params
->fds
[0];
2939 n_fds
= params
->n_fds
;
2942 r
= exec_context_named_iofds(unit
, context
, params
, named_iofds
);
2944 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
2946 r
= exec_context_load_environment(unit
, context
, &files_env
);
2948 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
2950 argv
= params
->argv
?: command
->argv
;
2951 line
= exec_command_line(argv
);
2955 log_struct(LOG_DEBUG
,
2956 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
2957 "EXECUTABLE=%s", command
->path
,
2962 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
2966 _cleanup_free_
char *error_message
= NULL
;
2968 r
= exec_child(unit
,
2979 unit
->manager
->user_lookup_fds
[1],
2985 log_struct_errno(LOG_ERR
, r
,
2986 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
2988 LOG_UNIT_MESSAGE(unit
, "%s: %m",
2990 "EXECUTABLE=%s", command
->path
,
2992 else if (r
== -ENOENT
&& command
->ignore
)
2993 log_struct_errno(LOG_INFO
, r
,
2994 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
2996 LOG_UNIT_MESSAGE(unit
, "Skipped spawning %s: %m",
2998 "EXECUTABLE=%s", command
->path
,
3001 log_struct_errno(LOG_ERR
, r
,
3002 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3004 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3005 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3007 "EXECUTABLE=%s", command
->path
,
3014 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3016 /* We add the new process to the cgroup both in the child (so
3017 * that we can be sure that no user code is ever executed
3018 * outside of the cgroup) and in the parent (so that we can be
3019 * sure that when we kill the cgroup the process will be
3021 if (params
->cgroup_path
)
3022 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3024 exec_status_start(&command
->exec_status
, pid
);
3030 void exec_context_init(ExecContext
*c
) {
3034 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3035 c
->cpu_sched_policy
= SCHED_OTHER
;
3036 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3037 c
->syslog_level_prefix
= true;
3038 c
->ignore_sigpipe
= true;
3039 c
->timer_slack_nsec
= NSEC_INFINITY
;
3040 c
->personality
= PERSONALITY_INVALID
;
3041 c
->runtime_directory_mode
= 0755;
3042 c
->capability_bounding_set
= CAP_ALL
;
3043 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
3046 void exec_context_done(ExecContext
*c
) {
3051 c
->environment
= strv_free(c
->environment
);
3052 c
->environment_files
= strv_free(c
->environment_files
);
3053 c
->pass_environment
= strv_free(c
->pass_environment
);
3055 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
3056 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
3058 for (l
= 0; l
< 3; l
++)
3059 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3061 c
->working_directory
= mfree(c
->working_directory
);
3062 c
->root_directory
= mfree(c
->root_directory
);
3063 c
->root_image
= mfree(c
->root_image
);
3064 c
->tty_path
= mfree(c
->tty_path
);
3065 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3066 c
->user
= mfree(c
->user
);
3067 c
->group
= mfree(c
->group
);
3069 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3071 c
->pam_name
= mfree(c
->pam_name
);
3073 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3074 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3075 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3077 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3080 CPU_FREE(c
->cpuset
);
3082 c
->utmp_id
= mfree(c
->utmp_id
);
3083 c
->selinux_context
= mfree(c
->selinux_context
);
3084 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3086 c
->syscall_filter
= set_free(c
->syscall_filter
);
3087 c
->syscall_archs
= set_free(c
->syscall_archs
);
3088 c
->address_families
= set_free(c
->address_families
);
3090 c
->runtime_directory
= strv_free(c
->runtime_directory
);
3093 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
3098 if (!runtime_prefix
)
3101 STRV_FOREACH(i
, c
->runtime_directory
) {
3102 _cleanup_free_
char *p
;
3104 p
= strjoin(runtime_prefix
, "/", *i
);
3108 /* We execute this synchronously, since we need to be
3109 * sure this is gone when we start the service
3111 (void) rm_rf(p
, REMOVE_ROOT
);
3117 void exec_command_done(ExecCommand
*c
) {
3120 c
->path
= mfree(c
->path
);
3122 c
->argv
= strv_free(c
->argv
);
3125 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3128 for (i
= 0; i
< n
; i
++)
3129 exec_command_done(c
+i
);
3132 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3136 LIST_REMOVE(command
, c
, i
);
3137 exec_command_done(i
);
3144 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3147 for (i
