1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <sys/capability.h>
11 #include <sys/eventfd.h>
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
16 #include <sys/socket.h>
18 #include <sys/types.h>
24 #include <security/pam_appl.h>
28 #include <selinux/selinux.h>
36 #include <sys/apparmor.h>
39 #include "sd-messages.h"
42 #include "alloc-util.h"
44 #include "apparmor-util.h"
49 #include "capability-util.h"
50 #include "chown-recursive.h"
51 #include "cpu-set-util.h"
54 #include "errno-list.h"
56 #include "exit-status.h"
59 #include "format-util.h"
61 #include "glob-util.h"
70 #include "namespace.h"
71 #include "parse-util.h"
72 #include "path-util.h"
73 #include "process-util.h"
74 #include "rlimit-util.h"
77 #include "seccomp-util.h"
79 #include "securebits.h"
80 #include "securebits-util.h"
81 #include "selinux-util.h"
82 #include "signal-util.h"
83 #include "smack-util.h"
84 #include "socket-util.h"
86 #include "stat-util.h"
87 #include "string-table.h"
88 #include "string-util.h"
90 #include "syslog-util.h"
91 #include "terminal-util.h"
92 #include "umask-util.h"
94 #include "user-util.h"
96 #include "utmp-wtmp.h"
98 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
99 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
101 /* This assumes there is a 'tty' group */
102 #define TTY_MODE 0620
104 #define SNDBUF_SIZE (8*1024*1024)
106 static int shift_fds(int fds
[], size_t n_fds
) {
107 int start
, restart_from
;
112 /* Modifies the fds array! (sorts it) */
122 for (i
= start
; i
< (int) n_fds
; i
++) {
125 /* Already at right index? */
129 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
136 /* Hmm, the fd we wanted isn't free? Then
137 * let's remember that and try again from here */
138 if (nfd
!= i
+3 && restart_from
< 0)
142 if (restart_from
< 0)
145 start
= restart_from
;
151 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
155 n_fds
= n_socket_fds
+ n_storage_fds
;
161 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
162 * O_NONBLOCK only applies to socket activation though. */
164 for (i
= 0; i
< n_fds
; i
++) {
166 if (i
< n_socket_fds
) {
167 r
= fd_nonblock(fds
[i
], nonblock
);
172 /* We unconditionally drop FD_CLOEXEC from the fds,
173 * since after all we want to pass these fds to our
176 r
= fd_cloexec(fds
[i
], false);
184 static const char *exec_context_tty_path(const ExecContext
*context
) {
187 if (context
->stdio_as_fds
)
190 if (context
->tty_path
)
191 return context
->tty_path
;
193 return "/dev/console";
196 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
201 path
= exec_context_tty_path(context
);
203 if (context
->tty_vhangup
) {
204 if (p
&& p
->stdin_fd
>= 0)
205 (void) terminal_vhangup_fd(p
->stdin_fd
);
207 (void) terminal_vhangup(path
);
210 if (context
->tty_reset
) {
211 if (p
&& p
->stdin_fd
>= 0)
212 (void) reset_terminal_fd(p
->stdin_fd
, true);
214 (void) reset_terminal(path
);
217 if (context
->tty_vt_disallocate
&& path
)
218 (void) vt_disallocate(path
);
221 static bool is_terminal_input(ExecInput i
) {
224 EXEC_INPUT_TTY_FORCE
,
225 EXEC_INPUT_TTY_FAIL
);
228 static bool is_terminal_output(ExecOutput o
) {
231 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
232 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
233 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
236 static bool is_syslog_output(ExecOutput o
) {
239 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
242 static bool is_kmsg_output(ExecOutput o
) {
245 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
248 static bool exec_context_needs_term(const ExecContext
*c
) {
251 /* Return true if the execution context suggests we should set $TERM to something useful. */
253 if (is_terminal_input(c
->std_input
))
256 if (is_terminal_output(c
->std_output
))
259 if (is_terminal_output(c
->std_error
))
262 return !!c
->tty_path
;
265 static int open_null_as(int flags
, int nfd
) {
270 fd
= open("/dev/null", flags
|O_NOCTTY
);
274 return move_fd(fd
, nfd
, false);
277 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
278 static const union sockaddr_union sa
= {
279 .un
.sun_family
= AF_UNIX
,
280 .un
.sun_path
= "/run/systemd/journal/stdout",
282 uid_t olduid
= UID_INVALID
;
283 gid_t oldgid
= GID_INVALID
;
286 if (gid_is_valid(gid
)) {
289 if (setegid(gid
) < 0)
293 if (uid_is_valid(uid
)) {
296 if (seteuid(uid
) < 0) {
302 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
304 /* If we fail to restore the uid or gid, things will likely
305 fail later on. This should only happen if an LSM interferes. */
307 if (uid_is_valid(uid
))
308 (void) seteuid(olduid
);
311 if (gid_is_valid(gid
))
312 (void) setegid(oldgid
);
317 static int connect_logger_as(
319 const ExecContext
*context
,
320 const ExecParameters
*params
,
327 _cleanup_close_
int fd
= -1;
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)
347 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
357 context
->syslog_identifier
?: ident
,
358 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
359 context
->syslog_priority
,
360 !!context
->syslog_level_prefix
,
361 is_syslog_output(output
),
362 is_kmsg_output(output
),
363 is_terminal_output(output
)) < 0)
366 return move_fd(TAKE_FD(fd
), nfd
, false);
369 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
375 fd
= open_terminal(path
, flags
| O_NOCTTY
);
379 return move_fd(fd
, nfd
, false);
382 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
383 union sockaddr_union sa
= {};
384 _cleanup_close_
int fd
= -1;
389 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
392 fd
= open(path
, flags
|O_NOCTTY
, mode
);
396 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
398 if (strlen(path
) >= sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
401 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
403 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
407 salen
= sockaddr_un_set_path(&sa
.un
, path
);
411 if (connect(fd
, &sa
.sa
, salen
) < 0)
412 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
413 * indication that his wasn't an AF_UNIX socket after all */
415 if ((flags
& O_ACCMODE
) == O_RDONLY
)
416 r
= shutdown(fd
, SHUT_WR
);
417 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
418 r
= shutdown(fd
, SHUT_RD
);
427 static int fixup_input(
428 const ExecContext
*context
,
430 bool apply_tty_stdin
) {
436 std_input
= context
->std_input
;
438 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
439 return EXEC_INPUT_NULL
;
441 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
442 return EXEC_INPUT_NULL
;
444 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
445 return EXEC_INPUT_NULL
;
450 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
452 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
453 return EXEC_OUTPUT_INHERIT
;
458 static int setup_input(
459 const ExecContext
*context
,
460 const ExecParameters
*params
,
462 int named_iofds
[3]) {
469 if (params
->stdin_fd
>= 0) {
470 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
473 /* Try to make this the controlling tty, if it is a tty, and reset it */
474 if (isatty(STDIN_FILENO
)) {
475 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
476 (void) reset_terminal_fd(STDIN_FILENO
, true);
482 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
486 case EXEC_INPUT_NULL
:
487 return open_null_as(O_RDONLY
, STDIN_FILENO
);
490 case EXEC_INPUT_TTY_FORCE
:
491 case EXEC_INPUT_TTY_FAIL
: {
494 fd
= acquire_terminal(exec_context_tty_path(context
),
495 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
496 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
497 ACQUIRE_TERMINAL_WAIT
,
502 return move_fd(fd
, STDIN_FILENO
, false);
505 case EXEC_INPUT_SOCKET
:
506 assert(socket_fd
>= 0);
508 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
510 case EXEC_INPUT_NAMED_FD
:
511 assert(named_iofds
[STDIN_FILENO
] >= 0);
513 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
514 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
516 case EXEC_INPUT_DATA
: {
519 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
523 return move_fd(fd
, STDIN_FILENO
, false);
526 case EXEC_INPUT_FILE
: {
530 assert(context
->stdio_file
[STDIN_FILENO
]);
532 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
533 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
535 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
539 return move_fd(fd
, STDIN_FILENO
, false);
543 assert_not_reached("Unknown input type");
547 static int setup_output(
549 const ExecContext
*context
,
550 const ExecParameters
*params
,
557 dev_t
*journal_stream_dev
,
558 ino_t
*journal_stream_ino
) {
568 assert(journal_stream_dev
);
569 assert(journal_stream_ino
);
571 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
573 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
576 return STDOUT_FILENO
;
579 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
580 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
583 return STDERR_FILENO
;
586 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
587 o
= fixup_output(context
->std_output
, socket_fd
);
589 if (fileno
== STDERR_FILENO
) {
591 e
= fixup_output(context
->std_error
, socket_fd
);
593 /* This expects the input and output are already set up */
595 /* Don't change the stderr file descriptor if we inherit all
596 * the way and are not on a tty */
597 if (e
== EXEC_OUTPUT_INHERIT
&&
598 o
== EXEC_OUTPUT_INHERIT
&&
599 i
== EXEC_INPUT_NULL
&&
600 !is_terminal_input(context
->std_input
) &&
604 /* Duplicate from stdout if possible */
605 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
606 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
610 } else if (o
== EXEC_OUTPUT_INHERIT
) {
611 /* If input got downgraded, inherit the original value */
612 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
613 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
615 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
616 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
617 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
619 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
623 /* We need to open /dev/null here anew, to get the right access mode. */
624 return open_null_as(O_WRONLY
, fileno
);
629 case EXEC_OUTPUT_NULL
:
630 return open_null_as(O_WRONLY
, fileno
);
632 case EXEC_OUTPUT_TTY
:
633 if (is_terminal_input(i
))
634 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
636 /* We don't reset the terminal if this is just about output */
637 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
639 case EXEC_OUTPUT_SYSLOG
:
640 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
641 case EXEC_OUTPUT_KMSG
:
642 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
643 case EXEC_OUTPUT_JOURNAL
:
644 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
645 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
647 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
648 r
= open_null_as(O_WRONLY
, fileno
);
652 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
653 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
654 * services to detect whether they are connected to the journal or not.
