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"
55 #include "errno-list.h"
57 #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-util.h"
80 #include "selinux-util.h"
81 #include "signal-util.h"
82 #include "smack-util.h"
83 #include "socket-util.h"
85 #include "stat-util.h"
86 #include "string-table.h"
87 #include "string-util.h"
89 #include "syslog-util.h"
90 #include "terminal-util.h"
91 #include "umask-util.h"
93 #include "user-util.h"
95 #include "utmp-wtmp.h"
97 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
98 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
100 /* This assumes there is a 'tty' group */
101 #define TTY_MODE 0620
103 #define SNDBUF_SIZE (8*1024*1024)
105 static int shift_fds(int fds
[], size_t n_fds
) {
106 int start
, restart_from
;
111 /* Modifies the fds array! (sorts it) */
121 for (i
= start
; i
< (int) n_fds
; i
++) {
124 /* Already at right index? */
128 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
135 /* Hmm, the fd we wanted isn't free? Then
136 * let's remember that and try again from here */
137 if (nfd
!= i
+3 && restart_from
< 0)
141 if (restart_from
< 0)
144 start
= restart_from
;
150 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
154 n_fds
= n_socket_fds
+ n_storage_fds
;
160 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
161 * O_NONBLOCK only applies to socket activation though. */
163 for (i
= 0; i
< n_fds
; i
++) {
165 if (i
< n_socket_fds
) {
166 r
= fd_nonblock(fds
[i
], nonblock
);
171 /* We unconditionally drop FD_CLOEXEC from the fds,
172 * since after all we want to pass these fds to our
175 r
= fd_cloexec(fds
[i
], false);
183 static const char *exec_context_tty_path(const ExecContext
*context
) {
186 if (context
->stdio_as_fds
)
189 if (context
->tty_path
)
190 return context
->tty_path
;
192 return "/dev/console";
195 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
200 path
= exec_context_tty_path(context
);
202 if (context
->tty_vhangup
) {
203 if (p
&& p
->stdin_fd
>= 0)
204 (void) terminal_vhangup_fd(p
->stdin_fd
);
206 (void) terminal_vhangup(path
);
209 if (context
->tty_reset
) {
210 if (p
&& p
->stdin_fd
>= 0)
211 (void) reset_terminal_fd(p
->stdin_fd
, true);
213 (void) reset_terminal(path
);
216 if (context
->tty_vt_disallocate
&& path
)
217 (void) vt_disallocate(path
);
220 static bool is_terminal_input(ExecInput i
) {
223 EXEC_INPUT_TTY_FORCE
,
224 EXEC_INPUT_TTY_FAIL
);
227 static bool is_terminal_output(ExecOutput o
) {
230 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
231 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
232 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
235 static bool is_syslog_output(ExecOutput o
) {
238 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
241 static bool is_kmsg_output(ExecOutput o
) {
244 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
247 static bool exec_context_needs_term(const ExecContext
*c
) {
250 /* Return true if the execution context suggests we should set $TERM to something useful. */
252 if (is_terminal_input(c
->std_input
))
255 if (is_terminal_output(c
->std_output
))
258 if (is_terminal_output(c
->std_error
))
261 return !!c
->tty_path
;
264 static int open_null_as(int flags
, int nfd
) {
269 fd
= open("/dev/null", flags
|O_NOCTTY
);
273 return move_fd(fd
, nfd
, false);
276 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
277 static const union sockaddr_union sa
= {
278 .un
.sun_family
= AF_UNIX
,
279 .un
.sun_path
= "/run/systemd/journal/stdout",
281 uid_t olduid
= UID_INVALID
;
282 gid_t oldgid
= GID_INVALID
;
285 if (gid_is_valid(gid
)) {
288 if (setegid(gid
) < 0)
292 if (uid_is_valid(uid
)) {
295 if (seteuid(uid
) < 0) {
301 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
303 /* If we fail to restore the uid or gid, things will likely
304 fail later on. This should only happen if an LSM interferes. */
306 if (uid_is_valid(uid
))
307 (void) seteuid(olduid
);
310 if (gid_is_valid(gid
))
311 (void) setegid(oldgid
);
316 static int connect_logger_as(
318 const ExecContext
*context
,
319 const ExecParameters
*params
,
326 _cleanup_close_
int fd
= -1;
331 assert(output
< _EXEC_OUTPUT_MAX
);
335 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
339 r
= connect_journal_socket(fd
, uid
, gid
);
343 if (shutdown(fd
, SHUT_RD
) < 0)
346 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
356 context
->syslog_identifier
?: ident
,
357 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
358 context
->syslog_priority
,
359 !!context
->syslog_level_prefix
,
360 is_syslog_output(output
),
361 is_kmsg_output(output
),
362 is_terminal_output(output
)) < 0)
365 return move_fd(TAKE_FD(fd
), nfd
, false);
368 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
374 fd
= open_terminal(path
, flags
| O_NOCTTY
);
378 return move_fd(fd
, nfd
, false);
381 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
382 union sockaddr_union sa
= {};
383 _cleanup_close_
int fd
= -1;
388 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
391 fd
= open(path
, flags
|O_NOCTTY
, mode
);
395 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
397 if (strlen(path
) >= sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
400 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
402 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
406 salen
= sockaddr_un_set_path(&sa
.un
, path
);
410 if (connect(fd
, &sa
.sa
, salen
) < 0)
411 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
412 * indication that his wasn't an AF_UNIX socket after all */
414 if ((flags
& O_ACCMODE
) == O_RDONLY
)
415 r
= shutdown(fd
, SHUT_WR
);
416 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
417 r
= shutdown(fd
, SHUT_RD
);
426 static int fixup_input(
427 const ExecContext
*context
,
429 bool apply_tty_stdin
) {
435 std_input
= context
->std_input
;
437 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
438 return EXEC_INPUT_NULL
;
440 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
441 return EXEC_INPUT_NULL
;
443 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
444 return EXEC_INPUT_NULL
;
449 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
451 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
452 return EXEC_OUTPUT_INHERIT
;
457 static int setup_input(
458 const ExecContext
*context
,
459 const ExecParameters
*params
,
461 int named_iofds
[3]) {
468 if (params
->stdin_fd
>= 0) {
469 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
472 /* Try to make this the controlling tty, if it is a tty, and reset it */
473 if (isatty(STDIN_FILENO
)) {
474 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
475 (void) reset_terminal_fd(STDIN_FILENO
, true);
481 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
485 case EXEC_INPUT_NULL
:
486 return open_null_as(O_RDONLY
, STDIN_FILENO
);
489 case EXEC_INPUT_TTY_FORCE
:
490 case EXEC_INPUT_TTY_FAIL
: {
493 fd
= acquire_terminal(exec_context_tty_path(context
),
494 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
495 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
496 ACQUIRE_TERMINAL_WAIT
,
501 return move_fd(fd
, STDIN_FILENO
, false);
504 case EXEC_INPUT_SOCKET
:
505 assert(socket_fd
>= 0);
507 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
509 case EXEC_INPUT_NAMED_FD
:
510 assert(named_iofds
[STDIN_FILENO
] >= 0);
512 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
513 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
515 case EXEC_INPUT_DATA
: {
518 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
522 return move_fd(fd
, STDIN_FILENO
, false);
525 case EXEC_INPUT_FILE
: {
529 assert(context
->stdio_file
[STDIN_FILENO
]);
531 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
532 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
534 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
538 return move_fd(fd
, STDIN_FILENO
, false);
542 assert_not_reached("Unknown input type");
546 static bool can_inherit_stderr_from_stdout(
547 const ExecContext
*context
,
553 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
556 if (e
== EXEC_OUTPUT_INHERIT
)
561 if (e
== EXEC_OUTPUT_NAMED_FD
)
562 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
564 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
565 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
570 static int setup_output(
572 const ExecContext
*context
,
573 const ExecParameters
*params
,
580 dev_t
*journal_stream_dev
,
581 ino_t
*journal_stream_ino
) {
591 assert(journal_stream_dev
);
592 assert(journal_stream_ino
);
594 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
596 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
599 return STDOUT_FILENO
;
602 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
603 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
606 return STDERR_FILENO
;
609 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
610 o
= fixup_output(context
->std_output
, socket_fd
);
612 if (fileno
== STDERR_FILENO
) {
614 e
= fixup_output(context
->std_error
, socket_fd
);
616 /* This expects the input and output are already set up */
618 /* Don't change the stderr file descriptor if we inherit all
619 * the way and are not on a tty */
620 if (e
== EXEC_OUTPUT_INHERIT
&&
621 o
== EXEC_OUTPUT_INHERIT
&&
622 i
== EXEC_INPUT_NULL
&&
623 !is_terminal_input(context
->std_input
) &&
627 /* Duplicate from stdout if possible */
628 if (can_inherit_stderr_from_stdout(context
, o
, e
))
629 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
633 } else if (o
== EXEC_OUTPUT_INHERIT
) {
634 /* If input got downgraded, inherit the original value */
635 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
636 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
638 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
639 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
640 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
642 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
646 /* We need to open /dev/null here anew, to get the right access mode. */
647 return open_null_as(O_WRONLY
, fileno
);
652 case EXEC_OUTPUT_NULL
:
653 return open_null_as(O_WRONLY
, fileno
);
655 case EXEC_OUTPUT_TTY
:
656 if (is_terminal_input(i
))
657 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
659 /* We don't reset the terminal if this is just about output */
660 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
662 case EXEC_OUTPUT_SYSLOG
:
663 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
664 case EXEC_OUTPUT_KMSG
:
665 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
666 case EXEC_OUTPUT_JOURNAL
:
667 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
668 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
670 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
671 r
= open_null_as(O_WRONLY
, fileno
);
675 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
676 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
677 * services to detect whether they are connected to the journal or not.