= 0; i
< n
; i
++)
3148 c
[i
] = exec_command_free_list(c
[i
]);
3151 typedef struct InvalidEnvInfo
{
3156 static void invalid_env(const char *p
, void *userdata
) {
3157 InvalidEnvInfo
*info
= userdata
;
3159 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3162 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3167 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3169 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3171 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3173 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3175 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3177 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3183 int exec_context_named_iofds(Unit
*unit
, const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3184 unsigned i
, targets
;
3185 const char* stdio_fdname
[3];
3190 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3191 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3192 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3194 for (i
= 0; i
< 3; i
++)
3195 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3197 for (i
= 0; i
< p
->n_fds
&& targets
> 0; i
++)
3198 if (named_iofds
[STDIN_FILENO
] < 0 &&
3199 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3200 stdio_fdname
[STDIN_FILENO
] &&
3201 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3203 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3206 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3207 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3208 stdio_fdname
[STDOUT_FILENO
] &&
3209 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3211 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3214 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3215 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3216 stdio_fdname
[STDERR_FILENO
] &&
3217 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3219 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3223 return targets
== 0 ? 0 : -ENOENT
;
3226 int exec_context_load_environment(Unit
*unit
, const ExecContext
*c
, char ***l
) {
3227 char **i
, **r
= NULL
;
3232 STRV_FOREACH(i
, c
->environment_files
) {
3235 bool ignore
= false;
3237 _cleanup_globfree_ glob_t pglob
= {};
3247 if (!path_is_absolute(fn
)) {
3255 /* Filename supports globbing, take all matching files */
3257 if (glob(fn
, 0, NULL
, &pglob
) != 0) {
3262 return errno
> 0 ? -errno
: -EINVAL
;
3264 count
= pglob
.gl_pathc
;
3272 for (n
= 0; n
< count
; n
++) {
3273 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3281 /* Log invalid environment variables with filename */
3283 InvalidEnvInfo info
= {
3285 .path
= pglob
.gl_pathv
[n
]
3288 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3296 m
= strv_env_merge(2, r
, p
);
3312 static bool tty_may_match_dev_console(const char *tty
) {
3313 _cleanup_free_
char *active
= NULL
;
3319 if (startswith(tty
, "/dev/"))
3322 /* trivial identity? */
3323 if (streq(tty
, "console"))
3326 console
= resolve_dev_console(&active
);
3327 /* if we could not resolve, assume it may */
3331 /* "tty0" means the active VC, so it may be the same sometimes */
3332 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
3335 bool exec_context_may_touch_console(ExecContext
*ec
) {
3337 return (ec
->tty_reset
||
3339 ec
->tty_vt_disallocate
||
3340 is_terminal_input(ec
->std_input
) ||
3341 is_terminal_output(ec
->std_output
) ||
3342 is_terminal_output(ec
->std_error
)) &&
3343 tty_may_match_dev_console(exec_context_tty_path(ec
));
3346 static void strv_fprintf(FILE *f
, char **l
) {
3352 fprintf(f
, " %s", *g
);
3355 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
3363 prefix
= strempty(prefix
);
3367 "%sWorkingDirectory: %s\n"
3368 "%sRootDirectory: %s\n"
3369 "%sNonBlocking: %s\n"
3370 "%sPrivateTmp: %s\n"
3371 "%sPrivateDevices: %s\n"
3372 "%sProtectKernelTunables: %s\n"
3373 "%sProtectKernelModules: %s\n"
3374 "%sProtectControlGroups: %s\n"
3375 "%sPrivateNetwork: %s\n"
3376 "%sPrivateUsers: %s\n"
3377 "%sProtectHome: %s\n"
3378 "%sProtectSystem: %s\n"
3379 "%sMountAPIVFS: %s\n"
3380 "%sIgnoreSIGPIPE: %s\n"
3381 "%sMemoryDenyWriteExecute: %s\n"
3382 "%sRestrictRealtime: %s\n",
3384 prefix
, c
->working_directory
? c
->working_directory
: "/",
3385 prefix
, c
->root_directory
? c
->root_directory
: "/",
3386 prefix
, yes_no(c
->non_blocking
),
3387 prefix
, yes_no(c
->private_tmp
),
3388 prefix
, yes_no(c
->private_devices
),
3389 prefix
, yes_no(c
->protect_kernel_tunables
),
3390 prefix
, yes_no(c
->protect_kernel_modules
),
3391 prefix
, yes_no(c
->protect_control_groups
),
3392 prefix
, yes_no(c
->private_network
),