656 * If both stdout and stderr are connected to a stream then let's make sure to store the data
657 * about STDERR as that's usually the best way to do logging. */
659 if (fstat(fileno
, &st
) >= 0 &&
660 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
661 *journal_stream_dev
= st
.st_dev
;
662 *journal_stream_ino
= st
.st_ino
;
667 case EXEC_OUTPUT_SOCKET
:
668 assert(socket_fd
>= 0);
670 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
672 case EXEC_OUTPUT_NAMED_FD
:
673 assert(named_iofds
[fileno
] >= 0);
675 (void) fd_nonblock(named_iofds
[fileno
], false);
676 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
678 case EXEC_OUTPUT_FILE
:
679 case EXEC_OUTPUT_FILE_APPEND
: {
683 assert(context
->stdio_file
[fileno
]);
685 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
686 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
689 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
692 if (o
== EXEC_OUTPUT_FILE_APPEND
)
695 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
700 return move_fd(fd
, fileno
, 0);
704 assert_not_reached("Unknown error type");
708 static int chown_terminal(int fd
, uid_t uid
) {
713 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
717 /* This might fail. What matters are the results. */
718 (void) fchown(fd
, uid
, -1);
719 (void) fchmod(fd
, TTY_MODE
);
721 if (fstat(fd
, &st
) < 0)
724 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
730 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
731 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
734 assert(_saved_stdin
);
735 assert(_saved_stdout
);
737 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
741 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
742 if (saved_stdout
< 0)
745 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
749 r
= chown_terminal(fd
, getuid());
753 r
= reset_terminal_fd(fd
, true);
757 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
762 *_saved_stdin
= saved_stdin
;
763 *_saved_stdout
= saved_stdout
;
765 saved_stdin
= saved_stdout
= -1;
770 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
773 if (err
== -ETIMEDOUT
)
774 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
777 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
781 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
782 _cleanup_close_
int fd
= -1;
786 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
790 write_confirm_error_fd(err
, fd
, u
);
793 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
797 assert(saved_stdout
);
801 if (*saved_stdin
>= 0)
802 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
805 if (*saved_stdout
>= 0)
806 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
809 *saved_stdin
= safe_close(*saved_stdin
);
810 *saved_stdout
= safe_close(*saved_stdout
);
816 CONFIRM_PRETEND_FAILURE
= -1,
817 CONFIRM_PRETEND_SUCCESS
= 0,
821 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
822 int saved_stdout
= -1, saved_stdin
= -1, r
;
823 _cleanup_free_
char *e
= NULL
;
826 /* For any internal errors, assume a positive response. */
827 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
829 write_confirm_error(r
, vc
, u
);
830 return CONFIRM_EXECUTE
;
833 /* confirm_spawn might have been disabled while we were sleeping. */
834 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
839 e
= ellipsize(cmdline
, 60, 100);
847 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
849 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
856 printf("Resuming normal execution.\n");
857 manager_disable_confirm_spawn();
861 unit_dump(u
, stdout
, " ");
862 continue; /* ask again */
864 printf("Failing execution.\n");
865 r
= CONFIRM_PRETEND_FAILURE
;
868 printf(" c - continue, proceed without asking anymore\n"
869 " D - dump, show the state of the unit\n"
870 " f - fail, don't execute the command and pretend it failed\n"
872 " i - info, show a short summary of the unit\n"
873 " j - jobs, show jobs that are in progress\n"
874 " s - skip, don't execute the command and pretend it succeeded\n"
875 " y - yes, execute the command\n");
876 continue; /* ask again */
878 printf(" Description: %s\n"
881 u
->id
, u
->description
, cmdline
);
882 continue; /* ask again */
884 manager_dump_jobs(u
->manager
, stdout
, " ");
885 continue; /* ask again */
887 /* 'n' was removed in favor of 'f'. */
888 printf("Didn't understand 'n', did you mean 'f'?\n");
889 continue; /* ask again */
891 printf("Skipping execution.\n");
892 r
= CONFIRM_PRETEND_SUCCESS
;
898 assert_not_reached("Unhandled choice");
904 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
908 static int get_fixed_user(const ExecContext
*c
, const char **user
,
909 uid_t
*uid
, gid_t
*gid
,
910 const char **home
, const char **shell
) {
919 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
920 * (i.e. are "/" or "/bin/nologin"). */
923 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
931 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
941 r
= get_group_creds(&name
, gid
, 0);
949 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
950 const char *group
, gid_t gid
,
951 gid_t
**supplementary_gids
, int *ngids
) {
955 bool keep_groups
= false;
956 gid_t
*groups
= NULL
;
957 _cleanup_free_ gid_t
*l_gids
= NULL
;
962 * If user is given, then lookup GID and supplementary groups list.
963 * We avoid NSS lookups for gid=0. Also we have to initialize groups
964 * here and as early as possible so we keep the list of supplementary
965 * groups of the caller.
967 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
968 /* First step, initialize groups from /etc/groups */
969 if (initgroups(user
, gid
) < 0)
975 if (strv_isempty(c
->supplementary_groups
))
979 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
980 * be positive, otherwise fail.
983 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
984 if (ngroups_max
<= 0) {
988 return -EOPNOTSUPP
; /* For all other values */
991 l_gids
= new(gid_t
, ngroups_max
);
997 * Lookup the list of groups that the user belongs to, we
998 * avoid NSS lookups here too for gid=0.
1001 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1006 STRV_FOREACH(i
, c
->supplementary_groups
) {
1009 if (k
>= ngroups_max
)
1013 r
= get_group_creds(&g
, l_gids
+k
, 0);
1021 * Sets ngids to zero to drop all supplementary groups, happens
1022 * when we are under root and SupplementaryGroups= is empty.
1029 /* Otherwise get the final list of supplementary groups */
1030 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1034 *supplementary_gids
= groups
;
1042 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1045 /* Handle SupplementaryGroups= if it is not empty */
1047 r
= maybe_setgroups(ngids
, supplementary_gids
);
1052 if (gid_is_valid(gid
)) {
1053 /* Then set our gids */
1054 if (setresgid(gid
, gid
, gid
) < 0)
1061 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1064 if (!uid_is_valid(uid
))
1067 /* Sets (but doesn't look up) the uid and make sure we keep the
1068 * capabilities while doing so. */
1070 if (context
->capability_ambient_set
!= 0) {
1072 /* First step: If we need to keep capabilities but
1073 * drop privileges we need to make sure we keep our
1074 * caps, while we drop privileges. */
1076 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1078 if (prctl(PR_GET_SECUREBITS
) != sb
)
1079 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1084 /* Second step: actually set the uids */
1085 if (setresuid(uid
, uid
, uid
) < 0)
1088 /* At this point we should have all necessary capabilities but
1089 are otherwise a normal user. However, the caps might got
1090 corrupted due to the setresuid() so we need clean them up
1091 later. This is done outside of this call. */
1098 static int null_conv(
1100 const struct pam_message
**msg
,
1101 struct pam_response
**resp
,
1102 void *appdata_ptr
) {
1104 /* We don't support conversations */
1106 return PAM_CONV_ERR
;
1111 static int setup_pam(
1118 int fds
[], size_t n_fds
) {
1122 static const struct pam_conv conv
= {
1127 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1128 pam_handle_t
*handle
= NULL
;
1130 int pam_code
= PAM_SUCCESS
, r
;
1131 char **nv
, **e
= NULL
;
1132 bool close_session
= false;
1133 pid_t pam_pid
= 0, parent_pid
;
1140 /* We set up PAM in the parent process, then fork. The child
1141 * will then stay around until killed via PR_GET_PDEATHSIG or
1142 * systemd via the cgroup logic. It will then remove the PAM
1143 * session again. The parent process will exec() the actual
1144 * daemon. We do things this way to ensure that the main PID
1145 * of the daemon is the one we initially fork()ed. */
1147 r
= barrier_create(&barrier
);
1151 if (log_get_max_level() < LOG_DEBUG
)
1152 flags
|= PAM_SILENT
;
1154 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1155 if (pam_code
!= PAM_SUCCESS
) {
1161 _cleanup_free_
char *q
= NULL
;
1163 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1164 * out if that's the case, and read the TTY off it. */
1166 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1167 tty
= strjoina("/dev/", q
);
1171 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1172 if (pam_code
!= PAM_SUCCESS
)
1176 STRV_FOREACH(nv
, *env
) {
1177 pam_code
= pam_putenv(handle
, *nv
);
1178 if (pam_code
!= PAM_SUCCESS
)
1182 pam_code
= pam_acct_mgmt(handle
, flags
);
1183 if (pam_code
!= PAM_SUCCESS
)
1186 pam_code
= pam_open_session(handle
, flags
);
1187 if (pam_code
!= PAM_SUCCESS
)
1190 close_session
= true;
1192 e
= pam_getenvlist(handle
);
1194 pam_code
= PAM_BUF_ERR
;
1198 /* Block SIGTERM, so that we know that it won't get lost in
1201 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1203 parent_pid
= getpid_cached();
1205 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1209 int sig
, ret
= EXIT_PAM
;
1211 /* The child's job is to reset the PAM session on
1213 barrier_set_role(&barrier
, BARRIER_CHILD
);
1215 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1216 * are open here that have been opened by PAM. */
1217 (void) close_many(fds
, n_fds
);
1219 /* Drop privileges - we don't need any to pam_close_session
1220 * and this will make PR_SET_PDEATHSIG work in most cases.
1221 * If this fails, ignore the error - but expect sd-pam threads
1222 * to fail to exit normally */
1224 r
= maybe_setgroups(0, NULL
);
1226 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1227 if (setresgid(gid
, gid
, gid
) < 0)
1228 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1229 if (setresuid(uid
, uid
, uid
) < 0)
1230 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1232 (void) ignore_signals(SIGPIPE
, -1);
1234 /* Wait until our parent died. This will only work if
1235 * the above setresuid() succeeds, otherwise the kernel
1236 * will not allow unprivileged parents kill their privileged
1237 * children this way. We rely on the control groups kill logic
1238 * to do the rest for us. */
1239 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1242 /* Tell the parent that our setup is done. This is especially
1243 * important regarding dropping privileges. Otherwise, unit
1244 * setup might race against our setresuid(2) call.
