1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
29 #include <sys/socket.h>
31 #include <sys/prctl.h>
32 #include <linux/sched.h>
33 #include <sys/types.h>
37 #include <sys/mount.h>
39 #include <linux/oom.h>
42 #include <sys/personality.h>
47 #include <security/pam_appl.h>
51 #include <selinux/selinux.h>
59 #include <sys/apparmor.h>
65 #include "capability.h"
68 #include "sd-messages.h"
70 #include "securebits.h"
71 #include "namespace.h"
72 #include "exit-status.h"
74 #include "utmp-wtmp.h"
76 #include "path-util.h"
81 #include "selinux-util.h"
82 #include "errno-list.h"
85 #include "apparmor-util.h"
86 #include "bus-kernel.h"
89 #include "seccomp-util.h"
92 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
93 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
95 /* This assumes there is a 'tty' group */
98 #define SNDBUF_SIZE (8*1024*1024)
100 static int shift_fds(int fds
[], unsigned n_fds
) {
101 int start
, restart_from
;
106 /* Modifies the fds array! (sorts it) */
116 for (i
= start
; i
< (int) n_fds
; i
++) {
119 /* Already at right index? */
123 if ((nfd
= fcntl(fds
[i
], F_DUPFD
, i
+3)) < 0)
129 /* Hmm, the fd we wanted isn't free? Then
130 * let's remember that and try again from here*/
131 if (nfd
!= i
+3 && restart_from
< 0)
135 if (restart_from
< 0)
138 start
= restart_from
;
144 static int flags_fds(const int fds
[], unsigned n_fds
, bool nonblock
) {
153 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
155 for (i
= 0; i
< n_fds
; i
++) {
157 if ((r
= fd_nonblock(fds
[i
], nonblock
)) < 0)
160 /* We unconditionally drop FD_CLOEXEC from the fds,
161 * since after all we want to pass these fds to our
164 if ((r
= fd_cloexec(fds
[i
], false)) < 0)
171 _pure_
static const char *tty_path(const ExecContext
*context
) {
174 if (context
->tty_path
)
175 return context
->tty_path
;
177 return "/dev/console";
180 static void exec_context_tty_reset(const ExecContext
*context
) {
183 if (context
->tty_vhangup
)
184 terminal_vhangup(tty_path(context
));
186 if (context
->tty_reset
)
187 reset_terminal(tty_path(context
));
189 if (context
->tty_vt_disallocate
&& context
->tty_path
)
190 vt_disallocate(context
->tty_path
);
193 static bool is_terminal_output(ExecOutput o
) {
195 o
== EXEC_OUTPUT_TTY
||
196 o
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
197 o
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
198 o
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
;
201 static int open_null_as(int flags
, int nfd
) {
206 fd
= open("/dev/null", flags
|O_NOCTTY
);
211 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
219 static int connect_logger_as(const ExecContext
*context
, ExecOutput output
, const char *ident
, const char *unit_id
, int nfd
) {
221 union sockaddr_union sa
= {
222 .un
.sun_family
= AF_UNIX
,
223 .un
.sun_path
= "/run/systemd/journal/stdout",
227 assert(output
< _EXEC_OUTPUT_MAX
);
231 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
235 r
= connect(fd
, &sa
.sa
, offsetof(struct sockaddr_un
, sun_path
) + strlen(sa
.un
.sun_path
));
241 if (shutdown(fd
, SHUT_RD
) < 0) {
246 fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
256 context
->syslog_identifier
? context
->syslog_identifier
: ident
,
258 context
->syslog_priority
,
259 !!context
->syslog_level_prefix
,
260 output
== EXEC_OUTPUT_SYSLOG
|| output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
261 output
== EXEC_OUTPUT_KMSG
|| output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
,
262 is_terminal_output(output
));
265 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
272 static int open_terminal_as(const char *path
, mode_t mode
, int nfd
) {
278 if ((fd
= open_terminal(path
, mode
| O_NOCTTY
)) < 0)
282 r
= dup2(fd
, nfd
) < 0 ? -errno
: nfd
;
290 static bool is_terminal_input(ExecInput i
) {
292 i
== EXEC_INPUT_TTY
||
293 i
== EXEC_INPUT_TTY_FORCE
||
294 i
== EXEC_INPUT_TTY_FAIL
;
297 static int fixup_input(ExecInput std_input
, int socket_fd
, bool apply_tty_stdin
) {
299 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
300 return EXEC_INPUT_NULL
;
302 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
303 return EXEC_INPUT_NULL
;
308 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
310 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
311 return EXEC_OUTPUT_INHERIT
;
316 static int setup_input(const ExecContext
*context
, int socket_fd
, bool apply_tty_stdin
) {
321 i
= fixup_input(context
->std_input
, socket_fd
, apply_tty_stdin
);
325 case EXEC_INPUT_NULL
:
326 return open_null_as(O_RDONLY
, STDIN_FILENO
);
329 case EXEC_INPUT_TTY_FORCE
:
330 case EXEC_INPUT_TTY_FAIL
: {
333 fd
= acquire_terminal(tty_path(context
),
334 i
== EXEC_INPUT_TTY_FAIL
,
335 i
== EXEC_INPUT_TTY_FORCE
,
341 if (fd
!= STDIN_FILENO
) {
342 r
= dup2(fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
350 case EXEC_INPUT_SOCKET
:
351 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
354 assert_not_reached("Unknown input type");
358 static int setup_output(const ExecContext
*context
, int fileno
, int socket_fd
, const char *ident
, const char *unit_id
, bool apply_tty_stdin
) {
366 i
= fixup_input(context
->std_input
, socket_fd
, apply_tty_stdin
);
367 o
= fixup_output(context
->std_output
, socket_fd
);
369 if (fileno
== STDERR_FILENO
) {
371 e
= fixup_output(context
->std_error
, socket_fd
);
373 /* This expects the input and output are already set up */
375 /* Don't change the stderr file descriptor if we inherit all
376 * the way and are not on a tty */
377 if (e
== EXEC_OUTPUT_INHERIT
&&
378 o
== EXEC_OUTPUT_INHERIT
&&
379 i
== EXEC_INPUT_NULL
&&
380 !is_terminal_input(context
->std_input
) &&
384 /* Duplicate from stdout if possible */
385 if (e
== o
|| e
== EXEC_OUTPUT_INHERIT
)
386 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
390 } else if (o
== EXEC_OUTPUT_INHERIT
) {
391 /* If input got downgraded, inherit the original value */
392 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
393 return open_terminal_as(tty_path(context
), O_WRONLY
, fileno
);
395 /* If the input is connected to anything that's not a /dev/null, inherit that... */
396 if (i
!= EXEC_INPUT_NULL
)
397 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
399 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
403 /* We need to open /dev/null here anew, to get the right access mode. */
404 return open_null_as(O_WRONLY
, fileno
);
409 case EXEC_OUTPUT_NULL
:
410 return open_null_as(O_WRONLY
, fileno
);
412 case EXEC_OUTPUT_TTY
:
413 if (is_terminal_input(i
))
414 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
416 /* We don't reset the terminal if this is just about output */
417 return open_terminal_as(tty_path(context
), O_WRONLY
, fileno
);
419 case EXEC_OUTPUT_SYSLOG
:
420 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
421 case EXEC_OUTPUT_KMSG
:
422 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
423 case EXEC_OUTPUT_JOURNAL
:
424 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
425 r
= connect_logger_as(context
, o
, ident
, unit_id
, fileno
);
427 log_struct_unit(LOG_CRIT
, unit_id
,
428 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
429 fileno
== STDOUT_FILENO
? "out" : "err",
430 unit_id
, strerror(-r
),
433 r
= open_null_as(O_WRONLY
, fileno
);
437 case EXEC_OUTPUT_SOCKET
:
438 assert(socket_fd
>= 0);
439 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
442 assert_not_reached("Unknown error type");
446 static int chown_terminal(int fd
, uid_t uid
) {
451 /* This might fail. What matters are the results. */
452 (void) fchown(fd
, uid
, -1);
453 (void) fchmod(fd
, TTY_MODE
);
455 if (fstat(fd
, &st
) < 0)
458 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
464 static int setup_confirm_stdio(int *_saved_stdin
,
465 int *_saved_stdout
) {
466 int fd
= -1, saved_stdin
, saved_stdout
= -1, r
;
468 assert(_saved_stdin
);
469 assert(_saved_stdout
);
471 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
475 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
476 if (saved_stdout
< 0) {
481 fd
= acquire_terminal(
486 DEFAULT_CONFIRM_USEC
);
492 r
= chown_terminal(fd
, getuid());
496 if (dup2(fd
, STDIN_FILENO
) < 0) {
501 if (dup2(fd
, STDOUT_FILENO
) < 0) {
509 *_saved_stdin
= saved_stdin
;
510 *_saved_stdout
= saved_stdout
;
515 safe_close(saved_stdout
);
516 safe_close(saved_stdin
);
522 _printf_(1, 2) static int write_confirm_message(const char *format
, ...) {
523 _cleanup_close_
int fd
= -1;
528 fd
= open_terminal("/dev/console", O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
532 va_start(ap
, format
);
533 vdprintf(fd
, format
, ap
);
539 static int restore_confirm_stdio(int *saved_stdin
,
545 assert(saved_stdout
);
549 if (*saved_stdin
>= 0)
550 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
553 if (*saved_stdout
>= 0)
554 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
557 safe_close(*saved_stdin
);
558 safe_close(*saved_stdout
);
563 static int ask_for_confirmation(char *response
, char **argv
) {
564 int saved_stdout
= -1, saved_stdin
= -1, r
;
565 _cleanup_free_
char *line
= NULL
;
567 r
= setup_confirm_stdio(&saved_stdin
, &saved_stdout
);
571 line
= exec_command_line(argv
);
575 r
= ask_char(response
, "yns", "Execute %s? [Yes, No, Skip] ", line
);
577 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
582 static int enforce_groups(const ExecContext
*context
, const char *username
, gid_t gid
) {
583 bool keep_groups
= false;
588 /* Lookup and set GID and supplementary group list. Here too
589 * we avoid NSS lookups for gid=0. */
591 if (context
->group
|| username
) {
593 if (context
->group
) {
594 const char *g
= context
->group
;
596 if ((r
= get_group_creds(&g
, &gid
)) < 0)
600 /* First step, initialize groups from /etc/groups */
601 if (username
&& gid
!= 0) {
602 if (initgroups(username
, gid
) < 0)
608 /* Second step, set our gids */
609 if (setresgid(gid
, gid
, gid
) < 0)
613 if (context
->supplementary_groups
) {
618 /* Final step, initialize any manually set supplementary groups */
619 assert_se((ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
)) > 0);
621 if (!(gids
= new(gid_t
, ngroups_max
)))
625 if ((k
= getgroups(ngroups_max
, gids
)) < 0) {
632 STRV_FOREACH(i
, context
->supplementary_groups
) {
635 if (k
>= ngroups_max
) {
641 r
= get_group_creds(&g
, gids
+k
);
650 if (setgroups(k
, gids
) < 0) {
661 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
664 /* Sets (but doesn't lookup) the uid and make sure we keep the
665 * capabilities while doing so. */
667 if (context
->capabilities
) {
668 _cleanup_cap_free_ cap_t d
= NULL
;
669 static const cap_value_t bits
[] = {
670 CAP_SETUID
, /* Necessary so that we can run setresuid() below */
671 CAP_SETPCAP
/* Necessary so that we can set PR_SET_SECUREBITS later on */
674 /* First step: If we need to keep capabilities but
675 * drop privileges we need to make sure we keep our
676 * caps, while we drop privileges. */
678 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
680 if (prctl(PR_GET_SECUREBITS
) != sb
)
681 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
685 /* Second step: set the capabilities. This will reduce
686 * the capabilities to the minimum we need. */
688 d
= cap_dup(context
->capabilities
);
692 if (cap_set_flag(d
, CAP_EFFECTIVE
, ELEMENTSOF(bits
), bits
, CAP_SET
) < 0 ||
693 cap_set_flag(d
, CAP_PERMITTED
, ELEMENTSOF(bits
), bits
, CAP_SET
) < 0)
696 if (cap_set_proc(d
) < 0)
700 /* Third step: actually set the uids */
701 if (setresuid(uid
, uid
, uid
) < 0)
704 /* At this point we should have all necessary capabilities but
705 are otherwise a normal user. However, the caps might got
706 corrupted due to the setresuid() so we need clean them up
707 later. This is done outside of this call. */
714 static int null_conv(
716 const struct pam_message
**msg
,
717 struct pam_response
**resp
,
720 /* We don't support conversations */
725 static int setup_pam(
731 int fds
[], unsigned n_fds
) {
733 static const struct pam_conv conv
= {
738 pam_handle_t
*handle
= NULL
;
740 int pam_code
= PAM_SUCCESS
;
743 bool close_session
= false;
744 pid_t pam_pid
= 0, parent_pid
;
751 /* We set up PAM in the parent process, then fork. The child
752 * will then stay around until killed via PR_GET_PDEATHSIG or
753 * systemd via the cgroup logic. It will then remove the PAM
754 * session again. The parent process will exec() the actual
755 * daemon. We do things this way to ensure that the main PID
756 * of the daemon is the one we initially fork()ed. */
758 if (log_get_max_level() < LOG_PRI(LOG_DEBUG
))
761 pam_code
= pam_start(name
, user
, &conv
, &handle
);
762 if (pam_code
!= PAM_SUCCESS
) {
768 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
769 if (pam_code
!= PAM_SUCCESS
)
773 pam_code
= pam_acct_mgmt(handle
, flags
);
774 if (pam_code
!= PAM_SUCCESS
)
777 pam_code
= pam_open_session(handle
, flags
);
778 if (pam_code
!= PAM_SUCCESS
)
781 close_session
= true;
783 e
= pam_getenvlist(handle
);
785 pam_code
= PAM_BUF_ERR
;
789 /* Block SIGTERM, so that we know that it won't get lost in
791 if (sigemptyset(&ss
) < 0 ||
792 sigaddset(&ss
, SIGTERM
) < 0 ||
793 sigprocmask(SIG_BLOCK
, &ss
, &old_ss
) < 0)
796 parent_pid
= getpid();
806 /* The child's job is to reset the PAM session on
809 /* This string must fit in 10 chars (i.e. the length
810 * of "/sbin/init"), to look pretty in /bin/ps */
811 rename_process("(sd-pam)");
813 /* Make sure we don't keep open the passed fds in this
814 child. We assume that otherwise only those fds are
815 open here that have been opened by PAM. */
816 close_many(fds
, n_fds
);
818 /* Drop privileges - we don't need any to pam_close_session
819 * and this will make PR_SET_PDEATHSIG work in most cases.