679 * If both stdout and stderr are connected to a stream then let's make sure to store the data
680 * about STDERR as that's usually the best way to do logging. */
682 if (fstat(fileno
, &st
) >= 0 &&
683 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
684 *journal_stream_dev
= st
.st_dev
;
685 *journal_stream_ino
= st
.st_ino
;
690 case EXEC_OUTPUT_SOCKET
:
691 assert(socket_fd
>= 0);
693 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
695 case EXEC_OUTPUT_NAMED_FD
:
696 assert(named_iofds
[fileno
] >= 0);
698 (void) fd_nonblock(named_iofds
[fileno
], false);
699 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
701 case EXEC_OUTPUT_FILE
:
702 case EXEC_OUTPUT_FILE_APPEND
: {
706 assert(context
->stdio_file
[fileno
]);
708 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
709 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
712 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
715 if (o
== EXEC_OUTPUT_FILE_APPEND
)
718 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
722 return move_fd(fd
, fileno
, 0);
726 assert_not_reached("Unknown error type");
730 static int chown_terminal(int fd
, uid_t uid
) {
735 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
739 /* This might fail. What matters are the results. */
740 (void) fchown(fd
, uid
, -1);
741 (void) fchmod(fd
, TTY_MODE
);
743 if (fstat(fd
, &st
) < 0)
746 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
752 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
753 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
756 assert(_saved_stdin
);
757 assert(_saved_stdout
);
759 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
763 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
764 if (saved_stdout
< 0)
767 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
771 r
= chown_terminal(fd
, getuid());
775 r
= reset_terminal_fd(fd
, true);
779 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
784 *_saved_stdin
= saved_stdin
;
785 *_saved_stdout
= saved_stdout
;
787 saved_stdin
= saved_stdout
= -1;
792 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
795 if (err
== -ETIMEDOUT
)
796 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
799 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
803 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
804 _cleanup_close_
int fd
= -1;
808 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
812 write_confirm_error_fd(err
, fd
, u
);
815 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
819 assert(saved_stdout
);
823 if (*saved_stdin
>= 0)
824 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
827 if (*saved_stdout
>= 0)
828 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
831 *saved_stdin
= safe_close(*saved_stdin
);
832 *saved_stdout
= safe_close(*saved_stdout
);
838 CONFIRM_PRETEND_FAILURE
= -1,
839 CONFIRM_PRETEND_SUCCESS
= 0,
843 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
844 int saved_stdout
= -1, saved_stdin
= -1, r
;
845 _cleanup_free_
char *e
= NULL
;
848 /* For any internal errors, assume a positive response. */
849 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
851 write_confirm_error(r
, vc
, u
);
852 return CONFIRM_EXECUTE
;
855 /* confirm_spawn might have been disabled while we were sleeping. */
856 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
861 e
= ellipsize(cmdline
, 60, 100);
869 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
871 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
878 printf("Resuming normal execution.\n");
879 manager_disable_confirm_spawn();
883 unit_dump(u
, stdout
, " ");
884 continue; /* ask again */
886 printf("Failing execution.\n");
887 r
= CONFIRM_PRETEND_FAILURE
;
890 printf(" c - continue, proceed without asking anymore\n"
891 " D - dump, show the state of the unit\n"
892 " f - fail, don't execute the command and pretend it failed\n"
894 " i - info, show a short summary of the unit\n"
895 " j - jobs, show jobs that are in progress\n"
896 " s - skip, don't execute the command and pretend it succeeded\n"
897 " y - yes, execute the command\n");
898 continue; /* ask again */
900 printf(" Description: %s\n"
903 u
->id
, u
->description
, cmdline
);
904 continue; /* ask again */
906 manager_dump_jobs(u
->manager
, stdout
, " ");
907 continue; /* ask again */
909 /* 'n' was removed in favor of 'f'. */
910 printf("Didn't understand 'n', did you mean 'f'?\n");
911 continue; /* ask again */
913 printf("Skipping execution.\n");
914 r
= CONFIRM_PRETEND_SUCCESS
;
920 assert_not_reached("Unhandled choice");
926 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
930 static int get_fixed_user(const ExecContext
*c
, const char **user
,
931 uid_t
*uid
, gid_t
*gid
,
932 const char **home
, const char **shell
) {
941 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
942 * (i.e. are "/" or "/bin/nologin"). */
945 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
953 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
963 r
= get_group_creds(&name
, gid
, 0);
971 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
972 const char *group
, gid_t gid
,
973 gid_t
**supplementary_gids
, int *ngids
) {
977 bool keep_groups
= false;
978 gid_t
*groups
= NULL
;
979 _cleanup_free_ gid_t
*l_gids
= NULL
;
984 * If user is given, then lookup GID and supplementary groups list.
985 * We avoid NSS lookups for gid=0. Also we have to initialize groups
986 * here and as early as possible so we keep the list of supplementary
987 * groups of the caller.
989 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
990 /* First step, initialize groups from /etc/groups */
991 if (initgroups(user
, gid
) < 0)
997 if (strv_isempty(c
->supplementary_groups
))
1001 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1002 * be positive, otherwise fail.
1005 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1006 if (ngroups_max
<= 0) {
1010 return -EOPNOTSUPP
; /* For all other values */
1013 l_gids
= new(gid_t
, ngroups_max
);
1019 * Lookup the list of groups that the user belongs to, we
1020 * avoid NSS lookups here too for gid=0.
1023 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1028 STRV_FOREACH(i
, c
->supplementary_groups
) {
1031 if (k
>= ngroups_max
)
1035 r
= get_group_creds(&g
, l_gids
+k
, 0);
1043 * Sets ngids to zero to drop all supplementary groups, happens
1044 * when we are under root and SupplementaryGroups= is empty.
1051 /* Otherwise get the final list of supplementary groups */
1052 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1056 *supplementary_gids
= groups
;
1064 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1067 /* Handle SupplementaryGroups= if it is not empty */
1069 r
= maybe_setgroups(ngids
, supplementary_gids
);
1074 if (gid_is_valid(gid
)) {
1075 /* Then set our gids */
1076 if (setresgid(gid
, gid
, gid
) < 0)
1083 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1086 if (!uid_is_valid(uid
))
1089 /* Sets (but doesn't look up) the uid and make sure we keep the
1090 * capabilities while doing so. */
1092 if (context
->capability_ambient_set
!= 0) {
1094 /* First step: If we need to keep capabilities but
1095 * drop privileges we need to make sure we keep our
1096 * caps, while we drop privileges. */
1098 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1100 if (prctl(PR_GET_SECUREBITS
) != sb
)
1101 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1106 /* Second step: actually set the uids */
1107 if (setresuid(uid
, uid
, uid
) < 0)
1110 /* At this point we should have all necessary capabilities but
1111 are otherwise a normal user. However, the caps might got
1112 corrupted due to the setresuid() so we need clean them up
1113 later. This is done outside of this call. */
1120 static int null_conv(
1122 const struct pam_message
**msg
,
1123 struct pam_response
**resp
,
1124 void *appdata_ptr
) {
1126 /* We don't support conversations */
1128 return PAM_CONV_ERR
;
1133 static int setup_pam(
1140 int fds
[], size_t n_fds
) {
1144 static const struct pam_conv conv
= {
1149 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1150 pam_handle_t
*handle
= NULL
;
1152 int pam_code
= PAM_SUCCESS
, r
;
1153 char **nv
, **e
= NULL
;
1154 bool close_session
= false;
1155 pid_t pam_pid
= 0, parent_pid
;
1162 /* We set up PAM in the parent process, then fork. The child
1163 * will then stay around until killed via PR_GET_PDEATHSIG or
1164 * systemd via the cgroup logic. It will then remove the PAM
1165 * session again. The parent process will exec() the actual
1166 * daemon. We do things this way to ensure that the main PID
1167 * of the daemon is the one we initially fork()ed. */
1169 r
= barrier_create(&barrier
);
1173 if (log_get_max_level() < LOG_DEBUG
)
1174 flags
|= PAM_SILENT
;
1176 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1177 if (pam_code
!= PAM_SUCCESS
) {
1183 _cleanup_free_
char *q
= NULL
;
1185 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1186 * out if that's the case, and read the TTY off it. */
1188 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1189 tty
= strjoina("/dev/", q
);
1193 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1194 if (pam_code
!= PAM_SUCCESS
)
1198 STRV_FOREACH(nv
, *env
) {
1199 pam_code
= pam_putenv(handle
, *nv
);
1200 if (pam_code
!= PAM_SUCCESS
)
1204 pam_code
= pam_acct_mgmt(handle
, flags
);
1205 if (pam_code
!= PAM_SUCCESS
)
1208 pam_code
= pam_open_session(handle
, flags
);
1209 if (pam_code
!= PAM_SUCCESS
)
1212 close_session
= true;
1214 e
= pam_getenvlist(handle
);
1216 pam_code
= PAM_BUF_ERR
;
1220 /* Block SIGTERM, so that we know that it won't get lost in
1223 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1225 parent_pid
= getpid_cached();
1227 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1231 int sig
, ret
= EXIT_PAM
;
1233 /* The child's job is to reset the PAM session on
1235 barrier_set_role(&barrier
, BARRIER_CHILD
);
1237 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1238 * are open here that have been opened by PAM. */
1239 (void) close_many(fds
, n_fds
);
1241 /* Drop privileges - we don't need any to pam_close_session
1242 * and this will make PR_SET_PDEATHSIG work in most cases.