1246 * If the parent aborted, we'll detect this below, hence ignore
1247 * return failure here. */
1248 (void) barrier_place(&barrier
);
1250 /* Check if our parent process might already have died? */
1251 if (getppid() == parent_pid
) {
1254 assert_se(sigemptyset(&ss
) >= 0);
1255 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1258 if (sigwait(&ss
, &sig
) < 0) {
1265 assert(sig
== SIGTERM
);
1270 /* If our parent died we'll end the session */
1271 if (getppid() != parent_pid
) {
1272 pam_code
= pam_close_session(handle
, flags
);
1273 if (pam_code
!= PAM_SUCCESS
)
1280 pam_end(handle
, pam_code
| flags
);
1284 barrier_set_role(&barrier
, BARRIER_PARENT
);
1286 /* If the child was forked off successfully it will do all the
1287 * cleanups, so forget about the handle here. */
1290 /* Unblock SIGTERM again in the parent */
1291 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1293 /* We close the log explicitly here, since the PAM modules
1294 * might have opened it, but we don't want this fd around. */
1297 /* Synchronously wait for the child to initialize. We don't care for
1298 * errors as we cannot recover. However, warn loudly if it happens. */
1299 if (!barrier_place_and_sync(&barrier
))
1300 log_error("PAM initialization failed");
1302 return strv_free_and_replace(*env
, e
);
1305 if (pam_code
!= PAM_SUCCESS
) {
1306 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1307 r
= -EPERM
; /* PAM errors do not map to errno */
1309 log_error_errno(r
, "PAM failed: %m");
1313 pam_code
= pam_close_session(handle
, flags
);
1315 pam_end(handle
, pam_code
| flags
);
1327 static void rename_process_from_path(const char *path
) {
1328 char process_name
[11];
1332 /* This resulting string must fit in 10 chars (i.e. the length
1333 * of "/sbin/init") to look pretty in /bin/ps */
1337 rename_process("(...)");
1343 /* The end of the process name is usually more
1344 * interesting, since the first bit might just be
1350 process_name
[0] = '(';
1351 memcpy(process_name
+1, p
, l
);
1352 process_name
[1+l
] = ')';
1353 process_name
[1+l
+1] = 0;
1355 rename_process(process_name
);
1358 static bool context_has_address_families(const ExecContext
*c
) {
1361 return c
->address_families_whitelist
||
1362 !set_isempty(c
->address_families
);
1365 static bool context_has_syscall_filters(const ExecContext
*c
) {
1368 return c
->syscall_whitelist
||
1369 !hashmap_isempty(c
->syscall_filter
);
1372 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1375 if (c
->no_new_privileges
)
1378 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1381 /* We need NNP if we have any form of seccomp and are unprivileged */
1382 return context_has_address_families(c
) ||
1383 c
->memory_deny_write_execute
||
1384 c
->restrict_realtime
||
1385 exec_context_restrict_namespaces_set(c
) ||
1386 c
->protect_kernel_tunables
||
1387 c
->protect_kernel_modules
||
1388 c
->private_devices
||
1389 context_has_syscall_filters(c
) ||
1390 !set_isempty(c
->syscall_archs
) ||
1391 c
->lock_personality
;
1396 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1398 if (is_seccomp_available())
1401 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1405 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1406 uint32_t negative_action
, default_action
, action
;
1412 if (!context_has_syscall_filters(c
))
1415 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1418 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1420 if (c
->syscall_whitelist
) {
1421 default_action
= negative_action
;
1422 action
= SCMP_ACT_ALLOW
;
1424 default_action
= SCMP_ACT_ALLOW
;
1425 action
= negative_action
;
1428 if (needs_ambient_hack
) {
1429 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1434 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1437 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1441 if (set_isempty(c
->syscall_archs
))
1444 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1447 return seccomp_restrict_archs(c
->syscall_archs
);
1450 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1454 if (!context_has_address_families(c
))
1457 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1460 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1463 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1467 if (!c
->memory_deny_write_execute
)
1470 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1473 return seccomp_memory_deny_write_execute();
1476 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1480 if (!c
->restrict_realtime
)
1483 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1486 return seccomp_restrict_realtime();
1489 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1493 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1494 * let's protect even those systems where this is left on in the kernel. */
1496 if (!c
->protect_kernel_tunables
)
1499 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1502 return seccomp_protect_sysctl();
1505 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1509 /* Turn off module syscalls on ProtectKernelModules=yes */
1511 if (!c
->protect_kernel_modules
)
1514 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1517 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1520 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1524 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1526 if (!c
->private_devices
)
1529 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1532 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1535 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1539 if (!exec_context_restrict_namespaces_set(c
))
1542 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1545 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1548 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1549 unsigned long personality
;
1555 if (!c
->lock_personality
)
1558 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1561 personality
= c
->personality
;
1563 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1564 if (personality
== PERSONALITY_INVALID
) {
1566 r
= opinionated_personality(&personality
);
1571 return seccomp_lock_personality(personality
);
1576 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1579 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1580 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1582 if (idle_pipe
[0] >= 0) {
1585 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1587 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1590 /* Signal systemd that we are bored and want to continue. */
1591 n
= write(idle_pipe
[3], "x", 1);
1593 /* Wait for systemd to react to the signal above. */
1594 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1597 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1601 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1604 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1606 static int build_environment(
1608 const ExecContext
*c
,
1609 const ExecParameters
*p
,
1612 const char *username
,
1614 dev_t journal_stream_dev
,
1615 ino_t journal_stream_ino
,
1618 _cleanup_strv_free_
char **our_env
= NULL
;
1619 ExecDirectoryType t
;
1628 our_env
= new0(char*, 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1633 _cleanup_free_
char *joined
= NULL
;
1635 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1637 our_env
[n_env
++] = x
;
1639 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1641 our_env
[n_env
++] = x
;
1643 joined
= strv_join(p
->fd_names
, ":");
1647 x
= strjoin("LISTEN_FDNAMES=", joined
);
1650 our_env
[n_env
++] = x
;
1653 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1654 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1656 our_env
[n_env
++] = x
;
1658 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1660 our_env
[n_env
++] = x
;
1663 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1664 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1665 * check the database directly. */
1666 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1667 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1670 our_env
[n_env
++] = x
;
1674 x
= strappend("HOME=", home
);
1677 our_env
[n_env
++] = x
;
1681 x
= strappend("LOGNAME=", username
);
1684 our_env
[n_env
++] = x
;
1686 x
= strappend("USER=", username
);
1689 our_env
[n_env
++] = x
;
1693 x
= strappend("SHELL=", shell
);
1696 our_env
[n_env
++] = x
;
1699 if (!sd_id128_is_null(u
->invocation_id
)) {
1700 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1703 our_env
[n_env
++] = x
;
1706 if (exec_context_needs_term(c
)) {
1707 const char *tty_path
, *term
= NULL
;
1709 tty_path
= exec_context_tty_path(c
);
1711 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1712 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1713 * passes to PID 1 ends up all the way in the console login shown. */
1715 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1716 term
= getenv("TERM");
1718 term
= default_term_for_tty(tty_path
);
1720 x
= strappend("TERM=", term
);
1723 our_env
[n_env
++] = x
;
1726 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1727 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1730 our_env
[n_env
++] = x
;
1733 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1734 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1740 if (strv_isempty(c
->directories
[t
].paths
))
1743 n
= exec_directory_env_name_to_string(t
);
1747 pre
= strjoin(p
->prefix
[t
], "/");
1751 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1755 x
= strjoin(n
, "=", joined
);
1759 our_env
[n_env
++] = x
;
1762 our_env
[n_env
++] = NULL
;
1763 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1765 *ret
= TAKE_PTR(our_env
);
1770 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1771 _cleanup_strv_free_
char **pass_env
= NULL
;
1772 size_t n_env
= 0, n_bufsize
= 0;
1775 STRV_FOREACH(i
, c
->pass_environment
) {
1776 _cleanup_free_
char *x
= NULL
;
1782 x
= strjoin(*i
, "=", v
);
1786 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1789 pass_env
[n_env
++] = TAKE_PTR(x
);
1790 pass_env
[n_env
] = NULL
;
1793 *ret
= TAKE_PTR(pass_env
);
1798 static bool exec_needs_mount_namespace(
1799 const ExecContext
*context
,
1800 const ExecParameters
*params
,
1801 const ExecRuntime
*runtime
) {
1806 if (context
->root_image
)
1809 if (!strv_isempty(context
->read_write_paths
) ||
1810 !strv_isempty(context
->read_only_paths
) ||
1811 !strv_isempty(context
->inaccessible_paths
))
1814 if (context
->n_bind_mounts
> 0)
1817 if (context
->n_temporary_filesystems
> 0)
1820 if (context
->mount_flags
!= 0)
1823 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1826 if (context
->private_devices
||
1827 context
->private_mounts
||
1828 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1829 context
->protect_home
!= PROTECT_HOME_NO
||
1830 context
->protect_kernel_tunables
||
1831 context
->protect_kernel_modules
||
1832 context
->protect_control_groups
)
1835 if (context
->root_directory
) {
1836 ExecDirectoryType t
;
1838 if (context
->mount_apivfs
)
1841 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1842 if (!params
->prefix
[t
])
1845 if (!strv_isempty(context
->directories
[t
].paths
))
1850 if (context
->dynamic_user
&&
1851 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1852 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1853 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1859 static int setup_private_users(uid_t uid
, gid_t gid
) {
1860 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1861 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1862 _cleanup_close_
int unshare_ready_fd
= -1;
1863 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1868 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1869 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1870 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1871 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1872 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1873 * continues execution normally. */
1875 if (uid
!= 0 && uid_is_valid(uid
)) {
1876 r
= asprintf(&uid_map
,
1877 "0 0 1\n" /* Map root → root */
1878 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1883 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1888 if (gid
!= 0 && gid_is_valid(gid
)) {
1889 r
= asprintf(&gid_map
,
1890 "0 0 1\n" /* Map root → root */
1891 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1896 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1901 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1903 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1904 if (unshare_ready_fd
< 0)
1907 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1909 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1912 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1916 _cleanup_close_
int fd
= -1;
1920 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1921 * here, after the parent opened its own user namespace. */
1924 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1926 /* Wait until the parent unshared the user namespace */
1927 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1932 /* Disable the setgroups() system call in the child user namespace, for good. */
1933 a
= procfs_file_alloca(ppid
, "setgroups");
1934 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1936 if (errno
!= ENOENT
) {
1941 /* If the file is missing the kernel is too old, let's continue anyway. */
1943 if (write(fd
, "deny\n", 5) < 0) {
1948 fd
= safe_close(fd
);
1951 /* First write the GID map */
1952 a
= procfs_file_alloca(ppid
, "gid_map");
1953 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1958 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1962 fd
= safe_close(fd
);
1964 /* The write the UID map */
1965 a
= procfs_file_alloca(ppid
, "uid_map");
1966 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1971 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1976 _exit(EXIT_SUCCESS
);
1979 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1980 _exit(EXIT_FAILURE
);
1983 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1985 if (unshare(CLONE_NEWUSER
) < 0)
1988 /* Let the child know that the namespace is ready now */
1989 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1992 /* Try to read an error code from the child */
1993 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1996 if (n
== sizeof(r
)) { /* an error code was sent to us */
2001 if (n
!= 0) /* on success we should have read 0 bytes */
2004 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2008 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2014 static int setup_exec_directory(
2015 const ExecContext
*context
,
2016 const ExecParameters
*params
,
2019 ExecDirectoryType type
,
2022 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2023 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2024 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2025 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2026 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2027 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2034 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2035 assert(exit_status
);
2037 if (!params
->prefix
[type
])
2040 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2041 if (!uid_is_valid(uid
))
2043 if (!gid_is_valid(gid
))
2047 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2048 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2050 p
= strjoin(params
->prefix
[type
], "/", *rt
);
2056 r
= mkdir_parents_label(p
, 0755);
2060 if (context
->dynamic_user
&&
2061 !IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2062 _cleanup_free_
char *private_root
= NULL
;
2064 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2065 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2066 * whose UID is later on reused. To lock this down we use the same trick used by container
2067 * managers to prohibit host users to get access to files of the same UID in containers: we
2068 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2069 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2070 * to make this directory permeable for the service itself.