820 * If this fails, ignore the error - but expect sd-pam threads
821 * to fail to exit normally */
822 if (setresuid(uid
, uid
, uid
) < 0)
823 log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r
));
825 /* Wait until our parent died. This will only work if
826 * the above setresuid() succeeds, otherwise the kernel
827 * will not allow unprivileged parents kill their privileged
828 * children this way. We rely on the control groups kill logic
829 * to do the rest for us. */
830 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
833 /* Check if our parent process might already have
835 if (getppid() == parent_pid
) {
837 if (sigwait(&ss
, &sig
) < 0) {
844 assert(sig
== SIGTERM
);
849 /* If our parent died we'll end the session */
850 if (getppid() != parent_pid
) {
851 pam_code
= pam_close_session(handle
, flags
);
852 if (pam_code
!= PAM_SUCCESS
)
859 pam_end(handle
, pam_code
| flags
);
863 /* If the child was forked off successfully it will do all the
864 * cleanups, so forget about the handle here. */
867 /* Unblock SIGTERM again in the parent */
868 if (sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) < 0)
871 /* We close the log explicitly here, since the PAM modules
872 * might have opened it, but we don't want this fd around. */
881 if (pam_code
!= PAM_SUCCESS
) {
882 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
883 err
= -EPERM
; /* PAM errors do not map to errno */
885 log_error("PAM failed: %m");
891 pam_code
= pam_close_session(handle
, flags
);
893 pam_end(handle
, pam_code
| flags
);
901 kill(pam_pid
, SIGTERM
);
902 kill(pam_pid
, SIGCONT
);
909 static void rename_process_from_path(const char *path
) {
910 char process_name
[11];
914 /* This resulting string must fit in 10 chars (i.e. the length
915 * of "/sbin/init") to look pretty in /bin/ps */
919 rename_process("(...)");
925 /* The end of the process name is usually more
926 * interesting, since the first bit might just be
932 process_name
[0] = '(';
933 memcpy(process_name
+1, p
, l
);
934 process_name
[1+l
] = ')';
935 process_name
[1+l
+1] = 0;
937 rename_process(process_name
);
942 static int apply_seccomp(const ExecContext
*c
) {
943 uint32_t negative_action
, action
;
944 scmp_filter_ctx
*seccomp
;
951 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
953 seccomp
= seccomp_init(c
->syscall_whitelist
? negative_action
: SCMP_ACT_ALLOW
);
957 if (c
->syscall_archs
) {
959 SET_FOREACH(id
, c
->syscall_archs
, i
) {
960 r
= seccomp_arch_add(seccomp
, PTR_TO_UINT32(id
) - 1);
968 r
= seccomp_add_secondary_archs(seccomp
);
973 action
= c
->syscall_whitelist
? SCMP_ACT_ALLOW
: negative_action
;
974 SET_FOREACH(id
, c
->syscall_filter
, i
) {
975 r
= seccomp_rule_add(seccomp
, action
, PTR_TO_INT(id
) - 1, 0);
980 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
984 r
= seccomp_load(seccomp
);
987 seccomp_release(seccomp
);
991 static int apply_address_families(const ExecContext
*c
) {
992 scmp_filter_ctx
*seccomp
;
998 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1002 r
= seccomp_add_secondary_archs(seccomp
);
1006 if (c
->address_families_whitelist
) {
1007 int af
, first
= 0, last
= 0;
1010 /* If this is a whitelist, we first block the address
1011 * families that are out of range and then everything
1012 * that is not in the set. First, we find the lowest
1013 * and highest address family in the set. */
1015 SET_FOREACH(afp
, c
->address_families
, i
) {
1016 af
= PTR_TO_INT(afp
);
1018 if (af
<= 0 || af
>= af_max())
1021 if (first
== 0 || af
< first
)
1024 if (last
== 0 || af
> last
)
1028 assert((first
== 0) == (last
== 0));
1032 /* No entries in the valid range, block everything */
1033 r
= seccomp_rule_add(
1035 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1043 /* Block everything below the first entry */
1044 r
= seccomp_rule_add(
1046 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1049 SCMP_A0(SCMP_CMP_LT
, first
));
1053 /* Block everything above the last entry */
1054 r
= seccomp_rule_add(
1056 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1059 SCMP_A0(SCMP_CMP_GT
, last
));
1063 /* Block everything between the first and last
1065 for (af
= 1; af
< af_max(); af
++) {
1067 if (set_contains(c
->address_families
, INT_TO_PTR(af
)))
1070 r
= seccomp_rule_add(
1072 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1075 SCMP_A0(SCMP_CMP_EQ
, af
));
1084 /* If this is a blacklist, then generate one rule for
1085 * each address family that are then combined in OR
1088 SET_FOREACH(af
, c
->address_families
, i
) {
1090 r
= seccomp_rule_add(
1092 SCMP_ACT_ERRNO(EPROTONOSUPPORT
),
1095 SCMP_A0(SCMP_CMP_EQ
, PTR_TO_INT(af
)));
1101 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1105 r
= seccomp_load(seccomp
);
1108 seccomp_release(seccomp
);
1114 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1118 safe_close(idle_pipe
[1]);
1119 safe_close(idle_pipe
[2]);
1121 if (idle_pipe
[0] >= 0) {