1243 * If this fails, ignore the error - but expect sd-pam threads
1244 * to fail to exit normally */
1246 r
= maybe_setgroups(0, NULL
);
1248 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1249 if (setresgid(gid
, gid
, gid
) < 0)
1250 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1251 if (setresuid(uid
, uid
, uid
) < 0)
1252 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1254 (void) ignore_signals(SIGPIPE
, -1);
1256 /* Wait until our parent died. This will only work if
1257 * the above setresuid() succeeds, otherwise the kernel
1258 * will not allow unprivileged parents kill their privileged
1259 * children this way. We rely on the control groups kill logic
1260 * to do the rest for us. */
1261 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1264 /* Tell the parent that our setup is done. This is especially
1265 * important regarding dropping privileges. Otherwise, unit
1266 * setup might race against our setresuid(2) call.
1268 * If the parent aborted, we'll detect this below, hence ignore
1269 * return failure here. */
1270 (void) barrier_place(&barrier
);
1272 /* Check if our parent process might already have died? */
1273 if (getppid() == parent_pid
) {
1276 assert_se(sigemptyset(&ss
) >= 0);
1277 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1280 if (sigwait(&ss
, &sig
) < 0) {
1287 assert(sig
== SIGTERM
);
1292 /* If our parent died we'll end the session */
1293 if (getppid() != parent_pid
) {
1294 pam_code
= pam_close_session(handle
, flags
);
1295 if (pam_code
!= PAM_SUCCESS
)
1302 pam_end(handle
, pam_code
| flags
);
1306 barrier_set_role(&barrier
, BARRIER_PARENT
);
1308 /* If the child was forked off successfully it will do all the
1309 * cleanups, so forget about the handle here. */
1312 /* Unblock SIGTERM again in the parent */
1313 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1315 /* We close the log explicitly here, since the PAM modules
1316 * might have opened it, but we don't want this fd around. */
1319 /* Synchronously wait for the child to initialize. We don't care for
1320 * errors as we cannot recover. However, warn loudly if it happens. */
1321 if (!barrier_place_and_sync(&barrier
))
1322 log_error("PAM initialization failed");
1324 return strv_free_and_replace(*env
, e
);
1327 if (pam_code
!= PAM_SUCCESS
) {
1328 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1329 r
= -EPERM
; /* PAM errors do not map to errno */
1331 log_error_errno(r
, "PAM failed: %m");
1335 pam_code
= pam_close_session(handle
, flags
);
1337 pam_end(handle
, pam_code
| flags
);
1349 static void rename_process_from_path(const char *path
) {
1350 char process_name
[11];
1354 /* This resulting string must fit in 10 chars (i.e. the length
1355 * of "/sbin/init") to look pretty in /bin/ps */
1359 rename_process("(...)");
1365 /* The end of the process name is usually more
1366 * interesting, since the first bit might just be
1372 process_name
[0] = '(';
1373 memcpy(process_name
+1, p
, l
);
1374 process_name
[1+l
] = ')';
1375 process_name
[1+l
+1] = 0;
1377 rename_process(process_name
);
1380 static bool context_has_address_families(const ExecContext
*c
) {
1383 return c
->address_families_whitelist
||
1384 !set_isempty(c
->address_families
);
1387 static bool context_has_syscall_filters(const ExecContext
*c
) {
1390 return c
->syscall_whitelist
||
1391 !hashmap_isempty(c
->syscall_filter
);
1394 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1397 if (c
->no_new_privileges
)
1400 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1403 /* We need NNP if we have any form of seccomp and are unprivileged */
1404 return context_has_address_families(c
) ||
1405 c
->memory_deny_write_execute
||
1406 c
->restrict_realtime
||
1407 exec_context_restrict_namespaces_set(c
) ||
1408 c
->protect_kernel_tunables
||
1409 c
->protect_kernel_modules
||
1410 c
->private_devices
||
1411 context_has_syscall_filters(c
) ||
1412 !set_isempty(c
->syscall_archs
) ||
1413 c
->lock_personality
;
1418 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1420 if (is_seccomp_available())
1423 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1427 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1428 uint32_t negative_action
, default_action
, action
;
1434 if (!context_has_syscall_filters(c
))
1437 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1440 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1442 if (c
->syscall_whitelist
) {
1443 default_action
= negative_action
;
1444 action
= SCMP_ACT_ALLOW
;
1446 default_action
= SCMP_ACT_ALLOW
;
1447 action
= negative_action
;
1450 if (needs_ambient_hack
) {
1451 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1456 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1459 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1463 if (set_isempty(c
->syscall_archs
))
1466 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1469 return seccomp_restrict_archs(c
->syscall_archs
);
1472 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1476 if (!context_has_address_families(c
))
1479 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1482 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1485 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1489 if (!c
->memory_deny_write_execute
)
1492 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1495 return seccomp_memory_deny_write_execute();
1498 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1502 if (!c
->restrict_realtime
)
1505 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1508 return seccomp_restrict_realtime();
1511 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1515 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1516 * let's protect even those systems where this is left on in the kernel. */
1518 if (!c
->protect_kernel_tunables
)
1521 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1524 return seccomp_protect_sysctl();
1527 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1531 /* Turn off module syscalls on ProtectKernelModules=yes */
1533 if (!c
->protect_kernel_modules
)
1536 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1539 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1542 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1546 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1548 if (!c
->private_devices
)
1551 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1554 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1557 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1561 if (!exec_context_restrict_namespaces_set(c
))
1564 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1567 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1570 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1571 unsigned long personality
;
1577 if (!c
->lock_personality
)
1580 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1583 personality
= c
->personality
;
1585 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1586 if (personality
== PERSONALITY_INVALID
) {
1588 r
= opinionated_personality(&personality
);
1593 return seccomp_lock_personality(personality
);
1598 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1601 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1602 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1604 if (idle_pipe
[0] >= 0) {
1607 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1609 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1612 /* Signal systemd that we are bored and want to continue. */
1613 n
= write(idle_pipe
[3], "x", 1);
1615 /* Wait for systemd to react to the signal above. */
1616 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1619 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1623 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1626 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1628 static int build_environment(
1630 const ExecContext
*c
,
1631 const ExecParameters
*p
,
1634 const char *username
,
1636 dev_t journal_stream_dev
,
1637 ino_t journal_stream_ino
,
1640 _cleanup_strv_free_
char **our_env
= NULL
;
1641 ExecDirectoryType t
;
1650 our_env
= new0(char*, 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1655 _cleanup_free_
char *joined
= NULL
;
1657 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1659 our_env
[n_env
++] = x
;
1661 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1663 our_env
[n_env
++] = x
;
1665 joined
= strv_join(p
->fd_names
, ":");
1669 x
= strjoin("LISTEN_FDNAMES=", joined
);
1672 our_env
[n_env
++] = x
;
1675 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1676 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1678 our_env
[n_env
++] = x
;
1680 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1682 our_env
[n_env
++] = x
;
1685 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1686 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1687 * check the database directly. */
1688 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1689 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1692 our_env
[n_env
++] = x
;
1696 x
= strappend("HOME=", home
);
1699 our_env
[n_env
++] = x
;
1703 x
= strappend("LOGNAME=", username
);
1706 our_env
[n_env
++] = x
;
1708 x
= strappend("USER=", username
);
1711 our_env
[n_env
++] = x
;
1715 x
= strappend("SHELL=", shell
);
1718 our_env
[n_env
++] = x
;
1721 if (!sd_id128_is_null(u
->invocation_id
)) {
1722 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1725 our_env
[n_env
++] = x
;
1728 if (exec_context_needs_term(c
)) {
1729 const char *tty_path
, *term
= NULL
;
1731 tty_path
= exec_context_tty_path(c
);
1733 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1734 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1735 * passes to PID 1 ends up all the way in the console login shown. */
1737 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1738 term
= getenv("TERM");
1740 term
= default_term_for_tty(tty_path
);
1742 x
= strappend("TERM=", term
);
1745 our_env
[n_env
++] = x
;
1748 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1749 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1752 our_env
[n_env
++] = x
;
1755 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1756 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1762 if (strv_isempty(c
->directories
[t
].paths
))
1765 n
= exec_directory_env_name_to_string(t
);
1769 pre
= strjoin(p
->prefix
[t
], "/");
1773 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1777 x
= strjoin(n
, "=", joined
);
1781 our_env
[n_env
++] = x
;
1784 our_env
[n_env
++] = NULL
;
1785 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1787 *ret
= TAKE_PTR(our_env
);
1792 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1793 _cleanup_strv_free_
char **pass_env
= NULL
;
1794 size_t n_env
= 0, n_bufsize
= 0;
1797 STRV_FOREACH(i
, c
->pass_environment
) {
1798 _cleanup_free_
char *x
= NULL
;
1804 x
= strjoin(*i
, "=", v
);
1808 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1811 pass_env
[n_env
++] = TAKE_PTR(x
);
1812 pass_env
[n_env
] = NULL
;
1815 *ret
= TAKE_PTR(pass_env
);
1820 static bool exec_needs_mount_namespace(
1821 const ExecContext
*context
,
1822 const ExecParameters
*params
,
1823 const ExecRuntime
*runtime
) {
1828 if (context
->root_image
)
1831 if (!strv_isempty(context
->read_write_paths
) ||
1832 !strv_isempty(context
->read_only_paths
) ||
1833 !strv_isempty(context
->inaccessible_paths
))
1836 if (context
->n_bind_mounts
> 0)
1839 if (context
->n_temporary_filesystems
> 0)
1842 if (context
->mount_flags
!= 0)
1845 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1848 if (context
->private_devices
||
1849 context
->private_mounts
||
1850 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1851 context
->protect_home
!= PROTECT_HOME_NO
||
1852 context
->protect_kernel_tunables
||
1853 context
->protect_kernel_modules
||
1854 context
->protect_control_groups
)
1857 if (context
->root_directory
) {
1858 ExecDirectoryType t
;
1860 if (context
->mount_apivfs
)
1863 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1864 if (!params
->prefix
[t
])
1867 if (!strv_isempty(context
->directories
[t
].paths
))
1872 if (context
->dynamic_user
&&