3393 prefix
, yes_no(c
->private_users
),
3394 prefix
, protect_home_to_string(c
->protect_home
),
3395 prefix
, protect_system_to_string(c
->protect_system
),
3396 prefix
, yes_no(c
->mount_apivfs
),
3397 prefix
, yes_no(c
->ignore_sigpipe
),
3398 prefix
, yes_no(c
->memory_deny_write_execute
),
3399 prefix
, yes_no(c
->restrict_realtime
));
3402 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3404 STRV_FOREACH(e
, c
->environment
)
3405 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3407 STRV_FOREACH(e
, c
->environment_files
)
3408 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3410 STRV_FOREACH(e
, c
->pass_environment
)
3411 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3413 fprintf(f
, "%sRuntimeDirectoryMode: %04o\n", prefix
, c
->runtime_directory_mode
);
3415 STRV_FOREACH(d
, c
->runtime_directory
)
3416 fprintf(f
, "%sRuntimeDirectory: %s\n", prefix
, *d
);
3423 if (c
->oom_score_adjust_set
)
3425 "%sOOMScoreAdjust: %i\n",
3426 prefix
, c
->oom_score_adjust
);
3428 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3430 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3431 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3432 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3433 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3436 if (c
->ioprio_set
) {
3437 _cleanup_free_
char *class_str
= NULL
;
3439 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
3441 "%sIOSchedulingClass: %s\n"
3442 "%sIOPriority: %i\n",
3443 prefix
, strna(class_str
),
3444 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
3447 if (c
->cpu_sched_set
) {
3448 _cleanup_free_
char *policy_str
= NULL
;
3450 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
3452 "%sCPUSchedulingPolicy: %s\n"
3453 "%sCPUSchedulingPriority: %i\n"
3454 "%sCPUSchedulingResetOnFork: %s\n",
3455 prefix
, strna(policy_str
),
3456 prefix
, c
->cpu_sched_priority
,
3457 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
3461 fprintf(f
, "%sCPUAffinity:", prefix
);
3462 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
3463 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
3464 fprintf(f
, " %u", i
);
3468 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
3469 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
3472 "%sStandardInput: %s\n"
3473 "%sStandardOutput: %s\n"
3474 "%sStandardError: %s\n",
3475 prefix
, exec_input_to_string(c
->std_input
),
3476 prefix
, exec_output_to_string(c
->std_output
),
3477 prefix
, exec_output_to_string(c
->std_error
));
3483 "%sTTYVHangup: %s\n"
3484 "%sTTYVTDisallocate: %s\n",
3485 prefix
, c
->tty_path
,
3486 prefix
, yes_no(c
->tty_reset
),
3487 prefix
, yes_no(c
->tty_vhangup
),
3488 prefix
, yes_no(c
->tty_vt_disallocate
));
3490 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
3491 c
->std_output
== EXEC_OUTPUT_KMSG
||
3492 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
3493 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3494 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3495 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
3496 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
3497 c
->std_error
== EXEC_OUTPUT_KMSG
||
3498 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
3499 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
3500 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
3501 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
3503 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
3505 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
3506 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
3509 "%sSyslogFacility: %s\n"
3510 "%sSyslogLevel: %s\n",
3511 prefix
, strna(fac_str
),
3512 prefix
, strna(lvl_str
));
3516 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
3518 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
3519 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
3520 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
3521 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
3522 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
3523 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
3525 if (c
->capability_bounding_set
!= CAP_ALL
) {
3527 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
3529 for (l
= 0; l
<= cap_last_cap(); l
++)
3530 if (c
->capability_bounding_set
& (UINT64_C(1) << l
))
3531 fprintf(f
, " %s", strna(capability_to_name(l
)));
3536 if (c
->capability_ambient_set
!= 0) {
3538 fprintf(f
, "%sAmbientCapabilities:", prefix
);
3540 for (l
= 0; l
<= cap_last_cap(); l
++)
3541 if (c
->capability_ambient_set
& (UINT64_C(1) << l
))
3542 fprintf(f
, " %s", strna(capability_to_name(l