1124 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1126 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1127 /* Signal systemd that we are bored and want to continue. */
1128 write(idle_pipe
[3], "x", 1);
1130 /* Wait for systemd to react to the signal above. */
1131 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1134 safe_close(idle_pipe
[0]);
1138 safe_close(idle_pipe
[3]);
1141 static int build_environment(
1142 const ExecContext
*c
,
1144 usec_t watchdog_usec
,
1146 const char *username
,
1150 _cleanup_strv_free_
char **our_env
= NULL
;
1157 our_env
= new0(char*, 10);
1162 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid()) < 0)
1164 our_env
[n_env
++] = x
;
1166 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1168 our_env
[n_env
++] = x
;
1171 if (watchdog_usec
> 0) {
1172 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid()) < 0)
1174 our_env
[n_env
++] = x
;
1176 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, watchdog_usec
) < 0)
1178 our_env
[n_env
++] = x
;
1182 x
= strappend("HOME=", home
);
1185 our_env
[n_env
++] = x
;
1189 x
= strappend("LOGNAME=", username
);
1192 our_env
[n_env
++] = x
;
1194 x
= strappend("USER=", username
);
1197 our_env
[n_env
++] = x
;
1201 x
= strappend("SHELL=", shell
);
1204 our_env
[n_env
++] = x
;
1207 if (is_terminal_input(c
->std_input
) ||
1208 c
->std_output
== EXEC_OUTPUT_TTY
||
1209 c
->std_error
== EXEC_OUTPUT_TTY
||
1212 x
= strdup(default_term_for_tty(tty_path(c
)));
1215 our_env
[n_env
++] = x
;
1218 our_env
[n_env
++] = NULL
;
1219 assert(n_env
<= 10);
1227 static int exec_child(ExecCommand
*command
,
1228 const ExecContext
*context
,
1229 const ExecParameters
*params
,
1230 ExecRuntime
*runtime
,
1233 int *fds
, unsigned n_fds
,
1237 _cleanup_strv_free_
char **our_env
= NULL
, **pam_env
= NULL
, **final_env
= NULL
, **final_argv
= NULL
;
1238 const char *username
= NULL
, *home
= NULL
, *shell
= NULL
;
1239 unsigned n_dont_close
= 0;
1240 int dont_close
[n_fds
+ 4];
1241 uid_t uid
= (uid_t
) -1;
1242 gid_t gid
= (gid_t
) -1;
1250 rename_process_from_path(command
->path
);
1252 /* We reset exactly these signals, since they are the
1253 * only ones we set to SIG_IGN in the main daemon. All
1254 * others we leave untouched because we set them to
1255 * SIG_DFL or a valid handler initially, both of which
1256 * will be demoted to SIG_DFL. */
1257 default_signals(SIGNALS_CRASH_HANDLER
,
1258 SIGNALS_IGNORE
, -1);
1260 if (context
->ignore_sigpipe
)
1261 ignore_signals(SIGPIPE
, -1);
1263 err
= reset_signal_mask();
1265 *error
= EXIT_SIGNAL_MASK
;
1269 if (params
->idle_pipe
)
1270 do_idle_pipe_dance(params
->idle_pipe
);
1272 /* Close sockets very early to make sure we don't
1273 * block init reexecution because it cannot bind its
1278 dont_close
[n_dont_close
++] = socket_fd
;
1280 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
1281 n_dont_close
+= n_fds
;
1283 if (params
->bus_endpoint_fd
>= 0)
1284 dont_close
[n_dont_close
++] = params
->bus_endpoint_fd
;
1286 if (runtime
->netns_storage_socket
[0] >= 0)
1287 dont_close
[n_dont_close
++] = runtime
->netns_storage_socket
[0];
1288 if (runtime
->netns_storage_socket
[1] >= 0)
1289 dont_close
[n_dont_close
++] = runtime
->netns_storage_socket
[1];
1292 err
= close_all_fds(dont_close
, n_dont_close
);
1298 if (!context
->same_pgrp
)
1300 *error
= EXIT_SETSID
;
1304 exec_context_tty_reset(context
);
1306 if (params
->confirm_spawn
) {
1309 err
= ask_for_confirmation(&response
, argv
);
1310 if (err
== -ETIMEDOUT
)
1311 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1313 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err
));
1314 else if (response
== 's') {
1315 write_confirm_message("Skipping execution.\n");
1316 *error
= EXIT_CONFIRM
;
1318 } else if (response
== 'n') {
1319 write_confirm_message("Failing execution.\n");
1325 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1326 * must sure to drop O_NONBLOCK */
1328 fd_nonblock(socket_fd
, false);
1330 err
= setup_input(context
, socket_fd
, params
->apply_tty_stdin
);
1332 *error
= EXIT_STDIN
;
1336 err
= setup_output(context
, STDOUT_FILENO
, socket_fd
, basename(command
->path
), params
->unit_id
, params
->apply_tty_stdin
);
1338 *error
= EXIT_STDOUT
;
1342 err
= setup_output(context
, STDERR_FILENO
, socket_fd
, basename(command
->path
), params
->unit_id
, params
->apply_tty_stdin
);
1344 *error
= EXIT_STDERR
;
1348 if (params
->cgroup_path
) {
1349 err
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0);
1351 *error
= EXIT_CGROUP
;
1356 if (context
->oom_score_adjust_set
) {
1359 snprintf(t
, sizeof(t
), "%i", context
->oom_score_adjust
);
1362 if (write_string_file("/proc/self/oom_score_adj", t
) < 0) {
1363 *error
= EXIT_OOM_ADJUST
;
1368 if (context
->nice_set
)
1369 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
1374 if (context
->cpu_sched_set
) {
1375 struct sched_param param
= {
1376 .sched_priority
= context
->cpu_sched_priority
,
1379 err
= sched_setscheduler(0,
1380 context
->cpu_sched_policy
|
1381 (context
->cpu_sched_reset_on_fork
?