2072 * Specifically: for a service which wants a special directory "foo/" we first create a
2073 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2074 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2075 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2076 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2077 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2078 * disabling the access boundary for the service and making sure it only gets access to the
2079 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2081 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2082 * owned by the service itself.
2083 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2084 * files or sockets with other services. */
2086 private_root
= strjoin(params
->prefix
[type
], "/private");
2087 if (!private_root
) {
2092 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2093 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2097 pp
= strjoin(private_root
, "/", *rt
);
2103 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2104 r
= mkdir_parents_label(pp
, 0755);
2108 if (is_dir(p
, false) > 0 &&
2109 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2111 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2112 * it over. Most likely the service has been upgraded from one that didn't use
2113 * DynamicUser=1, to one that does. */
2115 if (rename(p
, pp
) < 0) {
2120 /* Otherwise, create the actual directory for the service */
2122 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2123 if (r
< 0 && r
!= -EEXIST
)
2127 /* And link it up from the original place */
2128 r
= symlink_idempotent(pp
, p
, true);
2132 /* Lock down the access mode */
2133 if (chmod(pp
, context
->directories
[type
].mode
) < 0) {
2138 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2139 if (r
< 0 && r
!= -EEXIST
)
2141 if (r
== -EEXIST
&& !context
->dynamic_user
)
2145 /* Don't change the owner of the configuration directory, as in the common case it is not written to by
2146 * a service, and shall not be writable. */
2147 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2150 /* Then, change the ownership of the whole tree, if necessary */
2151 r
= path_chown_recursive(pp
?: p
, uid
, gid
);
2159 *exit_status
= exit_status_table
[type
];
2164 static int setup_smack(
2165 const ExecContext
*context
,
2166 const ExecCommand
*command
) {
2173 if (context
->smack_process_label
) {
2174 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2178 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2180 _cleanup_free_
char *exec_label
= NULL
;
2182 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2183 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2186 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2196 static int compile_bind_mounts(
2197 const ExecContext
*context
,
2198 const ExecParameters
*params
,
2199 BindMount
**ret_bind_mounts
,
2200 size_t *ret_n_bind_mounts
,
2201 char ***ret_empty_directories
) {
2203 _cleanup_strv_free_
char **empty_directories
= NULL
;
2204 BindMount
*bind_mounts
;
2206 ExecDirectoryType t
;
2211 assert(ret_bind_mounts
);
2212 assert(ret_n_bind_mounts
);
2213 assert(ret_empty_directories
);
2215 n
= context
->n_bind_mounts
;
2216 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2217 if (!params
->prefix
[t
])
2220 n
+= strv_length(context
->directories
[t
].paths
);
2224 *ret_bind_mounts
= NULL
;
2225 *ret_n_bind_mounts
= 0;
2226 *ret_empty_directories
= NULL
;
2230 bind_mounts
= new(BindMount
, n
);
2234 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2235 BindMount
*item
= context
->bind_mounts
+ i
;
2238 s
= strdup(item
->source
);
2244 d
= strdup(item
->destination
);
2251 bind_mounts
[h
++] = (BindMount
) {
2254 .read_only
= item
->read_only
,
2255 .recursive
= item
->recursive
,
2256 .ignore_enoent
= item
->ignore_enoent
,
2260 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2263 if (!params
->prefix
[t
])
2266 if (strv_isempty(context
->directories
[t
].paths
))
2269 if (context
->dynamic_user
&&
2270 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2271 !(context
->root_directory
|| context
->root_image
)) {
2274 /* So this is for a dynamic user, and we need to make sure the process can access its own
2275 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2276 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2278 private_root
= strjoin(params
->prefix
[t
], "/private");
2279 if (!private_root
) {
2284 r
= strv_consume(&empty_directories
, private_root
);
2289 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2292 if (context
->dynamic_user
&&
2293 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2294 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2296 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2302 if (context
->dynamic_user
&&
2303 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2304 (context
->root_directory
|| context
->root_image
))
2305 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2306 * directory is not created on the root directory. So, let's bind-mount the directory
2307 * on the 'non-private' place. */
2308 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2317 bind_mounts
[h
++] = (BindMount
) {
2322 .ignore_enoent
= false,
2329 *ret_bind_mounts
= bind_mounts
;
2330 *ret_n_bind_mounts
= n
;
2331 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2336 bind_mount_free_many(bind_mounts
, h
);
2340 static int apply_mount_namespace(
2342 const ExecCommand
*command
,
2343 const ExecContext
*context
,
2344 const ExecParameters
*params
,
2345 const ExecRuntime
*runtime
) {
2347 _cleanup_strv_free_
char **empty_directories
= NULL
;
2348 char *tmp
= NULL
, *var
= NULL
;
2349 const char *root_dir
= NULL
, *root_image
= NULL
;
2350 NamespaceInfo ns_info
;
2351 bool needs_sandboxing
;
2352 BindMount
*bind_mounts
= NULL
;
2353 size_t n_bind_mounts
= 0;
2358 /* The runtime struct only contains the parent of the private /tmp,
2359 * which is non-accessible to world users. Inside of it there's a /tmp
2360 * that is sticky, and that's the one we want to use here. */
2362 if (context
->private_tmp
&& runtime
) {
2363 if (runtime
->tmp_dir
)
2364 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2365 if (runtime
->var_tmp_dir
)
2366 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2369 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2370 root_image
= context
->root_image
;
2373 root_dir
= context
->root_directory
;
2376 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2380 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2381 if (needs_sandboxing
)
2382 ns_info
= (NamespaceInfo
) {
2383 .ignore_protect_paths
= false,
2384 .private_dev
= context
->private_devices
,
2385 .protect_control_groups
= context
->protect_control_groups
,
2386 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2387 .protect_kernel_modules
= context
->protect_kernel_modules
,
2388 .mount_apivfs
= context
->mount_apivfs
,
2389 .private_mounts
= context
->private_mounts
,
2391 else if (!context
->dynamic_user
&& root_dir
)
2393 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2394 * sandbox info, otherwise enforce it, don't ignore protected paths and
2395 * fail if we are enable to apply the sandbox inside the mount namespace.
2397 ns_info
= (NamespaceInfo
) {
2398 .ignore_protect_paths
= true,
2401 ns_info
= (NamespaceInfo
) {};
2403 r
= setup_namespace(root_dir
, root_image
,
2404 &ns_info
, context
->read_write_paths
,
2405 needs_sandboxing
? context
->read_only_paths
: NULL
,
2406 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2410 context
->temporary_filesystems
,
2411 context
->n_temporary_filesystems
,
2414 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2415 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2416 context
->mount_flags
,
2417 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2419 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2421 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2422 * that with a special, recognizable error ENOANO. In this case, silently proceeed, but only if exclusively
2423 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2424 * completely different execution environment. */
2426 if (n_bind_mounts
== 0 &&
2427 context
->n_temporary_filesystems
== 0 &&
2428 !root_dir
&& !root_image
&&
2429 !context
->dynamic_user
) {
2430 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2434 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2435 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2436 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2444 static int apply_working_directory(
2445 const ExecContext
*context
,
2446 const ExecParameters
*params
,
2448 const bool needs_mount_ns
,
2454 assert(exit_status
);
2456 if (context
->working_directory_home
) {
2459 *exit_status
= EXIT_CHDIR
;
2465 } else if (context
->working_directory
)
2466 wd
= context
->working_directory
;
2470 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2471 if (!needs_mount_ns
&& context
->root_directory
)
2472 if (chroot(context
->root_directory
) < 0) {
2473 *exit_status
= EXIT_CHROOT
;
2479 d
= prefix_roota(context
->root_directory
, wd
);
2481 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2482 *exit_status
= EXIT_CHDIR
;
2489 static int setup_keyring(
2491 const ExecContext
*context
,
2492 const ExecParameters
*p
,
2493 uid_t uid
, gid_t gid
) {
2495 key_serial_t keyring
;
2504 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2505 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2506 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2507 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2508 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2509 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2511 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2514 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2517 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2518 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2519 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2520 * & group is just as nasty as acquiring a reference to the user keyring. */
2522 saved_uid
= getuid();
2523 saved_gid
= getgid();
2525 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2526 if (setregid(gid
, -1) < 0)
2527 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2530 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2531 if (setreuid(uid
, -1) < 0) {
2532 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2537 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2538 if (keyring
== -1) {
2539 if (errno
== ENOSYS
)
2540 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2541 else if (IN_SET(errno
, EACCES
, EPERM
))
2542 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2543 else if (errno
== EDQUOT
)
2544 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2546 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2551 /* When requested link the user keyring into the session keyring. */
2552 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2554 if (keyctl(KEYCTL_LINK
,
2555 KEY_SPEC_USER_KEYRING
,
2556 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2557 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2562 /* Restore uid/gid back */
2563 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2564 if (setreuid(saved_uid
, -1) < 0) {
2565 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2570 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2571 if (setregid(saved_gid
, -1) < 0)
2572 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2575 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2576 if (!sd_id128_is_null(u
->invocation_id
)) {
2579 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2581 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2583 if (keyctl(KEYCTL_SETPERM
, key
,
2584 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2585 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2586 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2591 /* Revert back uid & gid for the the last time, and exit */
2592 /* no extra logging, as only the first already reported error matters */
2593 if (getuid() != saved_uid
)
2594 (void) setreuid(saved_uid
, -1);
2596 if (getgid() != saved_gid
)
2597 (void) setregid(saved_gid
, -1);
2602 static void append_socket_pair(int *array
, size_t *n
, const int pair
[2]) {
2610 array
[(*n
)++] = pair
[0];
2612 array
[(*n
)++] = pair
[1];
2615 static int close_remaining_fds(
2616 const ExecParameters
*params
,
2617 const ExecRuntime
*runtime
,
2618 const DynamicCreds
*dcreds
,
2622 int *fds
, size_t n_fds
) {
2624 size_t n_dont_close
= 0;
2625 int dont_close
[n_fds
+ 12];
2629 if (params
->stdin_fd
>= 0)
2630 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2631 if (params
->stdout_fd
>= 0)
2632 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2633 if (params
->stderr_fd
>= 0)
2634 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2637 dont_close
[n_dont_close
++] = socket_fd
;
2639 dont_close
[n_dont_close
++] = exec_fd
;
2641 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2642 n_dont_close
+= n_fds
;
2646 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2650 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2652 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2655 if (user_lookup_fd
>= 0)
2656 dont_close
[n_dont_close
++] = user_lookup_fd
;
2658 return close_all_fds(dont_close
, n_dont_close
);
2661 static int send_user_lookup(
2669 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2670 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2673 if (user_lookup_fd