)));
3548 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
3550 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
3552 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
3554 if (strv_length(c
->supplementary_groups
) > 0) {
3555 fprintf(f
, "%sSupplementaryGroups:", prefix
);
3556 strv_fprintf(f
, c
->supplementary_groups
);
3561 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
3563 if (strv_length(c
->read_write_paths
) > 0) {
3564 fprintf(f
, "%sReadWritePaths:", prefix
);
3565 strv_fprintf(f
, c
->read_write_paths
);
3569 if (strv_length(c
->read_only_paths
) > 0) {
3570 fprintf(f
, "%sReadOnlyPaths:", prefix
);
3571 strv_fprintf(f
, c
->read_only_paths
);
3575 if (strv_length(c
->inaccessible_paths
) > 0) {
3576 fprintf(f
, "%sInaccessiblePaths:", prefix
);
3577 strv_fprintf(f
, c
->inaccessible_paths
);
3581 if (c
->n_bind_mounts
> 0)
3582 for (i
= 0; i
< c
->n_bind_mounts
; i
++) {
3583 fprintf(f
, "%s%s: %s:%s:%s\n", prefix
,
3584 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
3585 c
->bind_mounts
[i
].source
,
3586 c
->bind_mounts
[i
].destination
,
3587 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
3592 "%sUtmpIdentifier: %s\n",
3593 prefix
, c
->utmp_id
);
3595 if (c
->selinux_context
)
3597 "%sSELinuxContext: %s%s\n",
3598 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
3600 if (c
->personality
!= PERSONALITY_INVALID
)
3602 "%sPersonality: %s\n",
3603 prefix
, strna(personality_to_string(c
->personality
)));
3605 if (c
->syscall_filter
) {
3613 "%sSystemCallFilter: ",
3616 if (!c
->syscall_whitelist
)
3620 SET_FOREACH(id
, c
->syscall_filter
, j
) {
3621 _cleanup_free_
char *name
= NULL
;
3628 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
3629 fputs(strna(name
), f
);
3636 if (c
->syscall_archs
) {
3643 "%sSystemCallArchitectures:",
3647 SET_FOREACH(id
, c
->syscall_archs
, j
)
3648 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
3653 if (exec_context_restrict_namespaces_set(c
)) {
3654 _cleanup_free_
char *s
= NULL
;
3656 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
3658 fprintf(f
, "%sRestrictNamespaces: %s\n",
3662 if (c
->syscall_errno
> 0)
3664 "%sSystemCallErrorNumber: %s\n",
3665 prefix
, strna(errno_to_name(c
->syscall_errno
)));
3667 if (c
->apparmor_profile
)
3669 "%sAppArmorProfile: %s%s\n",
3670 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
3673 bool exec_context_maintains_privileges(ExecContext
*c
) {
3676 /* Returns true if the process forked off would run under
3677 * an unchanged UID or as root. */
3682 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
3688 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
3693 dual_timestamp_get(&s
->start_timestamp
);
3696 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
3699 if (s
->pid
&& s
->pid
!= pid
)
3703 dual_timestamp_get(&s
->exit_timestamp
);
3709 if (context
->utmp_id
)
3710 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
3712 exec_context_tty_reset(context
, NULL
);
3716 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
3717 char buf
[FORMAT_TIMESTAMP_MAX
];
3725 prefix
= strempty(prefix
);
3728 "%sPID: "PID_FMT
"\n",
3731 if (dual_timestamp_is_set(&s
->start_timestamp
))
3733 "%sStart Timestamp: %s\n",
3734 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
3736 if (dual_timestamp_is_set(&s
->exit_timestamp
))
3738 "%sExit Timestamp: %s\n"
3740 "%sExit Status: %i\n",
3741 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
3742 prefix
, sigchld_code_to_string(s
->code
),
3746 char *exec_command_line(char **argv
) {
3754 STRV_FOREACH(a
, argv
)
3762 STRV_FOREACH(a
, argv
) {
3769 if (strpbrk(*a
, WHITESPACE
)) {
3780 /* FIXME: this doesn't really handle arguments that have
3781 * spaces and ticks in them */
3786 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3787 _cleanup_free_
char *cmd
= NULL
;
3788 const char *prefix2
;
3793 prefix
= strempty(prefix
);
3794 prefix2
= strjoina(prefix
, "\t");
3796 cmd
= exec_command_line(c
->argv
);
3798 "%sCommand Line: %s\n",
3799 prefix
, cmd
? cmd
: strerror(ENOMEM
));
3801 exec_status_dump(&c
->exec_status
, f
, prefix2
);
3804 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
3807 prefix
= strempty(prefix
);
3809 LIST_FOREACH(command
, c
, c
)
3810 exec_command_dump(c
, f
, prefix
);
3813 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
3820 /* It's kind of important, that we keep the order here */
3821 LIST_FIND_TAIL(command
, *l
, end
);
3822 LIST_INSERT_AFTER(command
, *l
, end
, e
);
3827 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
3835 l
= strv_new_ap(path
, ap
);