1382 SCHED_RESET_ON_FORK
: 0),
1385 *error
= EXIT_SETSCHEDULER
;
1390 if (context
->cpuset
)
1391 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
1392 *error
= EXIT_CPUAFFINITY
;
1396 if (context
->ioprio_set
)
1397 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
1398 *error
= EXIT_IOPRIO
;
1402 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
1403 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
1404 *error
= EXIT_TIMERSLACK
;
1408 if (context
->personality
!= 0xffffffffUL
)
1409 if (personality(context
->personality
) < 0) {
1410 *error
= EXIT_PERSONALITY
;
1414 if (context
->utmp_id
)
1415 utmp_put_init_process(context
->utmp_id
, getpid(), getsid(0), context
->tty_path
);
1417 if (context
->user
) {
1418 username
= context
->user
;
1419 err
= get_user_creds(&username
, &uid
, &gid
, &home
, &shell
);
1425 if (is_terminal_input(context
->std_input
)) {
1426 err
= chown_terminal(STDIN_FILENO
, uid
);
1428 *error
= EXIT_STDIN
;
1435 if (params
->bus_endpoint_fd
>= 0 && context
->bus_endpoint
) {
1436 uid_t ep_uid
= (uid
== (uid_t
) -1) ? 0 : uid
;
1438 err
= bus_kernel_set_endpoint_policy(params
->bus_endpoint_fd
, ep_uid
, context
->bus_endpoint
);
1440 *error
= EXIT_BUS_ENDPOINT
;
1447 if (params
->cgroup_path
&& context
->user
&& context
->pam_name
) {
1448 err
= cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0644, uid
, gid
);
1450 *error
= EXIT_CGROUP
;
1455 err
= cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, 0755, uid
, gid
);
1457 *error
= EXIT_CGROUP
;
1463 if (!strv_isempty(context
->runtime_directory
) && params
->runtime_prefix
) {
1466 STRV_FOREACH(rt
, context
->runtime_directory
) {
1467 _cleanup_free_
char *p
;
1469 p
= strjoin(params
->runtime_prefix
, "/", *rt
, NULL
);
1471 *error
= EXIT_RUNTIME_DIRECTORY
;
1475 err
= mkdir_safe(p
, context
->runtime_directory_mode
, uid
, gid
);
1477 *error
= EXIT_RUNTIME_DIRECTORY
;
1483 if (params
->apply_permissions
) {
1484 err
= enforce_groups(context
, username
, gid
);
1486 *error
= EXIT_GROUP
;
1491 umask(context
->umask
);
1494 if (params
->apply_permissions
&& context
->pam_name
&& username
) {
1495 err
= setup_pam(context
->pam_name
, username
, uid
, context
->tty_path
, &pam_env
, fds
, n_fds
);
1503 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
1504 err
= setup_netns(runtime
->netns_storage_socket
);
1506 *error
= EXIT_NETWORK
;
1511 if (!strv_isempty(context
->read_write_dirs
) ||
1512 !strv_isempty(context
->read_only_dirs
) ||
1513 !strv_isempty(context
->inaccessible_dirs
) ||
1514 context
->mount_flags
!= 0 ||
1515 (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
)) ||
1516 params
->bus_endpoint_path
||
1517 context
->private_devices
||
1518 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1519 context
->protect_home
!= PROTECT_HOME_NO
) {
1521 char *tmp
= NULL
, *var
= NULL
;
1523 /* The runtime struct only contains the parent
1524 * of the private /tmp, which is
1525 * non-accessible to world users. Inside of it
1526 * there's a /tmp that is sticky, and that's
1527 * the one we want to use here. */
1529 if (context
->private_tmp
&& runtime
) {
1530 if (runtime
->tmp_dir
)
1531 tmp
= strappenda(runtime
->tmp_dir
, "/tmp");
1532 if (runtime
->var_tmp_dir
)
1533 var
= strappenda(runtime
->var_tmp_dir
, "/tmp");
1536 err
= setup_namespace(
1537 context
->read_write_dirs
,
1538 context
->read_only_dirs
,
1539 context
->inaccessible_dirs
,
1542 params
->bus_endpoint_path
,
1543 context
->private_devices
,
1544 context
->protect_home
,
1545 context
->protect_system
,
1546 context
->mount_flags
);
1548 *error
= EXIT_NAMESPACE
;
1553 if (params
->apply_chroot
) {
1554 if (context
->root_directory
)
1555 if (chroot(context
->root_directory
) < 0) {
1556 *error
= EXIT_CHROOT
;
1560 if (chdir(context
->working_directory
? context
->working_directory
: "/") < 0) {
1561 *error
= EXIT_CHDIR
;
1565 _cleanup_free_
char *d
= NULL
;
1567 if (asprintf(&d
, "%s/%s",
1568 context
->root_directory
? context
->root_directory
: "",
1569 context
->working_directory
? context
->working_directory
: "") < 0) {
1570 *error
= EXIT_MEMORY
;
1575 *error
= EXIT_CHDIR
;
1580 /* We repeat the fd closing here, to make sure that
1581 * nothing is leaked from the PAM modules. Note that
1582 * we are more aggressive this time since socket_fd
1583 * and the netns fds we don't need anymore. The custom
1584 * endpoint fd was needed to upload the policy and can
1585 * now be closed as well. */
1586 err
= close_all_fds(fds
, n_fds
);
1588 err
= shift_fds(fds
, n_fds
);
1590 err
= flags_fds(fds
, n_fds
, context
->non_blocking
);
1596 if (params
->apply_permissions
) {
1598 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
1599 if (!context
->rlimit
[i
])
1602 if (setrlimit_closest(i
, context
->rlimit
[i
]) < 0) {
1603 *error
= EXIT_LIMITS
;
1608 if (context
->capability_bounding_set_drop
) {
1609 err
= capability_bounding_set_drop(context
->capability_bounding_set_drop
, false);
1611 *error
= EXIT_CAPABILITIES
;
1616 if (context
->user
) {
1617 err
= enforce_user(context
, uid
);
1624 /* PR_GET_SECUREBITS is not privileged, while
1625 * PR_SET_SECUREBITS is. So to suppress
1626 * potential EPERMs we'll try not to call
1627 * PR_SET_SECUREBITS unless necessary. */
1628 if (prctl(PR_GET_SECUREBITS
) != context
->secure_bits
)
1629 if (prctl(PR_SET_SECUREBITS
, context
->secure_bits
) < 0) {
1630 *error
= EXIT_SECUREBITS
;
1634 if (context
->capabilities
)
1635 if (cap_set_proc(context
->capabilities
) < 0) {
1636 *error
= EXIT_CAPABILITIES
;
1640 if (context
->no_new_privileges
)
1641 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
1642 *error
= EXIT_NO_NEW_PRIVILEGES
;
1647 if (context
->address_families_whitelist
||
1648 !set_isempty(context
->address_families
)) {
1649 err
= apply_address_families(context
);
1651 *error
= EXIT_ADDRESS_FAMILIES
;
1656 if (context
->syscall_whitelist
||
1657 !set_isempty(context
->syscall_filter
) ||
1658 !set_isempty(context
->syscall_archs
)) {
1659 err
= apply_seccomp(context
);
1661 *error
= EXIT_SECCOMP
;
1668 if (context
->selinux_context
&& use_selinux()) {
1669 err
= setexeccon(context
->selinux_context
);
1670 if (err
< 0 && !context
->selinux_context_ignore
) {
1671 *error
= EXIT_SELINUX_CONTEXT
;
1677 #ifdef HAVE_APPARMOR
1678 if (context
->apparmor_profile
&& use_apparmor()) {
1679 err
= aa_change_onexec(context
->apparmor_profile
);
1680 if (err
< 0 && !context
->apparmor_profile_ignore
) {
1681 *error