< 0)
2676 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2679 if (writev(user_lookup_fd
,
2681 IOVEC_INIT(&uid
, sizeof(uid
)),
2682 IOVEC_INIT(&gid
, sizeof(gid
)),
2683 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2689 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2696 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2701 if (!c
->working_directory_home
)
2705 /* Hardcode /root as home directory for UID 0 */
2710 r
= get_home_dir(buf
);
2718 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2719 _cleanup_strv_free_
char ** list
= NULL
;
2720 ExecDirectoryType t
;
2727 assert(c
->dynamic_user
);
2729 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2730 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2733 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2736 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2742 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2745 if (t
== EXEC_DIRECTORY_RUNTIME
)
2746 e
= strjoin(p
->prefix
[t
], "/", *i
);
2748 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2752 r
= strv_consume(&list
, e
);
2758 *ret
= TAKE_PTR(list
);
2763 static char *exec_command_line(char **argv
);
2765 static int exec_child(
2767 const ExecCommand
*command
,
2768 const ExecContext
*context
,
2769 const ExecParameters
*params
,
2770 ExecRuntime
*runtime
,
2771 DynamicCreds
*dcreds
,
2775 size_t n_socket_fds
,
2776 size_t n_storage_fds
,
2781 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2782 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2783 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2784 const char *username
= NULL
, *groupname
= NULL
;
2785 _cleanup_free_
char *home_buffer
= NULL
;
2786 const char *home
= NULL
, *shell
= NULL
;
2787 dev_t journal_stream_dev
= 0;
2788 ino_t journal_stream_ino
= 0;
2789 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2790 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2791 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2792 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2794 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2795 bool use_selinux
= false;
2798 bool use_smack
= false;
2801 bool use_apparmor
= false;
2803 uid_t uid
= UID_INVALID
;
2804 gid_t gid
= GID_INVALID
;
2806 ExecDirectoryType dt
;
2813 assert(exit_status
);
2815 rename_process_from_path(command
->path
);
2817 /* We reset exactly these signals, since they are the
2818 * only ones we set to SIG_IGN in the main daemon. All
2819 * others we leave untouched because we set them to
2820 * SIG_DFL or a valid handler initially, both of which
2821 * will be demoted to SIG_DFL. */
2822 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2823 SIGNALS_IGNORE
, -1);
2825 if (context
->ignore_sigpipe
)
2826 (void) ignore_signals(SIGPIPE
, -1);
2828 r
= reset_signal_mask();
2830 *exit_status
= EXIT_SIGNAL_MASK
;
2831 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2834 if (params
->idle_pipe
)
2835 do_idle_pipe_dance(params
->idle_pipe
);
2837 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2838 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2839 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2840 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2843 log_set_open_when_needed(true);
2845 /* In case anything used libc syslog(), close this here, too */
2848 n_fds
= n_socket_fds
+ n_storage_fds
;
2849 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2851 *exit_status
= EXIT_FDS
;
2852 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2855 if (!context
->same_pgrp
)
2857 *exit_status
= EXIT_SETSID
;
2858 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2861 exec_context_tty_reset(context
, params
);
2863 if (unit_shall_confirm_spawn(unit
)) {
2864 const char *vc
= params
->confirm_spawn
;
2865 _cleanup_free_
char *cmdline
= NULL
;
2867 cmdline
= exec_command_line(command
->argv
);
2869 *exit_status
= EXIT_MEMORY
;
2873 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2874 if (r
!= CONFIRM_EXECUTE
) {
2875 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2876 *exit_status
= EXIT_SUCCESS
;
2879 *exit_status
= EXIT_CONFIRM
;
2880 log_unit_error(unit
, "Execution cancelled by the user");
2885 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2886 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2887 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2888 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2889 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2890 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2891 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2892 *exit_status
= EXIT_MEMORY
;
2893 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2896 if (context
->dynamic_user
&& dcreds
) {
2897 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2899 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2900 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2901 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2902 *exit_status
= EXIT_USER
;
2903 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2906 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2908 *exit_status
= EXIT_MEMORY
;
2912 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2914 *exit_status
= EXIT_USER
;
2916 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2919 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2922 if (!uid_is_valid(uid
)) {
2923 *exit_status
= EXIT_USER
;
2924 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
2928 if (!gid_is_valid(gid
)) {
2929 *exit_status
= EXIT_USER
;
2930 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
2935 username
= dcreds
->user
->name
;
2938 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2940 *exit_status
= EXIT_USER
;
2941 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
2944 r
= get_fixed_group(context
, &groupname
, &gid
);
2946 *exit_status
= EXIT_GROUP
;
2947 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
2951 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2952 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2953 &supplementary_gids
, &ngids
);
2955 *exit_status
= EXIT_GROUP
;
2956 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
2959 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2961 *exit_status
= EXIT_USER
;
2962 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
2965 user_lookup_fd
= safe_close(user_lookup_fd
);
2967 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2969 *exit_status
= EXIT_CHDIR
;
2970 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
2973 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2974 * must sure to drop O_NONBLOCK */
2976 (void) fd_nonblock(socket_fd
, false);
2978 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2980 *exit_status
= EXIT_STDIN
;
2981 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
2984 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2986 *exit_status
= EXIT_STDOUT
;
2987 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
2990 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2992 *exit_status
= EXIT_STDERR
;
2993 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
2996 if (params
->cgroup_path
) {
2997 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2999 *exit_status
= EXIT_CGROUP
;
3000 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", params
->cgroup_path
);
3004 if (context
->oom_score_adjust_set
) {
3005 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3006 * prohibit write access to this file, and we shouldn't trip up over that. */
3007 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3008 if (IN_SET(r
, -EPERM
, -EACCES
))
3009 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3011 *exit_status
= EXIT_OOM_ADJUST
;
3012 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3016 if (context
->nice_set
)
3017 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3018 *exit_status
= EXIT_NICE
;
3019 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3022 if (context
->cpu_sched_set
) {
3023 struct sched_param param
= {
3024 .sched_priority
= context
->cpu_sched_priority
,
3027 r
= sched_setscheduler(0,
3028 context
->cpu_sched_policy
|
3029 (context
->cpu_sched_reset_on_fork
?
3030 SCHED_RESET_ON_FORK
: 0),
3033 *exit_status
= EXIT_SETSCHEDULER
;
3034 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3038 if (context
->cpuset
)
3039 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3040 *exit_status
= EXIT_CPUAFFINITY
;
3041 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3044 if (context
->ioprio_set
)
3045 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3046 *exit_status
= EXIT_IOPRIO
;
3047 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3050 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3051 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3052 *exit_status
= EXIT_TIMERSLACK
;
3053 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3056 if (context
->personality
!= PERSONALITY_INVALID
) {
3057 r
= safe_personality(context
->personality
);
3059 *exit_status
= EXIT_PERSONALITY
;
3060 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3064 if (context
->utmp_id
)
3065 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3067 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3068 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3072 if (context
->user
) {
3073 r
= chown_terminal(STDIN_FILENO
, uid
);
3075 *exit_status
= EXIT_STDIN
;
3076 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3080 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroupsv1
3081 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3082 * safe. On cgroupsv2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3083 * touch a single hierarchy too. */
3084 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3085 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3087 *exit_status
= EXIT_CGROUP
;
3088 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3092 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3093 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3095 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3098 r
= build_environment(
3110 *exit_status
= EXIT_MEMORY
;
3114 r
= build_pass_environment(context
, &pass_env
);
3116 *exit_status
= EXIT_MEMORY
;
3120 accum_env
= strv_env_merge(5,
3121 params
->environment
,
3124 context
->environment
,
3128 *exit_status
= EXIT_MEMORY
;
3131 accum_env
= strv_env_clean(accum_env
);
3133 (void) umask(context
->umask
);
3135 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3137 *exit_status
= EXIT_KEYRING
;
3138 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3141 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3142 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3144 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked for it, and the kernel doesn't actually support ambient caps */
3145 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3147 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not desired */
3148 if (needs_ambient_hack
)
3149 needs_setuid
= false;
3151 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3153 if (needs_sandboxing
) {
3154 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3155 * present. The actual MAC context application will happen later, as late as possible, to avoid
3156 * impacting our own code paths. */
3159 use_selinux
= mac_selinux_use();
3162 use_smack
= mac_smack_use();
3165 use_apparmor
= mac_apparmor_use();
3170 if (context
->pam_name
&& username
) {
3171 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3173 *exit_status
= EXIT_PAM
;
3174 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3179 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3180 if (ns_type_supported(NAMESPACE_NET
)) {
3181 r
= setup_netns(runtime
->netns_storage_socket
);
3183 *exit_status
= EXIT_NETWORK
;
3184 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3187 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3190 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3191 if (needs_mount_namespace
) {
3192 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3194 *exit_status
= EXIT_NAMESPACE
;
3195 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3199 /* Apply just after mount namespace setup */
3200 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3202 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3204 /* Drop groups as early as possbile */
3206 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3208 *exit_status
= EXIT_GROUP
;
3209 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3213 if (needs_sandboxing
) {
3215 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3216 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3218 *exit_status
= EXIT_SELINUX_CONTEXT
;
3219 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3224 if (context
->private_users
) {
3225 r
= setup_private_users(uid
, gid
);
3227 *exit_status
= EXIT_USER
;
3228 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3233 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3234 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3235 * however if we have it as we want to keep it open until the final execve(). */
3237 if (params
->exec_fd
>= 0) {
3238 exec_fd
= params
->exec_fd
;
3240 if (exec_fd
< 3 + (int) n_fds
) {
3243 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3244 * process we are about to execute. */
3246 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3248 *exit_status
= EXIT_FDS
;
3249 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3252 safe_close(exec_fd
);