3856 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
3857 _cleanup_strv_free_
char **l
= NULL
;
3865 l
= strv_new_ap(path
, ap
);
3871 r
= strv_extend_strv(&c
->argv
, l
, false);
3879 static int exec_runtime_allocate(ExecRuntime
**rt
) {
3884 *rt
= new0(ExecRuntime
, 1);
3889 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
3894 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
3904 if (!c
->private_network
&& !c
->private_tmp
)
3907 r
= exec_runtime_allocate(rt
);
3911 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
3912 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, (*rt
)->netns_storage_socket
) < 0)
3916 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
3917 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
3925 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
3927 assert(r
->n_ref
> 0);
3933 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
3938 assert(r
->n_ref
> 0);
3945 free(r
->var_tmp_dir
);
3946 safe_close_pair(r
->netns_storage_socket
);
3950 int exec_runtime_serialize(Unit
*u
, ExecRuntime
*rt
, FILE *f
, FDSet
*fds
) {
3959 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
3961 if (rt
->var_tmp_dir
)
3962 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
3964 if (rt
->netns_storage_socket
[0] >= 0) {
3967 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
3971 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
3974 if (rt
->netns_storage_socket
[1] >= 0) {
3977 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
3981 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
3987 int exec_runtime_deserialize_item(Unit
*u
, ExecRuntime
**rt
, const char *key
, const char *value
, FDSet
*fds
) {
3994 if (streq(key
, "tmp-dir")) {
3997 r
= exec_runtime_allocate(rt
);
4001 copy
= strdup(value
);
4005 free((*rt
)->tmp_dir
);
4006 (*rt
)->tmp_dir
= copy
;
4008 } else if (streq(key
, "var-tmp-dir")) {
4011 r
= exec_runtime_allocate(rt
);
4015 copy
= strdup(value
);
4019 free((*rt
)->var_tmp_dir
);
4020 (*rt
)->var_tmp_dir
= copy
;
4022 } else if (streq(key
, "netns-socket-0")) {
4025 r
= exec_runtime_allocate(rt
);
4029 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4030 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4032 safe_close((*rt
)->netns_storage_socket
[0]);
4033 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4035 } else if (streq(key
, "netns-socket-1")) {
4038 r
= exec_runtime_allocate(rt
);
4042 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
4043 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4045 safe_close((*rt
)->netns_storage_socket
[1]);
4046 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
4054 static void *remove_tmpdir_thread(void *p
) {
4055 _cleanup_free_
char *path
= p
;
4057 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4061 void exec_runtime_destroy(ExecRuntime
*rt
) {
4067 /* If there are multiple users of this, let's leave the stuff around */
4072 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4074 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4076 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4083 if (rt
->var_tmp_dir
) {
4084 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4086 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4088 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4089 free(rt
->var_tmp_dir
);
4092 rt
->var_tmp_dir
= NULL
;
4095 safe_close_pair(rt
->netns_storage_socket
);
4098 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4099 [EXEC_INPUT_NULL
] = "null",
4100 [EXEC_INPUT_TTY
] = "tty",
4101 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4102 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4103 [EXEC_INPUT_SOCKET
] = "socket",
4104 [EXEC_INPUT_NAMED_FD
] = "fd",
4107 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4109 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4110 [EXEC_OUTPUT_INHERIT
] = "inherit",
4111 [EXEC_OUTPUT_NULL
] = "null",
4112 [EXEC_OUTPUT_TTY
] = "tty",
4113 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4114 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4115 [EXEC_OUTPUT_KMSG
] = "kmsg",
4116 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4117 [EXEC_OUTPUT_JOURNAL
] = "journal",
4118 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4119 [EXEC_OUTPUT_SOCKET
] = "socket",
4120 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4123 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4125 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4126 [EXEC_UTMP_INIT
] = "init",
4127 [EXEC_UTMP_LOGIN
] = "login",
4128 [EXEC_UTMP_USER
] = "user",
4131 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);