= EXIT_APPARMOR_PROFILE
;
1688 err
= build_environment(context
, n_fds
, params
->watchdog_usec
, home
, username
, shell
, &our_env
);
1690 *error
= EXIT_MEMORY
;
1694 final_env
= strv_env_merge(5,
1695 params
->environment
,
1697 context
->environment
,
1702 *error
= EXIT_MEMORY
;
1706 final_argv
= replace_env_argv(argv
, final_env
);
1708 *error
= EXIT_MEMORY
;
1712 final_env
= strv_env_clean(final_env
);
1714 if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG
))) {
1715 _cleanup_free_
char *line
;
1717 line
= exec_command_line(final_argv
);
1720 log_struct_unit(LOG_DEBUG
,
1722 "EXECUTABLE=%s", command
->path
,
1723 "MESSAGE=Executing: %s", line
,
1728 execve(command
->path
, final_argv
, final_env
);
1733 int exec_spawn(ExecCommand
*command
,
1734 const ExecContext
*context
,
1735 const ExecParameters
*params
,
1736 ExecRuntime
*runtime
,
1739 _cleanup_strv_free_
char **files_env
= NULL
;
1740 int *fds
= NULL
; unsigned n_fds
= 0;
1750 assert(params
->fds
|| params
->n_fds
<= 0);
1752 if (context
->std_input
== EXEC_INPUT_SOCKET
||
1753 context
->std_output
== EXEC_OUTPUT_SOCKET
||
1754 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
1756 if (params
->n_fds
!= 1)
1759 socket_fd
= params
->fds
[0];
1763 n_fds
= params
->n_fds
;
1766 err
= exec_context_load_environment(context
, &files_env
);
1768 log_struct_unit(LOG_ERR
,
1770 "MESSAGE=Failed to load environment files: %s", strerror(-err
),
1776 argv
= params
->argv
?: command
->argv
;
1778 line
= exec_command_line(argv
);
1782 log_struct_unit(LOG_DEBUG
,
1784 "EXECUTABLE=%s", command
->path
,
1785 "MESSAGE=About to execute: %s", line
,
1796 err
= exec_child(command
,
1807 log_struct(LOG_ERR
, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED
),
1808 "EXECUTABLE=%s", command
->path
,
1809 "MESSAGE=Failed at step %s spawning %s: %s",
1810 exit_status_to_string(r
, EXIT_STATUS_SYSTEMD
),
1811 command
->path
, strerror(-err
),
1820 log_struct_unit(LOG_DEBUG
,
1822 "MESSAGE=Forked %s as "PID_FMT
,
1826 /* We add the new process to the cgroup both in the child (so
1827 * that we can be sure that no user code is ever executed
1828 * outside of the cgroup) and in the parent (so that we can be
1829 * sure that when we kill the cgroup the process will be
1831 if (params
->cgroup_path
)
1832 cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
1834 exec_status_start(&command
->exec_status
, pid
);
1840 void exec_context_init(ExecContext
*c
) {
1844 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
1845 c
->cpu_sched_policy
= SCHED_OTHER
;
1846 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
1847 c
->syslog_level_prefix
= true;
1848 c
->ignore_sigpipe
= true;
1849 c
->timer_slack_nsec
= NSEC_INFINITY
;
1850 c
->personality
= 0xffffffffUL
;
1851 c
->runtime_directory_mode
= 0755;
1854 void exec_context_done(ExecContext
*c
) {
1859 strv_free(c
->environment
);
1860 c
->environment
= NULL
;
1862 strv_free(c
->environment_files
);
1863 c
->environment_files
= NULL
;
1865 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++) {
1867 c
->rlimit
[l
] = NULL
;
1870 free(c
->working_directory
);
1871 c
->working_directory
= NULL
;
1872 free(c
->root_directory
);
1873 c
->root_directory
= NULL
;
1878 free(c
->syslog_identifier
);
1879 c
->syslog_identifier
= NULL
;
1887 strv_free(c
->supplementary_groups
);
1888 c
->supplementary_groups
= NULL
;
1893 if (c
->capabilities
) {
1894 cap_free(c
->capabilities
);
1895 c
->capabilities
= NULL
;
1898 strv_free(c
->read_only_dirs
);
1899 c
->read_only_dirs
= NULL
;
1901 strv_free(c
->read_write_dirs
);
1902 c
->read_write_dirs
= NULL
;
1904 strv_free(c
->inaccessible_dirs
);
1905 c
->inaccessible_dirs
= NULL
;
1908 CPU_FREE(c
->cpuset
);
1913 free(c
->selinux_context
);
1914 c
->selinux_context
= NULL
;
1916 free(c
->apparmor_profile
);
1917 c
->apparmor_profile
= NULL
;
1919 set_free(c
->syscall_filter
);
1920 c
->syscall_filter
= NULL
;
1922 set_free(c
->syscall_archs
);
1923 c
->syscall_archs
= NULL
;
1925 set_free(c
->address_families
);
1926 c
->address_families
= NULL
;
1928 strv_free(c
->runtime_directory
);
1929 c
->runtime_directory
= NULL
;
1931 bus_endpoint_free(c
->bus_endpoint
);
1932 c
->bus_endpoint
= NULL
;
1935 int exec_context_destroy_runtime_directory(ExecContext
*c
, const char *runtime_prefix
) {
1940 if (!runtime_prefix
)
1943 STRV_FOREACH(i
, c
->runtime_directory
) {
1944 _cleanup_free_
char *p
;
1946 p
= strjoin(runtime_prefix
, "/", *i
, NULL
);
1950 /* We execute this synchronously, since we need to be
1951 * sure this is gone when we start the service
1953 rm_rf_dangerous(p
, false, true, false);
1959 void exec_command_done(ExecCommand
*c
) {
1969 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
1972 for (i
= 0; i
< n
; i
++)
1973 exec_command_done(c
+i
);
1976 void exec_command_free_list(ExecCommand
*c
) {
1980 LIST_REMOVE(command
, c
, i
);
1981 exec_command_done(i
);
1986 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
1989 for (i
= 0; i
< n
; i
++) {
1990 exec_command_free_list(c
[i
]);
1995 int exec_context_load_environment(const ExecContext
*c
, char ***l
) {
1996 char **i
, **r
= NULL
;
2001 STRV_FOREACH(i
, c
->environment_files
) {
2004 bool ignore
= false;
2006 _cleanup_globfree_ glob_t pglob
= {};
2016 if (!path_is_absolute(fn
)) {
2024 /* Filename supports globbing, take all matching files */
2026 if (glob(fn
, 0, NULL
, &pglob
) != 0) {
2031 return errno
? -errno
: -EINVAL
;
2033 count
= pglob
.gl_pathc
;
2041 for (n
= 0; n
< count
; n
++) {
2042 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
2050 /* Log invalid environment variables with filename */
2052 p
= strv_env_clean_log(p
, pglob
.gl_pathv
[n
]);
2059 m
= strv_env_merge(2, r
, p
);
2075 static bool tty_may_match_dev_console(const char *tty
) {
2076 _cleanup_free_
char *active
= NULL
;
2079 if (startswith(tty
, "/dev/"))
2082 /* trivial identity? */
2083 if (streq(tty
, "console"))
2086 console
= resolve_dev_console(&active
);
2087 /* if we could not resolve, assume it may */
2091 /* "tty0" means the active VC, so it may be the same sometimes */
2092 return streq(console
, tty
) || (streq(console
, "tty0") && tty_is_vc(tty
));
2095 bool exec_context_may_touch_console(ExecContext
*ec
) {
2096 return (ec
->tty_reset
|| ec
->tty_vhangup
|| ec
->tty_vt_disallocate
||
2097 is_terminal_input(ec
->std_input
) ||
2098 is_terminal_output(ec
->std_output
) ||
2099 is_terminal_output(ec
->std_error
)) &&
2100 tty_may_match_dev_console(tty_path(ec