3255 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3256 r
= fd_cloexec(exec_fd
, true);
3258 *exit_status
= EXIT_FDS
;
3259 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3263 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3264 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3265 fds_with_exec_fd
[n_fds
] = exec_fd
;
3266 n_fds_with_exec_fd
= n_fds
+ 1;
3268 fds_with_exec_fd
= fds
;
3269 n_fds_with_exec_fd
= n_fds
;
3272 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3274 r
= shift_fds(fds
, n_fds
);
3276 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3278 *exit_status
= EXIT_FDS
;
3279 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3282 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3283 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3284 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3287 secure_bits
= context
->secure_bits
;
3289 if (needs_sandboxing
) {
3293 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3295 *exit_status
= EXIT_LIMITS
;
3296 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3299 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
3300 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3301 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3302 *exit_status
= EXIT_LIMITS
;
3303 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3308 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3309 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3311 r
= setup_smack(context
, command
);
3313 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3314 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3319 bset
= context
->capability_bounding_set
;
3320 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3321 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3322 * instead of us doing that */
3323 if (needs_ambient_hack
)
3324 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3325 (UINT64_C(1) << CAP_SETUID
) |
3326 (UINT64_C(1) << CAP_SETGID
);
3328 if (!cap_test_all(bset
)) {
3329 r
= capability_bounding_set_drop(bset
, false);
3331 *exit_status
= EXIT_CAPABILITIES
;
3332 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3336 /* This is done before enforce_user, but ambient set
3337 * does not survive over setresuid() if keep_caps is not set. */
3338 if (!needs_ambient_hack
&&
3339 context
->capability_ambient_set
!= 0) {
3340 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3342 *exit_status
= EXIT_CAPABILITIES
;
3343 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3349 if (context
->user
) {
3350 r
= enforce_user(context
, uid
);
3352 *exit_status
= EXIT_USER
;
3353 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3356 if (!needs_ambient_hack
&&
3357 context
->capability_ambient_set
!= 0) {
3359 /* Fix the ambient capabilities after user change. */
3360 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3362 *exit_status
= EXIT_CAPABILITIES
;
3363 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3366 /* If we were asked to change user and ambient capabilities
3367 * were requested, we had to add keep-caps to the securebits
3368 * so that we would maintain the inherited capability set
3369 * through the setresuid(). Make sure that the bit is added
3370 * also to the context secure_bits so that we don't try to
3371 * drop the bit away next. */
3373 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3378 if (needs_sandboxing
) {
3379 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3380 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3381 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3382 * are restricted. */
3386 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3389 r
= setexeccon(exec_context
);
3391 *exit_status
= EXIT_SELINUX_CONTEXT
;
3392 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3399 if (use_apparmor
&& context
->apparmor_profile
) {
3400 r
= aa_change_onexec(context
->apparmor_profile
);
3401 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3402 *exit_status
= EXIT_APPARMOR_PROFILE
;
3403 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3408 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3409 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3410 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3411 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3412 *exit_status
= EXIT_SECUREBITS
;
3413 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3416 if (context_has_no_new_privileges(context
))
3417 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3418 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3419 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3423 r
= apply_address_families(unit
, context
);
3425 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3426 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3429 r
= apply_memory_deny_write_execute(unit
, context
);
3431 *exit_status
= EXIT_SECCOMP
;
3432 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3435 r
= apply_restrict_realtime(unit
, context
);
3437 *exit_status
= EXIT_SECCOMP
;
3438 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3441 r
= apply_restrict_namespaces(unit
, context
);
3443 *exit_status
= EXIT_SECCOMP
;
3444 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3447 r
= apply_protect_sysctl(unit
, context
);
3449 *exit_status
= EXIT_SECCOMP
;
3450 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3453 r
= apply_protect_kernel_modules(unit
, context
);
3455 *exit_status
= EXIT_SECCOMP
;
3456 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3459 r
= apply_private_devices(unit
, context
);
3461 *exit_status
= EXIT_SECCOMP
;
3462 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3465 r
= apply_syscall_archs(unit
, context
);
3467 *exit_status
= EXIT_SECCOMP
;
3468 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3471 r
= apply_lock_personality(unit
, context
);
3473 *exit_status
= EXIT_SECCOMP
;
3474 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3477 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3478 * by the filter as little as possible. */
3479 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3481 *exit_status
= EXIT_SECCOMP
;
3482 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3487 if (!strv_isempty(context
->unset_environment
)) {
3490 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3492 *exit_status
= EXIT_MEMORY
;
3496 strv_free_and_replace(accum_env
, ee
);
3499 final_argv
= replace_env_argv(command
->argv
, accum_env
);
3501 *exit_status
= EXIT_MEMORY
;
3505 if (DEBUG_LOGGING
) {
3506 _cleanup_free_
char *line
;
3508 line
= exec_command_line(final_argv
);
3510 log_struct(LOG_DEBUG
,
3511 "EXECUTABLE=%s", command
->path
,
3512 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3514 LOG_UNIT_INVOCATION_ID(unit
));
3520 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3521 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3523 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3524 *exit_status
= EXIT_EXEC
;
3525 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3529 execve(command
->path
, final_argv
, accum_env
);
3535 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3536 * that POLLHUP on it no longer means execve() succeeded. */
3538 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3539 *exit_status
= EXIT_EXEC
;
3540 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3544 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3545 log_struct_errno(LOG_INFO
, r
,
3546 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3548 LOG_UNIT_INVOCATION_ID(unit
),
3549 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3551 "EXECUTABLE=%s", command
->path
);
3555 *exit_status
= EXIT_EXEC
;
3556 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3559 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3560 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3562 int exec_spawn(Unit
*unit
,
3563 ExecCommand
*command
,
3564 const ExecContext
*context
,
3565 const ExecParameters
*params
,
3566 ExecRuntime
*runtime
,
3567 DynamicCreds
*dcreds
,
3570 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3571 _cleanup_strv_free_
char **files_env
= NULL
;
3572 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3573 _cleanup_free_
char *line
= NULL
;
3581 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3583 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3584 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3585 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3587 if (params
->n_socket_fds
> 1) {
3588 log_unit_error(unit
, "Got more than one socket.");
3592 if (params
->n_socket_fds
== 0) {
3593 log_unit_error(unit
, "Got no socket.");
3597 socket_fd
= params
->fds
[0];
3601 n_socket_fds
= params
->n_socket_fds
;
3602 n_storage_fds
= params
->n_storage_fds
;
3605 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3607 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3609 r
= exec_context_load_environment(unit
, context
, &files_env
);
3611 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3613 line
= exec_command_line(command
->argv
);
3617 log_struct(LOG_DEBUG
,
3618 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3619 "EXECUTABLE=%s", command
->path
,
3621 LOG_UNIT_INVOCATION_ID(unit
));
3625 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3628 int exit_status
= EXIT_SUCCESS
;
3630 r
= exec_child(unit
,
3642 unit
->manager
->user_lookup_fds
[1],
3646 log_struct_errno(LOG_ERR
, r
,
3647 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3649 LOG_UNIT_INVOCATION_ID(unit
),
3650 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3651 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3653 "EXECUTABLE=%s", command
->path
);
3658 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3660 /* We add the new process to the cgroup both in the child (so
3661 * that we can be sure that no user code is ever executed
3662 * outside of the cgroup) and in the parent (so that we can be
3663 * sure that when we kill the cgroup the process will be
3665 if (params
->cgroup_path
)
3666 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3668 exec_status_start(&command
->exec_status
, pid
);
3674 void exec_context_init(ExecContext
*c
) {
3675 ExecDirectoryType i
;
3680 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3681 c
->cpu_sched_policy
= SCHED_OTHER
;
3682 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3683 c
->syslog_level_prefix
= true;
3684 c
->ignore_sigpipe
= true;
3685 c
->timer_slack_nsec
= NSEC_INFINITY
;
3686 c
->personality
= PERSONALITY_INVALID
;
3687 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3688 c
->directories
[i
].mode
= 0755;
3689 c
->capability_bounding_set
= CAP_ALL
;
3690 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3691 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3692 c
->log_level_max
= -1;
3695 void exec_context_done(ExecContext
*c
) {
3696 ExecDirectoryType i
;
3701 c
->environment
= strv_free(c
->environment
);
3702 c
->environment_files
= strv_free(c
->environment_files
);
3703 c
->pass_environment
= strv_free(c
->pass_environment
);
3704 c
->unset_environment
= strv_free(c
->unset_environment
);
3706 rlimit_free_all(c
->rlimit
);
3708 for (l
= 0; l
< 3; l
++) {
3709 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3710 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3713 c
->working_directory
= mfree(c
->working_directory
);
3714 c
->root_directory
= mfree(c
->root_directory
);
3715 c
->root_image
= mfree(c
->root_image
);
3716 c
->tty_path
= mfree(c
->tty_path
);
3717 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3718 c
->user
= mfree(c
->user
);
3719 c
->group
= mfree(c
->group
);
3721 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3723 c
->pam_name
= mfree(c
->pam_name
);
3725 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3726 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3727 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3729 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3730 c
->bind_mounts
= NULL
;
3731 c
->n_bind_mounts
= 0;
3732 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3733 c
->temporary_filesystems
= NULL
;
3734 c
->n_temporary_filesystems
= 0;
3736 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3738 c
->utmp_id
= mfree(c
->utmp_id
);
3739 c
->selinux_context
= mfree(c
->selinux_context
);
3740 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3741 c
->smack_process_label
= mfree(c
->smack_process_label
);
3743 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3744 c
->syscall_archs
= set_free(c
->syscall_archs
);
3745 c
->address_families
= set_free(c
->address_families
);
3747 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3748 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3750 c
->log_level_max
= -1;
3752 exec_context_free_log_extra_fields(c
);
3754 c
->log_rate_limit_interval_usec
= 0;
3755 c
->log_rate_limit_burst
= 0;
3757 c
->stdin_data
= mfree(c
->stdin_data
);
3758 c
->stdin_data_size
= 0;
3761 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3766 if (!runtime_prefix
)
3769 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3770 _cleanup_free_
char *p
;
3772 p
= strjoin(runtime_prefix
, "/", *i
);
3776 /* We execute this synchronously, since we need to be sure this is gone when we start the service
3778 (void) rm_rf(p
, REMOVE_ROOT
);
3784 static void exec_command_done(ExecCommand
*c
) {
3787 c
->path
= mfree(c
->path
);
3788 c
->argv
= strv_free(c
->argv
);
3791 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3794 for (i
= 0; i
< n
; i
++)
3795 exec_command_done(c
+i
);