1873 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1874 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1875 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1881 static int setup_private_users(uid_t uid
, gid_t gid
) {
1882 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1883 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1884 _cleanup_close_
int unshare_ready_fd
= -1;
1885 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1890 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1891 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1892 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1893 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1894 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1895 * continues execution normally. */
1897 if (uid
!= 0 && uid_is_valid(uid
)) {
1898 r
= asprintf(&uid_map
,
1899 "0 0 1\n" /* Map root → root */
1900 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1905 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1910 if (gid
!= 0 && gid_is_valid(gid
)) {
1911 r
= asprintf(&gid_map
,
1912 "0 0 1\n" /* Map root → root */
1913 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1918 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1923 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1925 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1926 if (unshare_ready_fd
< 0)
1929 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1931 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1934 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1938 _cleanup_close_
int fd
= -1;
1942 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1943 * here, after the parent opened its own user namespace. */
1946 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1948 /* Wait until the parent unshared the user namespace */
1949 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1954 /* Disable the setgroups() system call in the child user namespace, for good. */
1955 a
= procfs_file_alloca(ppid
, "setgroups");
1956 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1958 if (errno
!= ENOENT
) {
1963 /* If the file is missing the kernel is too old, let's continue anyway. */
1965 if (write(fd
, "deny\n", 5) < 0) {
1970 fd
= safe_close(fd
);
1973 /* First write the GID map */
1974 a
= procfs_file_alloca(ppid
, "gid_map");
1975 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1980 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1984 fd
= safe_close(fd
);
1986 /* The write the UID map */
1987 a
= procfs_file_alloca(ppid
, "uid_map");
1988 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1993 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1998 _exit(EXIT_SUCCESS
);
2001 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2002 _exit(EXIT_FAILURE
);
2005 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2007 if (unshare(CLONE_NEWUSER
) < 0)
2010 /* Let the child know that the namespace is ready now */
2011 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2014 /* Try to read an error code from the child */
2015 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2018 if (n
== sizeof(r
)) { /* an error code was sent to us */
2023 if (n
!= 0) /* on success we should have read 0 bytes */
2026 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2030 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2036 static int setup_exec_directory(
2037 const ExecContext
*context
,
2038 const ExecParameters
*params
,
2041 ExecDirectoryType type
,
2044 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2045 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2046 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2047 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2048 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2049 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2056 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2057 assert(exit_status
);
2059 if (!params
->prefix
[type
])
2062 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2063 if (!uid_is_valid(uid
))
2065 if (!gid_is_valid(gid
))
2069 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2070 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2072 p
= strjoin(params
->prefix
[type
], "/", *rt
);
2078 r
= mkdir_parents_label(p
, 0755);
2082 if (context
->dynamic_user
&&
2083 !IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2084 _cleanup_free_
char *private_root
= NULL
;
2086 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2087 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2088 * whose UID is later on reused. To lock this down we use the same trick used by container
2089 * managers to prohibit host users to get access to files of the same UID in containers: we
2090 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2091 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2092 * to make this directory permeable for the service itself.
2094 * Specifically: for a service which wants a special directory "foo/" we first create a
2095 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2096 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2097 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2098 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2099 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2100 * disabling the access boundary for the service and making sure it only gets access to the
2101 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2103 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2104 * owned by the service itself.
2105 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2106 * files or sockets with other services. */
2108 private_root
= strjoin(params
->prefix
[type
], "/private");
2109 if (!private_root
) {
2114 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2115 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2119 pp
= strjoin(private_root
, "/", *rt
);
2125 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2126 r
= mkdir_parents_label(pp
, 0755);
2130 if (is_dir(p
, false) > 0 &&
2131 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2133 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2134 * it over. Most likely the service has been upgraded from one that didn't use
2135 * DynamicUser=1, to one that does. */
2137 if (rename(p
, pp
) < 0) {
2142 /* Otherwise, create the actual directory for the service */
2144 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2145 if (r
< 0 && r
!= -EEXIST
)
2149 /* And link it up from the original place */
2150 r
= symlink_idempotent(pp
, p
, true);
2154 /* Lock down the access mode */
2155 if (chmod(pp
, context
->directories
[type
].mode
) < 0) {
2160 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2161 if (r
< 0 && r
!= -EEXIST
)
2164 if (chmod(p
, context
->directories
[type
].mode
) < 0) {
2168 if (!context
->dynamic_user
)
2173 /* Don't change the owner of the configuration directory, as in the common case it is not written to by
2174 * a service, and shall not be writable. */
2175 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2178 /* Then, change the ownership of the whole tree, if necessary */
2179 r
= path_chown_recursive(pp
?: p
, uid
, gid
);
2187 *exit_status
= exit_status_table
[type
];
2192 static int setup_smack(
2193 const ExecContext
*context
,
2194 const ExecCommand
*command
) {
2201 if (context
->smack_process_label
) {
2202 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2206 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2208 _cleanup_free_
char *exec_label
= NULL
;
2210 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2211 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2214 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2224 static int compile_bind_mounts(
2225 const ExecContext
*context
,
2226 const ExecParameters
*params
,
2227 BindMount
**ret_bind_mounts
,
2228 size_t *ret_n_bind_mounts
,
2229 char ***ret_empty_directories
) {
2231 _cleanup_strv_free_
char **empty_directories
= NULL
;
2232 BindMount
*bind_mounts
;
2234 ExecDirectoryType t
;
2239 assert(ret_bind_mounts
);
2240 assert(ret_n_bind_mounts
);
2241 assert(ret_empty_directories
);
2243 n
= context
->n_bind_mounts
;
2244 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2245 if (!params
->prefix
[t
])
2248 n
+= strv_length(context
->directories
[t
].paths
);
2252 *ret_bind_mounts
= NULL
;
2253 *ret_n_bind_mounts
= 0;
2254 *ret_empty_directories
= NULL
;
2258 bind_mounts
= new(BindMount
, n
);
2262 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2263 BindMount
*item
= context
->bind_mounts
+ i
;
2266 s
= strdup(item
->source
);
2272 d
= strdup(item
->destination
);
2279 bind_mounts
[h
++] = (BindMount
) {
2282 .read_only
= item
->read_only
,
2283 .recursive
= item
->recursive
,
2284 .ignore_enoent
= item
->ignore_enoent
,
2288 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2291 if (!params
->prefix
[t
])
2294 if (strv_isempty(context
->directories
[t
].paths
))
2297 if (context
->dynamic_user
&&
2298 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2299 !(context
->root_directory
|| context
->root_image
)) {
2302 /* So this is for a dynamic user, and we need to make sure the process can access its own
2303 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2304 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2306 private_root
= strjoin(params
->prefix
[t
], "/private");
2307 if (!private_root
) {
2312 r
= strv_consume(&empty_directories
, private_root
);
2317 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2320 if (context
->dynamic_user
&&
2321 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2322 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2324 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2330 if (context
->dynamic_user
&&
2331 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2332 (context
->root_directory
|| context
->root_image
))
2333 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2334 * directory is not created on the root directory. So, let's bind-mount the directory
2335 * on the 'non-private' place. */
2336 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2345 bind_mounts
[h
++] = (BindMount
) {
2350 .ignore_enoent
= false,
2357 *ret_bind_mounts
= bind_mounts
;
2358 *ret_n_bind_mounts
= n
;
2359 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2364 bind_mount_free_many(bind_mounts
, h
);
2368 static int apply_mount_namespace(
2370 const ExecCommand
*command
,
2371 const ExecContext
*context
,
2372 const ExecParameters
*params
,
2373 const ExecRuntime
*runtime
) {
2375 _cleanup_strv_free_
char **empty_directories
= NULL
;
2376 char *tmp
= NULL
, *var
= NULL
;
2377 const char *root_dir
= NULL
, *root_image
= NULL
;
2378 NamespaceInfo ns_info
;
2379 bool needs_sandboxing
;
2380 BindMount
*bind_mounts
= NULL
;
2381 size_t n_bind_mounts
= 0;
2386 /* The runtime struct only contains the parent of the private /tmp,
2387 * which is non-accessible to world users. Inside of it there's a /tmp
2388 * that is sticky, and that's the one we want to use here. */
2390 if (context
->private_tmp
&& runtime
) {
2391 if (runtime
->tmp_dir
)
2392 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2393 if (runtime
->var_tmp_dir
)
2394 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2397 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2398 root_image
= context
->root_image
;
2401 root_dir
= context
->root_directory
;
2404 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2408 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2409 if (needs_sandboxing
)
2410 ns_info
= (NamespaceInfo
) {
2411 .ignore_protect_paths
= false,
2412 .private_dev
= context
->private_devices
,
2413 .protect_control_groups
= context
->protect_control_groups
,
2414 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2415 .protect_kernel_modules
= context
->protect_kernel_modules
,
2416 .mount_apivfs
= context
->mount_apivfs
,
2417 .private_mounts
= context
->private_mounts
,
2419 else if (!context
->dynamic_user
&& root_dir
)
2421 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2422 * sandbox info, otherwise enforce it, don't ignore protected paths and
2423 * fail if we are enable to apply the sandbox inside the mount namespace.