));
2103 static void strv_fprintf(FILE *f
, char **l
) {
2109 fprintf(f
, " %s", *g
);
2112 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
2119 prefix
= strempty(prefix
);
2123 "%sWorkingDirectory: %s\n"
2124 "%sRootDirectory: %s\n"
2125 "%sNonBlocking: %s\n"
2126 "%sPrivateTmp: %s\n"
2127 "%sPrivateNetwork: %s\n"
2128 "%sPrivateDevices: %s\n"
2129 "%sProtectHome: %s\n"
2130 "%sProtectSystem: %s\n"
2131 "%sIgnoreSIGPIPE: %s\n",
2133 prefix
, c
->working_directory
? c
->working_directory
: "/",
2134 prefix
, c
->root_directory
? c
->root_directory
: "/",
2135 prefix
, yes_no(c
->non_blocking
),
2136 prefix
, yes_no(c
->private_tmp
),
2137 prefix
, yes_no(c
->private_network
),
2138 prefix
, yes_no(c
->private_devices
),
2139 prefix
, protect_home_to_string(c
->protect_home
),
2140 prefix
, protect_system_to_string(c
->protect_system
),
2141 prefix
, yes_no(c
->ignore_sigpipe
));
2143 STRV_FOREACH(e
, c
->environment
)
2144 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
2146 STRV_FOREACH(e
, c
->environment_files
)
2147 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
2154 if (c
->oom_score_adjust_set
)
2156 "%sOOMScoreAdjust: %i\n",
2157 prefix
, c
->oom_score_adjust
);
2159 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
2161 fprintf(f
, "%s%s: "RLIM_FMT
"\n",
2162 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
2164 if (c
->ioprio_set
) {
2165 _cleanup_free_
char *class_str
= NULL
;
2167 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
2169 "%sIOSchedulingClass: %s\n"
2170 "%sIOPriority: %i\n",
2171 prefix
, strna(class_str
),
2172 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
2175 if (c
->cpu_sched_set
) {
2176 _cleanup_free_
char *policy_str
= NULL
;
2178 sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
2180 "%sCPUSchedulingPolicy: %s\n"
2181 "%sCPUSchedulingPriority: %i\n"
2182 "%sCPUSchedulingResetOnFork: %s\n",
2183 prefix
, strna(policy_str
),
2184 prefix
, c
->cpu_sched_priority
,
2185 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
2189 fprintf(f
, "%sCPUAffinity:", prefix
);
2190 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
2191 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
2192 fprintf(f
, " %u", i
);
2196 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
2197 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
2200 "%sStandardInput: %s\n"
2201 "%sStandardOutput: %s\n"
2202 "%sStandardError: %s\n",
2203 prefix
, exec_input_to_string(c
->std_input
),
2204 prefix
, exec_output_to_string(c
->std_output
),
2205 prefix
, exec_output_to_string(c
->std_error
));
2211 "%sTTYVHangup: %s\n"
2212 "%sTTYVTDisallocate: %s\n",
2213 prefix
, c
->tty_path
,
2214 prefix
, yes_no(c
->tty_reset
),
2215 prefix
, yes_no(c
->tty_vhangup
),
2216 prefix
, yes_no(c
->tty_vt_disallocate
));
2218 if (c
->std_output
== EXEC_OUTPUT_SYSLOG
||
2219 c
->std_output
== EXEC_OUTPUT_KMSG
||
2220 c
->std_output
== EXEC_OUTPUT_JOURNAL
||
2221 c
->std_output
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
2222 c
->std_output
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
2223 c
->std_output
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
||
2224 c
->std_error
== EXEC_OUTPUT_SYSLOG
||
2225 c
->std_error
== EXEC_OUTPUT_KMSG
||
2226 c
->std_error
== EXEC_OUTPUT_JOURNAL
||
2227 c
->std_error
== EXEC_OUTPUT_SYSLOG_AND_CONSOLE
||
2228 c
->std_error
== EXEC_OUTPUT_KMSG_AND_CONSOLE
||
2229 c
->std_error
== EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) {
2231 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
2233 log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
2234 log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
2237 "%sSyslogFacility: %s\n"
2238 "%sSyslogLevel: %s\n",
2239 prefix
, strna(fac_str
),
2240 prefix
, strna(lvl_str
));
2243 if (c
->capabilities
) {
2244 _cleanup_cap_free_charp_
char *t
;
2246 t
= cap_to_text(c
->capabilities
, NULL
);
2248 fprintf(f
, "%sCapabilities: %s\n", prefix
, t
);
2252 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
2254 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS
) ? " keep-caps" : "",
2255 (c
->secure_bits
& 1<<SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
2256 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
2257 (c
->secure_bits
& 1<<SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
2258 (c
->secure_bits
& 1<<SECURE_NOROOT
) ? " noroot" : "",
2259 (c
->secure_bits
& 1<<SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
2261 if (c
->capability_bounding_set_drop
) {
2263 fprintf(f
, "%sCapabilityBoundingSet:", prefix
);
2265 for (l
= 0; l
<= cap_last_cap(); l
++)
2266 if (!(c
->capability_bounding_set_drop
& ((uint64_t) 1ULL << (uint64_t) l
))) {
2267 _cleanup_cap_free_charp_
char *t
;
2271 fprintf(f
, " %s", t
);
2278 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
2280 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
2282 if (strv_length(c
->supplementary_groups
) > 0) {
2283 fprintf(f
, "%sSupplementaryGroups:", prefix
);
2284 strv_fprintf(f
, c
->supplementary_groups
);
2289 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
2291 if (strv_length(c
->read_write_dirs
) > 0) {
2292 fprintf(f
, "%sReadWriteDirs:", prefix
);
2293 strv_fprintf(f
, c
->read_write_dirs
);
2297 if (strv_length(c
->read_only_dirs
) > 0) {
2298 fprintf(f
, "%sReadOnlyDirs:", prefix
);
2299 strv_fprintf(f
, c
->read_only_dirs
);
2303 if (strv_length(c
->inaccessible_dirs
) > 0) {
2304 fprintf(f
, "%sInaccessibleDirs:", prefix
);
2305 strv_fprintf(f
, c
->inaccessible_dirs
);
2311 "%sUtmpIdentifier: %s\n",
2312 prefix
, c
->utmp_id
);
2314 if (c
->selinux_context
)
2316 "%sSELinuxContext: %s%s\n",
2317 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
2319 if (c
->personality
!= 0xffffffffUL
)
2321 "%sPersonality: %s\n",
2322 prefix
, strna(personality_to_string(c
->personality
)));
2324 if (c
->syscall_filter
) {
2332 "%sSystemCallFilter: ",
2335 if (!c
->syscall_whitelist
)
2339 SET_FOREACH(id
, c
->syscall_filter
, j
) {
2340 _cleanup_free_
char *name
= NULL
;
2347 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
2348 fputs(strna(name
), f
);
2355 if (c
->syscall_archs
) {
2362 "%sSystemCallArchitectures:",
2366 SET_FOREACH(id
, c
->syscall_archs
, j
)
2367 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
2372 if (c
->syscall_errno
!= 0)
2374 "%sSystemCallErrorNumber: %s\n",
2375 prefix
, strna(errno_to_name(c
->syscall_errno
)));
2377 if (c
->apparmor_profile