3798 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3802 LIST_REMOVE(command
, c
, i
);
3803 exec_command_done(i
);
3810 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3813 for (i
= 0; i
< n
; i
++)
3814 c
[i
] = exec_command_free_list(c
[i
]);
3817 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3820 for (i
= 0; i
< n
; i
++)
3821 exec_status_reset(&c
[i
].exec_status
);
3824 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3827 for (i
= 0; i
< n
; i
++) {
3830 LIST_FOREACH(command
, z
, c
[i
])
3831 exec_status_reset(&z
->exec_status
);
3835 typedef struct InvalidEnvInfo
{
3840 static void invalid_env(const char *p
, void *userdata
) {
3841 InvalidEnvInfo
*info
= userdata
;
3843 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3846 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3852 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3855 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3858 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3861 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3864 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3867 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3874 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3876 const char* stdio_fdname
[3];
3882 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3883 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3884 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3886 for (i
= 0; i
< 3; i
++)
3887 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3889 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3891 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3892 if (named_iofds
[STDIN_FILENO
] < 0 &&
3893 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3894 stdio_fdname
[STDIN_FILENO
] &&
3895 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3897 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3900 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3901 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3902 stdio_fdname
[STDOUT_FILENO
] &&
3903 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3905 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3908 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3909 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3910 stdio_fdname
[STDERR_FILENO
] &&
3911 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3913 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3917 return targets
== 0 ? 0 : -ENOENT
;
3920 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
3921 char **i
, **r
= NULL
;
3926 STRV_FOREACH(i
, c
->environment_files
) {
3930 bool ignore
= false;
3932 _cleanup_globfree_ glob_t pglob
= {};
3941 if (!path_is_absolute(fn
)) {
3949 /* Filename supports globbing, take all matching files */
3950 k
= safe_glob(fn
, 0, &pglob
);
3959 /* When we don't match anything, -ENOENT should be returned */
3960 assert(pglob
.gl_pathc
> 0);
3962 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
3963 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3971 /* Log invalid environment variables with filename */
3973 InvalidEnvInfo info
= {
3975 .path
= pglob
.gl_pathv
[n
]
3978 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3986 m
= strv_env_merge(2, r
, p
);
4002 static bool tty_may_match_dev_console(const char *tty
) {
4003 _cleanup_free_
char *resolved
= NULL
;
4008 tty
= skip_dev_prefix(tty
);
4010 /* trivial identity? */
4011 if (streq(tty
, "console"))
4014 if (resolve_dev_console(&resolved
) < 0)
4015 return true; /* if we could not resolve, assume it may */
4017 /* "tty0" means the active VC, so it may be the same sometimes */
4018 return streq(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4021 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4023 return (ec
->tty_reset
||
4025 ec
->tty_vt_disallocate
||
4026 is_terminal_input(ec
->std_input
) ||
4027 is_terminal_output(ec
->std_output
) ||
4028 is_terminal_output(ec
->std_error
)) &&
4029 tty_may_match_dev_console(exec_context_tty_path(ec
));
4032 static void strv_fprintf(FILE *f
, char **l
) {
4038 fprintf(f
, " %s", *g
);
4041 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4042 ExecDirectoryType dt
;
4050 prefix
= strempty(prefix
);
4054 "%sWorkingDirectory: %s\n"
4055 "%sRootDirectory: %s\n"
4056 "%sNonBlocking: %s\n"
4057 "%sPrivateTmp: %s\n"
4058 "%sPrivateDevices: %s\n"
4059 "%sProtectKernelTunables: %s\n"
4060 "%sProtectKernelModules: %s\n"
4061 "%sProtectControlGroups: %s\n"
4062 "%sPrivateNetwork: %s\n"
4063 "%sPrivateUsers: %s\n"
4064 "%sProtectHome: %s\n"
4065 "%sProtectSystem: %s\n"
4066 "%sMountAPIVFS: %s\n"
4067 "%sIgnoreSIGPIPE: %s\n"
4068 "%sMemoryDenyWriteExecute: %s\n"
4069 "%sRestrictRealtime: %s\n"
4070 "%sKeyringMode: %s\n",
4072 prefix
, c
->working_directory
? c
->working_directory
: "/",
4073 prefix
, c
->root_directory
? c
->root_directory
: "/",
4074 prefix
, yes_no(c
->non_blocking
),
4075 prefix
, yes_no(c
->private_tmp
),
4076 prefix
, yes_no(c
->private_devices
),
4077 prefix
, yes_no(c
->protect_kernel_tunables
),
4078 prefix
, yes_no(c
->protect_kernel_modules
),
4079 prefix
, yes_no(c
->protect_control_groups
),
4080 prefix
, yes_no(c
->private_network
),
4081 prefix
, yes_no(c
->private_users
),
4082 prefix
, protect_home_to_string(c
->protect_home
),
4083 prefix
, protect_system_to_string(c
->protect_system
),
4084 prefix
, yes_no(c
->mount_apivfs
),
4085 prefix
, yes_no(c
->ignore_sigpipe
),
4086 prefix
, yes_no(c
->memory_deny_write_execute
),
4087 prefix
, yes_no(c
->restrict_realtime
),
4088 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
));
4091 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4093 STRV_FOREACH(e
, c
->environment
)
4094 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4096 STRV_FOREACH(e
, c
->environment_files
)
4097 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4099 STRV_FOREACH(e
, c
->pass_environment
)
4100 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4102 STRV_FOREACH(e
, c
->unset_environment
)
4103 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4105 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4107 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4108 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4110 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4111 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4119 if (c
->oom_score_adjust_set
)
4121 "%sOOMScoreAdjust: %i\n",
4122 prefix
, c
->oom_score_adjust
);
4124 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4126 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4127 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4128 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4129 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4132 if (c
->ioprio_set
) {
4133 _cleanup_free_
char *class_str
= NULL
;
4135 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4137 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4139 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4142 if (c
->cpu_sched_set
) {
4143 _cleanup_free_
char *policy_str
= NULL
;
4145 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4147 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4150 "%sCPUSchedulingPriority: %i\n"
4151 "%sCPUSchedulingResetOnFork: %s\n",
4152 prefix
, c
->cpu_sched_priority
,
4153 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4157 fprintf(f
, "%sCPUAffinity:", prefix
);
4158 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4159 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4160 fprintf(f
, " %u", i
);
4164 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4165 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4168 "%sStandardInput: %s\n"
4169 "%sStandardOutput: %s\n"
4170 "%sStandardError: %s\n",
4171 prefix
, exec_input_to_string(c
->std_input
),
4172 prefix
, exec_output_to_string(c
->std_output
),
4173 prefix
, exec_output_to_string(c
->std_error
));
4175 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4176 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4177 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4178 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4179 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4180 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4182 if (c
->std_input
== EXEC_INPUT_FILE
)
4183 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4184 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4185 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4186 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4187 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4188 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4189 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4190 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4191 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4197 "%sTTYVHangup: %s\n"
4198 "%sTTYVTDisallocate: %s\n",
4199 prefix
, c
->tty_path
,
4200 prefix
, yes_no(c
->tty_reset
),
4201 prefix
, yes_no(c
->tty_vhangup
),
4202 prefix
, yes_no(c
->tty_vt_disallocate
));
4204 if (IN_SET(c
->std_output
,
4207 EXEC_OUTPUT_JOURNAL
,
4208 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4209 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4210 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4211 IN_SET(c
->std_error
,
4214 EXEC_OUTPUT_JOURNAL
,
4215 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4216 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4217 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4219 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4221 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4223 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4225 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4227 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4230 if (c
->log_level_max
>= 0) {
4231 _cleanup_free_
char *t
= NULL
;
4233 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4235 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4238 if (c
->log_rate_limit_interval_usec
> 0) {
4239 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4242 "%sLogRateLimitIntervalSec: %s\n",
4243 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4246 if (c
->log_rate_limit_burst
> 0)
4247 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4249 if (c
->n_log_extra_fields
> 0) {
4252 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4253 fprintf(f
, "%sLogExtraFields: ", prefix
);
4254 fwrite(c
->log_extra_fields
[j
].iov_base
,
4255 1, c
->log_extra_fields
[j
].iov_len
,
4261 if (c
->secure_bits
) {
4262 _cleanup_free_
char *str
= NULL
;
4264 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4266 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4269 if (c
->capability_bounding_set
!= CAP_ALL
) {
4270 _cleanup_free_
char *str
= NULL
;
4272 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4274 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4277 if (c
->capability_ambient_set
!= 0) {
4278 _cleanup_free_
char *str
= NULL
;
4280 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4282 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4286 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4288 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4290 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4292 if (!strv_isempty(c
->supplementary_groups
)) {
4293 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4294 strv_fprintf(f
, c
->supplementary_groups
);
4299 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4301 if (!strv_isempty(c
->read_write_paths
)) {
4302 fprintf(f
, "%sReadWritePaths:", prefix
);
4303 strv_fprintf(f
, c
->read_write_paths
);
4307 if (!strv_isempty(c
->read_only_paths
)) {
4308 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4309 strv_fprintf(f
, c
->read_only_paths
);
4313 if (!strv_isempty(c
->inaccessible_paths
)) {
4314 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4315 strv_fprintf(f
, c
->inaccessible_paths
);
4319 if (c
->n_bind_mounts
> 0)
4320 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4321 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4322 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4323 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4324 c
->bind_mounts
[i
].source
,
4325 c
->bind_mounts
[i
].destination
,
4326 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4328 if (c
->n_temporary_filesystems
> 0)
4329 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4330 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4332 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4334 isempty(t
->options
) ? "" : ":",
4335 strempty(t
->options
));
4340 "%sUtmpIdentifier: %s\n",
4341 prefix
, c
->utmp_id
);
4343 if (c
->selinux_context
)
4345 "%sSELinuxContext: %s%s\n",
4346 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4348 if (c
->apparmor_profile
)
4350 "%sAppArmorProfile: %s%s\n",
4351 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4353 if (c
->smack_process_label
)
4355 "%sSmackProcessLabel: %s%s\n",
4356 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4358 if (c
->personality
!= PERSONALITY_INVALID
)
4360 "%sPersonality: %s\n",
4361 prefix
, strna(personality_to_string(c
->personality
)));
4364 "%sLockPersonality: %s\n",
4365 prefix
, yes_no(c
->lock_personality
));
4367 if (c
->syscall_filter
) {
4375 "%sSystemCallFilter: ",
4378 if (!c
->syscall_whitelist
)
4382 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4383 _cleanup_free_
char *name
= NULL
;
4384 const char *errno_name
= NULL
;
4385 int num
= PTR_TO_INT(val
);