2425 ns_info
= (NamespaceInfo
) {
2426 .ignore_protect_paths
= true,
2429 ns_info
= (NamespaceInfo
) {};
2431 r
= setup_namespace(root_dir
, root_image
,
2432 &ns_info
, context
->read_write_paths
,
2433 needs_sandboxing
? context
->read_only_paths
: NULL
,
2434 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2438 context
->temporary_filesystems
,
2439 context
->n_temporary_filesystems
,
2442 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2443 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2444 context
->mount_flags
,
2445 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2447 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2449 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2450 * that with a special, recognizable error ENOANO. In this case, silently proceeed, but only if exclusively
2451 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2452 * completely different execution environment. */
2454 if (n_bind_mounts
== 0 &&
2455 context
->n_temporary_filesystems
== 0 &&
2456 !root_dir
&& !root_image
&&
2457 !context
->dynamic_user
) {
2458 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2462 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2463 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2464 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2472 static int apply_working_directory(
2473 const ExecContext
*context
,
2474 const ExecParameters
*params
,
2476 const bool needs_mount_ns
,
2482 assert(exit_status
);
2484 if (context
->working_directory_home
) {
2487 *exit_status
= EXIT_CHDIR
;
2493 } else if (context
->working_directory
)
2494 wd
= context
->working_directory
;
2498 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2499 if (!needs_mount_ns
&& context
->root_directory
)
2500 if (chroot(context
->root_directory
) < 0) {
2501 *exit_status
= EXIT_CHROOT
;
2507 d
= prefix_roota(context
->root_directory
, wd
);
2509 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2510 *exit_status
= EXIT_CHDIR
;
2517 static int setup_keyring(
2519 const ExecContext
*context
,
2520 const ExecParameters
*p
,
2521 uid_t uid
, gid_t gid
) {
2523 key_serial_t keyring
;
2532 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2533 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2534 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2535 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2536 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2537 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2539 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2542 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2543 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2544 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2545 * & group is just as nasty as acquiring a reference to the user keyring. */
2547 saved_uid
= getuid();
2548 saved_gid
= getgid();
2550 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2551 if (setregid(gid
, -1) < 0)
2552 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2555 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2556 if (setreuid(uid
, -1) < 0) {
2557 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2562 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2563 if (keyring
== -1) {
2564 if (errno
== ENOSYS
)
2565 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2566 else if (IN_SET(errno
, EACCES
, EPERM
))
2567 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2568 else if (errno
== EDQUOT
)
2569 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2571 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2576 /* When requested link the user keyring into the session keyring. */
2577 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2579 if (keyctl(KEYCTL_LINK
,
2580 KEY_SPEC_USER_KEYRING
,
2581 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2582 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2587 /* Restore uid/gid back */
2588 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2589 if (setreuid(saved_uid
, -1) < 0) {
2590 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2595 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2596 if (setregid(saved_gid
, -1) < 0)
2597 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2600 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2601 if (!sd_id128_is_null(u
->invocation_id
)) {
2604 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2606 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2608 if (keyctl(KEYCTL_SETPERM
, key
,
2609 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2610 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2611 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2616 /* Revert back uid & gid for the the last time, and exit */
2617 /* no extra logging, as only the first already reported error matters */
2618 if (getuid() != saved_uid
)
2619 (void) setreuid(saved_uid
, -1);
2621 if (getgid() != saved_gid
)
2622 (void) setregid(saved_gid
, -1);
2627 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2635 array
[(*n
)++] = pair
[0];
2637 array
[(*n
)++] = pair
[1];
2640 static int close_remaining_fds(
2641 const ExecParameters
*params
,
2642 const ExecRuntime
*runtime
,
2643 const DynamicCreds
*dcreds
,
2647 int *fds
, size_t n_fds
) {
2649 size_t n_dont_close
= 0;
2650 int dont_close
[n_fds
+ 12];
2654 if (params
->stdin_fd
>= 0)
2655 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2656 if (params
->stdout_fd
>= 0)
2657 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2658 if (params
->stderr_fd
>= 0)
2659 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2662 dont_close
[n_dont_close
++] = socket_fd
;
2664 dont_close
[n_dont_close
++] = exec_fd
;
2666 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2667 n_dont_close
+= n_fds
;
2671 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2675 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2677 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2680 if (user_lookup_fd
>= 0)
2681 dont_close
[n_dont_close
++] = user_lookup_fd
;
2683 return close_all_fds(dont_close
, n_dont_close
);
2686 static int send_user_lookup(
2694 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2695 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2698 if (user_lookup_fd
< 0)
2701 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2704 if (writev(user_lookup_fd
,
2706 IOVEC_INIT(&uid
, sizeof(uid
)),
2707 IOVEC_INIT(&gid
, sizeof(gid
)),
2708 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2714 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2721 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2726 if (!c
->working_directory_home
)
2730 /* Hardcode /root as home directory for UID 0 */
2735 r
= get_home_dir(buf
);
2743 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2744 _cleanup_strv_free_
char ** list
= NULL
;
2745 ExecDirectoryType t
;
2752 assert(c
->dynamic_user
);
2754 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2755 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2758 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2761 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2767 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2770 if (t
== EXEC_DIRECTORY_RUNTIME
)
2771 e
= strjoin(p
->prefix
[t
], "/", *i
);
2773 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2777 r
= strv_consume(&list
, e
);
2783 *ret
= TAKE_PTR(list
);
2788 static char *exec_command_line(char **argv
);
2790 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2791 bool using_subcgroup
;
2797 if (!params
->cgroup_path
)
2800 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2801 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2802 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2803 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2804 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2805 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2806 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2807 * flag, which is only passed for the former statements, not for the latter. */
2809 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2810 if (using_subcgroup
)
2811 p
= strjoin(params
->cgroup_path
, "/.control");
2813 p
= strdup(params
->cgroup_path
);
2818 return using_subcgroup
;
2821 static int exec_child(
2823 const ExecCommand
*command
,
2824 const ExecContext
*context
,
2825 const ExecParameters
*params
,
2826 ExecRuntime
*runtime
,
2827 DynamicCreds
*dcreds
,
2831 size_t n_socket_fds
,
2832 size_t n_storage_fds
,
2837 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2838 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2839 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2840 const char *username
= NULL
, *groupname
= NULL
;
2841 _cleanup_free_
char *home_buffer
= NULL
;
2842 const char *home
= NULL
, *shell
= NULL
;
2843 dev_t journal_stream_dev
= 0;
2844 ino_t journal_stream_ino
= 0;
2845 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2846 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2847 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2848 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2850 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2851 bool use_selinux
= false;
2854 bool use_smack
= false;
2857 bool use_apparmor
= false;
2859 uid_t uid
= UID_INVALID
;
2860 gid_t gid
= GID_INVALID
;
2862 ExecDirectoryType dt
;
2869 assert(exit_status
);
2871 rename_process_from_path(command
->path
);
2873 /* We reset exactly these signals, since they are the
2874 * only ones we set to SIG_IGN in the main daemon. All
2875 * others we leave untouched because we set them to
2876 * SIG_DFL or a valid handler initially, both of which
2877 * will be demoted to SIG_DFL. */
2878 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2879 SIGNALS_IGNORE
, -1);
2881 if (context
->ignore_sigpipe
)
2882 (void) ignore_signals(SIGPIPE
, -1);
2884 r
= reset_signal_mask();
2886 *exit_status
= EXIT_SIGNAL_MASK
;
2887 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2890 if (params
->idle_pipe
)
2891 do_idle_pipe_dance(params
->idle_pipe
);
2893 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2894 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2895 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2896 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2899 log_set_open_when_needed(true);
2901 /* In case anything used libc syslog(), close this here, too */
2904 n_fds
= n_socket_fds
+ n_storage_fds
;
2905 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2907 *exit_status
= EXIT_FDS
;
2908 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2911 if (!context
->same_pgrp
)
2913 *exit_status
= EXIT_SETSID
;
2914 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2917 exec_context_tty_reset(context
, params
);
2919 if (unit_shall_confirm_spawn(unit
)) {
2920 const char *vc
= params
->confirm_spawn
;
2921 _cleanup_free_
char *cmdline
= NULL
;
2923 cmdline
= exec_command_line(command
->argv
);
2925 *exit_status
= EXIT_MEMORY
;
2929 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2930 if (r
!= CONFIRM_EXECUTE
) {
2931 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2932 *exit_status
= EXIT_SUCCESS
;
2935 *exit_status
= EXIT_CONFIRM
;
2936 log_unit_error(unit
, "Execution cancelled by the user");
2941 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2942 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2943 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2944 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2945 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2946 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2947 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2948 *exit_status