)
2379 "%sAppArmorProfile: %s%s\n",
2380 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
2383 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
2388 dual_timestamp_get(&s
->start_timestamp
);
2391 void exec_status_exit(ExecStatus
*s
, ExecContext
*context
, pid_t pid
, int code
, int status
) {
2394 if (s
->pid
&& s
->pid
!= pid
)
2398 dual_timestamp_get(&s
->exit_timestamp
);
2404 if (context
->utmp_id
)
2405 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
2407 exec_context_tty_reset(context
);
2411 void exec_status_dump(ExecStatus
*s
, FILE *f
, const char *prefix
) {
2412 char buf
[FORMAT_TIMESTAMP_MAX
];
2420 prefix
= strempty(prefix
);
2423 "%sPID: "PID_FMT
"\n",
2426 if (s
->start_timestamp
.realtime
> 0)
2428 "%sStart Timestamp: %s\n",
2429 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
2431 if (s
->exit_timestamp
.realtime
> 0)
2433 "%sExit Timestamp: %s\n"
2435 "%sExit Status: %i\n",
2436 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
2437 prefix
, sigchld_code_to_string(s
->code
),
2441 char *exec_command_line(char **argv
) {
2449 STRV_FOREACH(a
, argv
)
2452 if (!(n
= new(char, k
)))
2456 STRV_FOREACH(a
, argv
) {
2463 if (strpbrk(*a
, WHITESPACE
)) {
2474 /* FIXME: this doesn't really handle arguments that have
2475 * spaces and ticks in them */
2480 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
2481 _cleanup_free_
char *cmd
= NULL
;
2482 const char *prefix2
;
2487 prefix
= strempty(prefix
);
2488 prefix2
= strappenda(prefix
, "\t");
2490 cmd
= exec_command_line(c
->argv
);
2492 "%sCommand Line: %s\n",
2493 prefix
, cmd
? cmd
: strerror(ENOMEM
));
2495 exec_status_dump(&c
->exec_status
, f
, prefix2
);
2498 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
2501 prefix
= strempty(prefix
);
2503 LIST_FOREACH(command
, c
, c
)
2504 exec_command_dump(c
, f
, prefix
);
2507 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
2514 /* It's kind of important, that we keep the order here */
2515 LIST_FIND_TAIL(command
, *l
, end
);
2516 LIST_INSERT_AFTER(command
, *l
, end
, e
);
2521 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
2529 l
= strv_new_ap(path
, ap
);
2550 static int exec_runtime_allocate(ExecRuntime
**rt
) {
2555 *rt
= new0(ExecRuntime
, 1);
2560 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
2565 int exec_runtime_make(ExecRuntime
**rt
, ExecContext
*c
, const char *id
) {
2575 if (!c
->private_network
&& !c
->private_tmp
)
2578 r
= exec_runtime_allocate(rt
);
2582 if (c
->private_network
&& (*rt
)->netns_storage_socket
[0] < 0) {
2583 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, (*rt
)->netns_storage_socket
) < 0)
2587 if (c
->private_tmp
&& !(*rt
)->tmp_dir
) {
2588 r
= setup_tmp_dirs(id
, &(*rt
)->tmp_dir
, &(*rt
)->var_tmp_dir
);
2596 ExecRuntime
*exec_runtime_ref(ExecRuntime
*r
) {
2598 assert(r
->n_ref
> 0);
2604 ExecRuntime
*exec_runtime_unref(ExecRuntime
*r
) {
2609 assert(r
->n_ref
> 0);
2612 if (r
->n_ref
<= 0) {
2614 free(r
->var_tmp_dir
);
2615 safe_close_pair(r
->netns_storage_socket
);
2622 int exec_runtime_serialize(ExecRuntime
*rt
, Unit
*u
, FILE *f
, FDSet
*fds
) {
2631 unit_serialize_item(u
, f
, "tmp-dir", rt
->tmp_dir
);
2633 if (rt
->var_tmp_dir
)
2634 unit_serialize_item(u
, f
, "var-tmp-dir", rt
->var_tmp_dir
);
2636 if (rt
->netns_storage_socket
[0] >= 0) {
2639 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
2643 unit_serialize_item_format(u
, f
, "netns-socket-0", "%i", copy
);
2646 if (rt
->netns_storage_socket
[1] >= 0) {
2649 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
2653 unit_serialize_item_format(u
, f
, "netns-socket-1", "%i", copy
);
2659 int exec_runtime_deserialize_item(ExecRuntime
**rt
, Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
2666 if (streq(key
, "tmp-dir")) {
2669 r
= exec_runtime_allocate(rt
);
2673 copy
= strdup(value
);
2677 free((*rt
)->tmp_dir
);
2678 (*rt
)->tmp_dir
= copy
;
2680 } else if (streq(key
, "var-tmp-dir")) {
2683 r
= exec_runtime_allocate(rt
);
2687 copy
= strdup(value
);
2691 free((*rt
)->var_tmp_dir
);
2692 (*rt
)->var_tmp_dir
= copy
;
2694 } else if (streq(key
, "netns-socket-0")) {
2697 r
= exec_runtime_allocate(rt
);
2701 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
2702 log_debug_unit(u
->id
, "Failed to parse netns socket value %s", value
);
2704 safe_close((*rt
)->netns_storage_socket
[0]);
2705 (*rt
)->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
2707 } else if (streq(key
, "netns-socket-1")) {
2710 r
= exec_runtime_allocate(rt
);
2714 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
))
2715 log_debug_unit(u
->id
, "Failed to parse netns socket value %s", value
);
2717 safe_close((*rt
)->netns_storage_socket
[1]);
2718 (*rt
)->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
2726 static void *remove_tmpdir_thread(void *p
) {
2727 _cleanup_free_
char *path
= p
;
2729 rm_rf_dangerous(path
, false, true, false);
2733 void exec_runtime_destroy(ExecRuntime
*rt
) {
2739 /* If there are multiple users of this, let's leave the stuff around */
2744 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
2746 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
2748 log_warning("Failed to nuke %s: %s", rt
->tmp_dir
, strerror(-r
));
2755 if (rt
->var_tmp_dir
) {
2756 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
2758 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
2760 log_warning("Failed to nuke %s: %s", rt
->var_tmp_dir
, strerror(-r
));
2761 free(rt
->var_tmp_dir
);
2764 rt
->var_tmp_dir
= NULL
;
2767 safe_close_pair(rt
->netns_storage_socket
);
2770 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
2771 [EXEC_INPUT_NULL
] = "null",
2772 [EXEC_INPUT_TTY
] = "tty",
2773 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
2774 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
2775 [EXEC_INPUT_SOCKET
] = "socket"
2778 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
2780 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
2781 [EXEC_OUTPUT_INHERIT
] = "inherit",
2782 [EXEC_OUTPUT_NULL
] = "null",
2783 [EXEC_OUTPUT_TTY
] = "tty",
2784 [EXEC_OUTPUT_SYSLOG
] = "syslog",
2785 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
2786 [EXEC_OUTPUT_KMSG
] = "kmsg",
2787 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
2788 [EXEC_OUTPUT_JOURNAL
] = "journal",
2789 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
2790 [EXEC_OUTPUT_SOCKET
] = "socket"
2793 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);