4392 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4393 fputs(strna(name
), f
);
4396 errno_name
= errno_to_name(num
);
4398 fprintf(f
, ":%s", errno_name
);
4400 fprintf(f
, ":%d", num
);
4408 if (c
->syscall_archs
) {
4415 "%sSystemCallArchitectures:",
4419 SET_FOREACH(id
, c
->syscall_archs
, j
)
4420 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4425 if (exec_context_restrict_namespaces_set(c
)) {
4426 _cleanup_free_
char *s
= NULL
;
4428 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4430 fprintf(f
, "%sRestrictNamespaces: %s\n",
4434 if (c
->syscall_errno
> 0) {
4435 const char *errno_name
;
4437 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4439 errno_name
= errno_to_name(c
->syscall_errno
);
4441 fprintf(f
, "%s\n", errno_name
);
4443 fprintf(f
, "%d\n", c
->syscall_errno
);
4446 if (c
->apparmor_profile
)
4448 "%sAppArmorProfile: %s%s\n",
4449 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4452 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4455 /* Returns true if the process forked off would run under
4456 * an unchanged UID or as root. */
4461 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4467 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4475 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4477 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4482 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4487 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4488 free(c
->log_extra_fields
[l
].iov_base
);
4489 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4490 c
->n_log_extra_fields
= 0;
4493 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4500 dual_timestamp_get(&s
->start_timestamp
);
4503 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4506 if (s
->pid
!= pid
) {
4512 dual_timestamp_get(&s
->exit_timestamp
);
4518 if (context
->utmp_id
)
4519 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4521 exec_context_tty_reset(context
, NULL
);
4525 void exec_status_reset(ExecStatus
*s
) {
4528 *s
= (ExecStatus
) {};
4531 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4532 char buf
[FORMAT_TIMESTAMP_MAX
];
4540 prefix
= strempty(prefix
);
4543 "%sPID: "PID_FMT
"\n",
4546 if (dual_timestamp_is_set(&s
->start_timestamp
))
4548 "%sStart Timestamp: %s\n",
4549 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4551 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4553 "%sExit Timestamp: %s\n"
4555 "%sExit Status: %i\n",
4556 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4557 prefix
, sigchld_code_to_string(s
->code
),
4561 static char *exec_command_line(char **argv
) {
4569 STRV_FOREACH(a
, argv
)
4577 STRV_FOREACH(a
, argv
) {
4584 if (strpbrk(*a
, WHITESPACE
)) {
4595 /* FIXME: this doesn't really handle arguments that have
4596 * spaces and ticks in them */
4601 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4602 _cleanup_free_
char *cmd
= NULL
;
4603 const char *prefix2
;
4608 prefix
= strempty(prefix
);
4609 prefix2
= strjoina(prefix
, "\t");
4611 cmd
= exec_command_line(c
->argv
);
4613 "%sCommand Line: %s\n",
4614 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4616 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4619 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4622 prefix
= strempty(prefix
);
4624 LIST_FOREACH(command
, c
, c
)
4625 exec_command_dump(c
, f
, prefix
);
4628 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4635 /* It's kind of important, that we keep the order here */
4636 LIST_FIND_TAIL(command
, *l
, end
);
4637 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4642 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4650 l
= strv_new_ap(path
, ap
);
4665 return strv_free_and_replace(c
->argv
, l
);
4668 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4669 _cleanup_strv_free_
char **l
= NULL
;
4677 l
= strv_new_ap(path
, ap
);
4683 r
= strv_extend_strv(&c
->argv
, l
, false);
4690 static void *remove_tmpdir_thread(void *p
) {
4691 _cleanup_free_
char *path
= p
;
4693 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4697 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4704 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4706 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4707 if (destroy
&& rt
->tmp_dir
) {
4708 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4710 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4712 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4719 if (destroy
&& rt
->var_tmp_dir
) {
4720 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4722 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4724 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4725 free(rt
->var_tmp_dir
);
4728 rt
->var_tmp_dir
= NULL
;
4731 rt
->id
= mfree(rt
->id
);
4732 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4733 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4734 safe_close_pair(rt
->netns_storage_socket
);
4738 static void exec_runtime_freep(ExecRuntime
**rt
) {
4740 (void) exec_runtime_free(*rt
, false);
4743 static int exec_runtime_allocate(ExecRuntime
**rt
) {
4746 *rt
= new0(ExecRuntime
, 1);
4750 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
4754 static int exec_runtime_add(
4757 const char *tmp_dir
,
4758 const char *var_tmp_dir
,
4759 const int netns_storage_socket
[2],
4760 ExecRuntime
**ret
) {
4762 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4768 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4772 r
= exec_runtime_allocate(&rt
);
4776 rt
->id
= strdup(id
);
4781 rt
->tmp_dir
= strdup(tmp_dir
);
4785 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4786 assert(var_tmp_dir
);
4787 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4788 if (!rt
->var_tmp_dir
)
4792 if (netns_storage_socket
) {
4793 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4794 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4797 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4806 /* do not remove created ExecRuntime object when the operation succeeds. */
4811 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4812 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4813 _cleanup_close_pair_
int netns_storage_socket
[2] = {-1, -1};
4820 /* It is not necessary to create ExecRuntime object. */
4821 if (!c
->private_network
&& !c
->private_tmp
)
4824 if (c
->private_tmp
) {
4825 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
4830 if (c
->private_network
) {
4831 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
4835 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
4840 netns_storage_socket
[0] = -1;
4841 netns_storage_socket
[1] = -1;
4845 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
4853 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
4855 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
4861 /* If not found, then create a new object. */
4862 r
= exec_runtime_make(m
, c
, id
, &rt
);
4864 /* When r == 0, it is not necessary to create ExecRuntime object. */
4868 /* increment reference counter. */
4874 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
4878 assert(rt
->n_ref
> 0);
4884 return exec_runtime_free(rt
, destroy
);
4887 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
4895 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
4896 fprintf(f
, "exec-runtime=%s", rt
->id
);
4899 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
4901 if (rt
->var_tmp_dir
)
4902 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
4904 if (rt
->netns_storage_socket
[0] >= 0) {
4907 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4911 fprintf(f
, " netns-socket-0=%i", copy
);
4914 if (rt
->netns_storage_socket
[1] >= 0) {
4917 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
4921 fprintf(f
, " netns-socket-1=%i", copy
);
4930 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
4931 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
4935 /* This is for the migration from old (v237 or earlier) deserialization text.
4936 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
4937 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
4938 * so or not from the serialized text, then we always creates a new object owned by this. */
4944 /* Manager manages ExecRuntime objects by the unit id.
4945 * So, we omit the serialized text when the unit does not have id (yet?)... */
4946 if (isempty(u
->id
)) {
4947 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
4951 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
4953 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
4957 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
4959 r
= exec_runtime_allocate(&rt_create
);
4963 rt_create
->id
= strdup(u
->id
);
4970 if (streq(key
, "tmp-dir")) {
4973 copy
= strdup(value
);
4977 free_and_replace(rt
->tmp_dir
, copy
);
4979 } else if (streq(key
, "var-tmp-dir")) {
4982 copy
= strdup(value
);
4986 free_and_replace(rt
->var_tmp_dir
, copy
);
4988 } else if (streq(key
, "netns-socket-0")) {
4991 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
4992 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4996 safe_close(rt
->netns_storage_socket
[0]);
4997 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4999 } else if (streq(key
, "netns-socket-1")) {
5002 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5003 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5007 safe_close(rt
->netns_storage_socket
[1]);
5008 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5012 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5014 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5016 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5020 rt_create
->manager
= u
->manager
;
5029 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5030 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5031 int r
, fd0
= -1, fd1
= -1;
5032 const char *p
, *v
= value
;
5039 n
= strcspn(v
, " ");
5040 id
= strndupa(v
, n
);
5045 v
= startswith(p
, "tmp-dir=");
5047 n
= strcspn(v
, " ");
5048 tmp_dir
= strndupa(v
, n
);
5054 v
= startswith(p
, "var-tmp-dir=");
5056 n
= strcspn(v
, " ");
5057 var_tmp_dir
= strndupa(v
, n
);
5063 v
= startswith(p
, "netns-socket-0=");
5067 n
= strcspn(v
, " ");
5068 buf
= strndupa(v
, n
);
5069 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5070 log_debug("Unable to process exec-runtime netns fd specification.");
5073 fd0
= fdset_remove(fds
, fd0
);
5079 v
= startswith(p
, "netns-socket-1=");
5083 n
= strcspn(v
, " ");
5084 buf
= strndupa(v
, n
);
5085 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5086 log_debug("Unable to process exec-runtime netns fd specification.");
5089 fd1
= fdset_remove(fds
, fd1
);
5094 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5096 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5099 void exec_runtime_vacuum(Manager
*m
) {
5105 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5107 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5111 (void) exec_runtime_free(rt
, false);
5115 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5116 [EXEC_INPUT_NULL
] = "null",
5117 [EXEC_INPUT_TTY
] = "tty",
5118 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5119 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5120 [EXEC_INPUT_SOCKET
] = "socket",
5121 [EXEC_INPUT_NAMED_FD
] = "fd",
5122 [EXEC_INPUT_DATA
] = "data",
5123 [EXEC_INPUT_FILE
] = "file",
5126 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5128 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5129 [EXEC_OUTPUT_INHERIT
] = "inherit",
5130 [EXEC_OUTPUT_NULL
] = "null",
5131 [EXEC_OUTPUT_TTY
] = "tty",
5132 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5133 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5134 [EXEC_OUTPUT_KMSG
] = "kmsg",
5135 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5136 [EXEC_OUTPUT_JOURNAL
] = "journal",
5137 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5138 [EXEC_OUTPUT_SOCKET
] = "socket",
5139 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5140 [EXEC_OUTPUT_FILE
] = "file",
5141 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5144 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5146 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5147 [EXEC_UTMP_INIT
] = "init",
5148 [EXEC_UTMP_LOGIN
] = "login",
5149 [EXEC_UTMP_USER
] = "user",
5152 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5154 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5155 [EXEC_PRESERVE_NO
] = "no",
5156 [EXEC_PRESERVE_YES
] = "yes",
5157 [EXEC_PRESERVE_RESTART
] = "restart",
5160 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5162 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5163 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5164 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5165 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5166 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5167 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5170 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5172 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5173 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5174 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5175 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5176 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5177 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5180 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5182 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5183 [EXEC_KEYRING_INHERIT
] = "inherit",
5184 [EXEC_KEYRING_PRIVATE
] = "private",
5185 [EXEC_KEYRING_SHARED
] = "shared",
5188 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);