= EXIT_MEMORY
;
2949 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2952 if (context
->dynamic_user
&& dcreds
) {
2953 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2955 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2956 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2957 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2958 *exit_status
= EXIT_USER
;
2959 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2962 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2964 *exit_status
= EXIT_MEMORY
;
2968 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2970 *exit_status
= EXIT_USER
;
2972 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2975 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2978 if (!uid_is_valid(uid
)) {
2979 *exit_status
= EXIT_USER
;
2980 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
2984 if (!gid_is_valid(gid
)) {
2985 *exit_status
= EXIT_USER
;
2986 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
2991 username
= dcreds
->user
->name
;
2994 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2996 *exit_status
= EXIT_USER
;
2997 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3000 r
= get_fixed_group(context
, &groupname
, &gid
);
3002 *exit_status
= EXIT_GROUP
;
3003 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3007 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3008 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3009 &supplementary_gids
, &ngids
);
3011 *exit_status
= EXIT_GROUP
;
3012 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3015 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3017 *exit_status
= EXIT_USER
;
3018 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3021 user_lookup_fd
= safe_close(user_lookup_fd
);
3023 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3025 *exit_status
= EXIT_CHDIR
;
3026 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3029 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3030 * must sure to drop O_NONBLOCK */
3032 (void) fd_nonblock(socket_fd
, false);
3034 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3035 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3036 if (params
->cgroup_path
) {
3037 _cleanup_free_
char *p
= NULL
;
3039 r
= exec_parameters_get_cgroup_path(params
, &p
);
3041 *exit_status
= EXIT_CGROUP
;
3042 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3045 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3047 *exit_status
= EXIT_CGROUP
;
3048 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3052 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3054 *exit_status
= EXIT_STDIN
;
3055 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3058 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3060 *exit_status
= EXIT_STDOUT
;
3061 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3064 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3066 *exit_status
= EXIT_STDERR
;
3067 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3070 if (context
->oom_score_adjust_set
) {
3071 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3072 * prohibit write access to this file, and we shouldn't trip up over that. */
3073 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3074 if (IN_SET(r
, -EPERM
, -EACCES
))
3075 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3077 *exit_status
= EXIT_OOM_ADJUST
;
3078 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3082 if (context
->nice_set
)
3083 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3084 *exit_status
= EXIT_NICE
;
3085 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3088 if (context
->cpu_sched_set
) {
3089 struct sched_param param
= {
3090 .sched_priority
= context
->cpu_sched_priority
,
3093 r
= sched_setscheduler(0,
3094 context
->cpu_sched_policy
|
3095 (context
->cpu_sched_reset_on_fork
?
3096 SCHED_RESET_ON_FORK
: 0),
3099 *exit_status
= EXIT_SETSCHEDULER
;
3100 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3104 if (context
->cpuset
)
3105 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3106 *exit_status
= EXIT_CPUAFFINITY
;
3107 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3110 if (context
->ioprio_set
)
3111 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3112 *exit_status
= EXIT_IOPRIO
;
3113 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3116 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3117 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3118 *exit_status
= EXIT_TIMERSLACK
;
3119 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3122 if (context
->personality
!= PERSONALITY_INVALID
) {
3123 r
= safe_personality(context
->personality
);
3125 *exit_status
= EXIT_PERSONALITY
;
3126 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3130 if (context
->utmp_id
)
3131 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3133 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3134 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3138 if (context
->user
) {
3139 r
= chown_terminal(STDIN_FILENO
, uid
);
3141 *exit_status
= EXIT_STDIN
;
3142 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3146 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3147 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3148 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3149 * touch a single hierarchy too. */
3150 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3151 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3153 *exit_status
= EXIT_CGROUP
;
3154 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3158 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3159 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3161 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3164 r
= build_environment(
3176 *exit_status
= EXIT_MEMORY
;
3180 r
= build_pass_environment(context
, &pass_env
);
3182 *exit_status
= EXIT_MEMORY
;
3186 accum_env
= strv_env_merge(5,
3187 params
->environment
,
3190 context
->environment
,
3194 *exit_status
= EXIT_MEMORY
;
3197 accum_env
= strv_env_clean(accum_env
);
3199 (void) umask(context
->umask
);
3201 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3203 *exit_status
= EXIT_KEYRING
;
3204 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3207 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3208 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3210 /* 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 */
3211 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3213 /* 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 */
3214 if (needs_ambient_hack
)
3215 needs_setuid
= false;
3217 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3219 if (needs_sandboxing
) {
3220 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3221 * present. The actual MAC context application will happen later, as late as possible, to avoid
3222 * impacting our own code paths. */
3225 use_selinux
= mac_selinux_use();
3228 use_smack
= mac_smack_use();
3231 use_apparmor
= mac_apparmor_use();
3235 if (needs_sandboxing
) {
3238 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3239 * is set here. (See below.) */
3241 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3243 *exit_status
= EXIT_LIMITS
;
3244 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3250 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3251 * wins here. (See above.) */
3253 if (context
->pam_name
&& username
) {
3254 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3256 *exit_status
= EXIT_PAM
;
3257 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3262 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3263 if (ns_type_supported(NAMESPACE_NET
)) {
3264 r
= setup_netns(runtime
->netns_storage_socket
);
3266 *exit_status
= EXIT_NETWORK
;
3267 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3270 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3273 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3274 if (needs_mount_namespace
) {
3275 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3277 *exit_status
= EXIT_NAMESPACE
;
3278 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3282 /* Drop groups as early as possbile */
3284 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3286 *exit_status
= EXIT_GROUP
;
3287 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3291 if (needs_sandboxing
) {
3293 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3294 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3296 *exit_status
= EXIT_SELINUX_CONTEXT
;
3297 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3302 if (context
->private_users
) {
3303 r
= setup_private_users(uid
, gid
);
3305 *exit_status
= EXIT_USER
;
3306 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3311 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3312 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3313 * however if we have it as we want to keep it open until the final execve(). */
3315 if (params
->exec_fd
>= 0) {
3316 exec_fd
= params
->exec_fd
;
3318 if (exec_fd
< 3 + (int) n_fds
) {
3321 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3322 * process we are about to execute. */
3324 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3326 *exit_status
= EXIT_FDS
;
3327 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3330 safe_close(exec_fd
);
3333 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3334 r
= fd_cloexec(exec_fd
, true);
3336 *exit_status
= EXIT_FDS
;
3337 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3341 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3342 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3343 fds_with_exec_fd
[n_fds
] = exec_fd
;
3344 n_fds_with_exec_fd
= n_fds
+ 1;
3346 fds_with_exec_fd
= fds
;
3347 n_fds_with_exec_fd
= n_fds
;
3350 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3352 r
= shift_fds(fds
, n_fds
);
3354 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3356 *exit_status
= EXIT_FDS
;
3357 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3360 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3361 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3362 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3365 secure_bits
= context
->secure_bits
;
3367 if (needs_sandboxing
) {
3370 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3371 * requested. (Note this is placed after the general resource limit initialization, see
3372 * above, in order to take precedence.) */
3373 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3374 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3375 *exit_status
= EXIT_LIMITS
;
3376 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3381 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3382 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3384 r
= setup_smack(context
, command
);
3386 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3387 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3392 bset
= context
->capability_bounding_set
;
3393 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3394 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3395 * instead of us doing that */
3396 if (needs_ambient_hack
)
3397 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3398 (UINT64_C(1) << CAP_SETUID
) |
3399 (UINT64_C(1) << CAP_SETGID
);
3401 if (!cap_test_all(bset
)) {
3402 r
= capability_bounding_set_drop(bset
, false);
3404 *exit_status
= EXIT_CAPABILITIES
;
3405 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3409 /* This is done before enforce_user, but ambient set
3410 * does not survive over setresuid() if keep_caps is not set. */
3411 if (!needs_ambient_hack
&&
3412 context
->capability_ambient_set
!= 0) {
3413 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3415 *exit_status
= EXIT_CAPABILITIES
;
3416 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3422 if (context
->user
) {
3423 r
= enforce_user(context
, uid
);
3425 *exit_status
= EXIT_USER
;
3426 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3429 if (!needs_ambient_hack
&&
3430 context
->capability_ambient_set
!= 0) {
3432 /* Fix the ambient capabilities after user change. */
3433 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3435 *exit_status
= EXIT_CAPABILITIES
;
3436 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3439 /* If we were asked to change user and ambient capabilities
3440 * were requested, we had to add keep-caps to the securebits
3441 * so that we would maintain the inherited capability set
3442 * through the setresuid(). Make sure that the bit is added
3443 * also to the context secure_bits so that we don't try to
3444 * drop the bit away next. */
3446 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3451 /* Apply working directory here, because the working directory might be on NFS and only the user running
3452 * this service might have the correct privilege to change to the working directory */
3453 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3455 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3457 if (needs_sandboxing
) {
3458 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3459 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3460 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3461 * are restricted. */
3465 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3468 r
= setexeccon(exec_context
);
3470 *exit_status
= EXIT_SELINUX_CONTEXT
;
3471 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3478 if (use_apparmor
&& context
->apparmor_profile
) {
3479 r
= aa_change_onexec(context
->apparmor_profile
);
3480 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3481 *exit_status
= EXIT_APPARMOR_PROFILE
;
3482 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3487 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3488 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3489 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3490 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3491 *exit_status
= EXIT_SECUREBITS
;
3492 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3495 if (context_has_no_new_privileges(context
))
3496 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3497 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3498 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3502 r
= apply_address_families(unit
, context
);
3504 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3505 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3508 r
= apply_memory_deny_write_execute(unit
, context
);
3510 *exit_status
= EXIT_SECCOMP
;
3511 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3514 r
= apply_restrict_realtime(unit
, context
);
3516 *exit_status
= EXIT_SECCOMP
;
3517 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3520 r
= apply_restrict_namespaces(unit
, context
);
3522 *exit_status
= EXIT_SECCOMP
;
3523 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3526 r
= apply_protect_sysctl(unit
, context
);
3528 *exit_status
= EXIT_SECCOMP
;
3529 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3532 r
= apply_protect_kernel_modules(unit
, context
);
3534 *exit_status
= EXIT_SECCOMP
;
3535 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3538 r
= apply_private_devices(unit
, context
);
3540 *exit_status
= EXIT_SECCOMP
;
3541 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3544 r
= apply_syscall_archs(unit
, context
);
3546 *exit_status
= EXIT_SECCOMP
;
3547 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3550 r
= apply_lock_personality(unit
, context
);
3552 *exit_status
= EXIT_SECCOMP
;
3553 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3556 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3557 * by the filter as little as possible. */
3558 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3560 *exit_status
= EXIT_SECCOMP
;
3561 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3566 if (!strv_isempty(context
->unset_environment
)) {
3569 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3571 *exit_status
= EXIT_MEMORY
;
3575 strv_free_and_replace(accum_env
, ee
);
3578 final_argv
= replace_env_argv(command
->argv
, accum_env
);
3580 *exit_status
= EXIT_MEMORY
;
3584 if (DEBUG_LOGGING
) {
3585 _cleanup_free_
char *line
;
3587 line
= exec_command_line(final_argv
);
3589 log_struct(LOG_DEBUG
,
3590 "EXECUTABLE=%s", command
->path
,
3591 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3593 LOG_UNIT_INVOCATION_ID(unit
));
3599 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3600 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3602 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3603 *exit_status
= EXIT_EXEC
;
3604 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3608 execve(command
->path
, final_argv
, accum_env
);
3614 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3615 * that POLLHUP on it no longer means execve() succeeded. */
3617 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3618 *exit_status
= EXIT_EXEC
;
3619 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3623 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3624 log_struct_errno(LOG_INFO
, r
,
3625 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3627 LOG_UNIT_INVOCATION_ID(unit
),
3628 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3630 "EXECUTABLE=%s", command
->path
);
3634 *exit_status
= EXIT_EXEC
;
3635 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3638 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3639 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3641 int exec_spawn(Unit
*unit
,
3642 ExecCommand
*command
,
3643 const ExecContext
*context
,
3644 const ExecParameters
*params
,
3645 ExecRuntime
*runtime
,
3646 DynamicCreds
*dcreds
,
3649 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3650 _cleanup_free_
char *subcgroup_path
= NULL
;
3651 _cleanup_strv_free_
char **files_env
= NULL
;
3652 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3653 _cleanup_free_
char *line
= NULL
;
3661 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3663 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3664 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3665 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3667 if (params
->n_socket_fds
> 1) {
3668 log_unit_error(unit
, "Got more than one socket.");
3672 if (params
->n_socket_fds
== 0) {
3673 log_unit_error(unit
, "Got no socket.");
3677 socket_fd
= params
->fds
[0];
3681 n_socket_fds
= params
->n_socket_fds
;
3682 n_storage_fds
= params
->n_storage_fds
;
3685 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3687 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3689 r
= exec_context_load_environment(unit
, context
, &files_env
);
3691 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3693 line
= exec_command_line(command
->argv
);
3697 log_struct(LOG_DEBUG
,
3698 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3699 "EXECUTABLE=%s", command
->path
,
3701 LOG_UNIT_INVOCATION_ID(unit
));
3703 if (params
->cgroup_path
) {
3704 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3706 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3707 if (r
> 0) { /* We are using a child cgroup */
3708 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3710 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3716 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3719 int exit_status
= EXIT_SUCCESS
;
3721 r
= exec_child(unit
,
3733 unit
->manager
->user_lookup_fds
[1],
3737 log_struct_errno(LOG_ERR
, r
,
3738 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3740 LOG_UNIT_INVOCATION_ID(unit
),
3741 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3742 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3744 "EXECUTABLE=%s", command
->path
);
3749 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3751 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3752 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3753 * process will be killed too). */
3755 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3757 exec_status_start(&command
->exec_status
, pid
);
3763 void exec_context_init(ExecContext
*c
) {
3764 ExecDirectoryType i
;
3769 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3770 c
->cpu_sched_policy
= SCHED_OTHER
;
3771 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3772 c
->syslog_level_prefix
= true;
3773 c
->ignore_sigpipe
= true;
3774 c
->timer_slack_nsec
= NSEC_INFINITY
;
3775 c
->personality
= PERSONALITY_INVALID
;
3776 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3777 c
->directories
[i
].mode
= 0755;
3778 c
->capability_bounding_set
= CAP_ALL
;
3779 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3780 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3781 c
->log_level_max
= -1;
3784 void exec_context_done(ExecContext
*c
) {
3785 ExecDirectoryType i
;
3790 c
->environment
= strv_free(c
->environment
);
3791 c
->environment_files
= strv_free(c
->environment_files
);
3792 c
->pass_environment
= strv_free(c
->pass_environment
);
3793 c
->unset_environment
= strv_free(c
->unset_environment
);
3795 rlimit_free_all(c
->rlimit
);
3797 for (l
= 0; l
< 3; l
++) {
3798 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3799 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3802 c
->working_directory
= mfree(c
->working_directory
);
3803 c
->root_directory
= mfree(c
->root_directory
);
3804 c
->root_image
= mfree(c
->root_image
);
3805 c
->tty_path
= mfree(c
->tty_path
);
3806 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3807 c
->user
= mfree(c
->user
);
3808 c
->group
= mfree(c
->group
);
3810 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3812 c
->pam_name
= mfree(c
->pam_name
);
3814 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3815 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3816 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3818 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3819 c
->bind_mounts
= NULL
;
3820 c
->n_bind_mounts
= 0;
3821 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3822 c
->temporary_filesystems
= NULL
;
3823 c
->n_temporary_filesystems
= 0;
3825 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3827 c
->utmp_id
= mfree(c
->utmp_id
);
3828 c
->selinux_context
= mfree(c
->selinux_context
);
3829 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3830 c
->smack_process_label
= mfree(c
->smack_process_label
);
3832 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3833 c
->syscall_archs
= set_free(c
->syscall_archs
);
3834 c
->address_families
= set_free(c
->address_families
);
3836 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3837 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3839 c
->log_level_max
= -1;
3841 exec_context_free_log_extra_fields(c
);
3843 c
->log_rate_limit_interval_usec
= 0;
3844 c
->log_rate_limit_burst
= 0;
3846 c
->stdin_data
= mfree(c
->stdin_data
);
3847 c
->stdin_data_size
= 0;
3850 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3855 if (!runtime_prefix
)
3858 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3859 _cleanup_free_
char *p
;
3861 p
= strjoin(runtime_prefix
, "/", *i
);
3865 /* We execute this synchronously, since we need to be sure this is gone when we start the service
3867 (void) rm_rf(p
, REMOVE_ROOT
);
3873 static void exec_command_done(ExecCommand
*c
) {
3876 c
->path
= mfree(c
->path
);
3877 c
->argv
= strv_free(c
->argv
);
3880 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3883 for (i
= 0; i
< n
; i
++)
3884 exec_command_done(c
+i
);
3887 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3891 LIST_REMOVE(command
, c
, i
);
3892 exec_command_done(i
);
3899 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3902 for (i
= 0; i
< n
; i
++)
3903 c
[i
] = exec_command_free_list(c
[i
]);
3906 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3909 for (i
= 0; i
< n
; i
++)
3910 exec_status_reset(&c
[i
].exec_status
);
3913 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3916 for (i
= 0; i
< n
; i
++) {
3919 LIST_FOREACH(command
, z
, c
[i
])
3920 exec_status_reset(&z
->exec_status
);
3924 typedef struct InvalidEnvInfo
{
3929 static void invalid_env(const char *p
, void *userdata
) {
3930 InvalidEnvInfo
*info
= userdata
;
3932 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3935 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3941 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3944 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3947 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3950 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3953 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3956 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3963 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]) {
3965 const char* stdio_fdname
[3];
3971 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3972 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3973 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3975 for (i
= 0; i
< 3; i
++)
3976 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3978 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3980 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3981 if (named_iofds
[STDIN_FILENO
] < 0 &&
3982 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3983 stdio_fdname
[STDIN_FILENO
] &&
3984 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3986 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3989 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3990 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3991 stdio_fdname
[STDOUT_FILENO
] &&
3992 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3994 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3997 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3998 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3999 stdio_fdname
[STDERR_FILENO
] &&
4000 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4002 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4006 return targets
== 0 ? 0 : -ENOENT
;
4009 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4010 char **i
, **r
= NULL
;
4015 STRV_FOREACH(i
, c
->environment_files
) {
4019 bool ignore
= false;
4021 _cleanup_globfree_ glob_t pglob
= {};
4030 if (!path_is_absolute(fn
)) {
4038 /* Filename supports globbing, take all matching files */
4039 k
= safe_glob(fn
, 0, &pglob
);
4048 /* When we don't match anything, -ENOENT should be returned */
4049 assert(pglob
.gl_pathc
> 0);
4051 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4052 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4060 /* Log invalid environment variables with filename */
4062 InvalidEnvInfo info
= {
4064 .path
= pglob
.gl_pathv
[n
]
4067 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4075 m
= strv_env_merge(2, r
, p
);
4091 static bool tty_may_match_dev_console(const char *tty
) {
4092 _cleanup_free_
char *resolved
= NULL
;
4097 tty
= skip_dev_prefix(tty
);
4099 /* trivial identity? */
4100 if (streq(tty
, "console"))
4103 if (resolve_dev_console(&resolved
) < 0)
4104 return true; /* if we could not resolve, assume it may */
4106 /* "tty0" means the active VC, so it may be the same sometimes */
4107 return streq(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4110 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4112 return (ec
->tty_reset
||
4114 ec
->tty_vt_disallocate
||
4115 is_terminal_input(ec
->std_input
) ||
4116 is_terminal_output(ec
->std_output
) ||
4117 is_terminal_output(ec
->std_error
)) &&
4118 tty_may_match_dev_console(exec_context_tty_path(ec
));
4121 static void strv_fprintf(FILE *f
, char **l
) {
4127 fprintf(f
, " %s", *g
);
4130 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4131 ExecDirectoryType dt
;
4139 prefix
= strempty(prefix
);
4143 "%sWorkingDirectory: %s\n"
4144 "%sRootDirectory: %s\n"
4145 "%sNonBlocking: %s\n"
4146 "%sPrivateTmp: %s\n"
4147 "%sPrivateDevices: %s\n"
4148 "%sProtectKernelTunables: %s\n"
4149 "%sProtectKernelModules: %s\n"
4150 "%sProtectControlGroups: %s\n"
4151 "%sPrivateNetwork: %s\n"
4152 "%sPrivateUsers: %s\n"
4153 "%sProtectHome: %s\n"
4154 "%sProtectSystem: %s\n"
4155 "%sMountAPIVFS: %s\n"
4156 "%sIgnoreSIGPIPE: %s\n"
4157 "%sMemoryDenyWriteExecute: %s\n"
4158 "%sRestrictRealtime: %s\n"
4159 "%sKeyringMode: %s\n",
4161 prefix
, c
->working_directory
? c
->working_directory
: "/",
4162 prefix
, c
->root_directory
? c
->root_directory
: "/",
4163 prefix
, yes_no(c
->non_blocking
),
4164 prefix
, yes_no(c
->private_tmp
),
4165 prefix
, yes_no(c
->private_devices
),
4166 prefix
, yes_no(c
->protect_kernel_tunables
),
4167 prefix
, yes_no(c
->protect_kernel_modules
),
4168 prefix
, yes_no(c
->protect_control_groups
),
4169 prefix
, yes_no(c
->private_network
),
4170 prefix
, yes_no(c
->private_users
),
4171 prefix
, protect_home_to_string(c
->protect_home
),
4172 prefix
, protect_system_to_string(c
->protect_system
),
4173 prefix
, yes_no(c
->mount_apivfs
),
4174 prefix
, yes_no(c
->ignore_sigpipe
),
4175 prefix
, yes_no(c
->memory_deny_write_execute
),
4176 prefix
, yes_no(c
->restrict_realtime
),
4177 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
));
4180 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4182 STRV_FOREACH(e
, c
->environment
)
4183 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4185 STRV_FOREACH(e
, c
->environment_files
)
4186 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4188 STRV_FOREACH(e
, c
->pass_environment
)
4189 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4191 STRV_FOREACH(e
, c
->unset_environment
)
4192 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4194 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4196 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4197 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4199 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4200 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4208 if (c
->oom_score_adjust_set
)
4210 "%sOOMScoreAdjust: %i\n",
4211 prefix
, c
->oom_score_adjust
);
4213 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4215 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4216 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4217 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4218 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4221 if (c
->ioprio_set
) {
4222 _cleanup_free_
char *class_str
= NULL
;
4224 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4226 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4228 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4231 if (c
->cpu_sched_set
) {
4232 _cleanup_free_
char *policy_str
= NULL
;
4234 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4236 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4239 "%sCPUSchedulingPriority: %i\n"
4240 "%sCPUSchedulingResetOnFork: %s\n",
4241 prefix
, c
->cpu_sched_priority
,
4242 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4246 fprintf(f
, "%sCPUAffinity:", prefix
);
4247 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4248 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4249 fprintf(f
, " %u", i
);
4253 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4254 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4257 "%sStandardInput: %s\n"
4258 "%sStandardOutput: %s\n"
4259 "%sStandardError: %s\n",
4260 prefix
, exec_input_to_string(c
->std_input
),
4261 prefix
, exec_output_to_string(c
->std_output
),
4262 prefix
, exec_output_to_string(c
->std_error
));
4264 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4265 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4266 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4267 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4268 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4269 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4271 if (c
->std_input
== EXEC_INPUT_FILE
)
4272 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4273 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4274 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4275 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4276 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4277 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4278 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4279 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4280 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4286 "%sTTYVHangup: %s\n"
4287 "%sTTYVTDisallocate: %s\n",
4288 prefix
, c
->tty_path
,
4289 prefix
, yes_no(c
->tty_reset
),
4290 prefix
, yes_no(c
->tty_vhangup
),
4291 prefix
, yes_no(c
->tty_vt_disallocate
));
4293 if (IN_SET(c
->std_output
,
4296 EXEC_OUTPUT_JOURNAL
,
4297 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4298 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4299 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4300 IN_SET(c
->std_error
,
4303 EXEC_OUTPUT_JOURNAL
,
4304 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4305 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4306 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4308 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4310 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4312 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4314 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4316 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4319 if (c
->log_level_max
>= 0) {
4320 _cleanup_free_
char *t
= NULL
;
4322 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4324 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4327 if (c
->log_rate_limit_interval_usec
> 0) {
4328 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4331 "%sLogRateLimitIntervalSec: %s\n",
4332 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4335 if (c
->log_rate_limit_burst
> 0)
4336 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4338 if (c
->n_log_extra_fields
> 0) {
4341 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4342 fprintf(f
, "%sLogExtraFields: ", prefix
);
4343 fwrite(c
->log_extra_fields
[j
].iov_base
,
4344 1, c
->log_extra_fields
[j
].iov_len
,
4350 if (c
->secure_bits
) {
4351 _cleanup_free_
char *str
= NULL
;
4353 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4355 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4358 if (c
->capability_bounding_set
!= CAP_ALL
) {
4359 _cleanup_free_
char *str
= NULL
;
4361 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4363 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4366 if (c
->capability_ambient_set
!= 0) {
4367 _cleanup_free_
char *str
= NULL
;
4369 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4371 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4375 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4377 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4379 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4381 if (!strv_isempty(c
->supplementary_groups
)) {
4382 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4383 strv_fprintf(f
, c
->supplementary_groups
);
4388 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4390 if (!strv_isempty(c
->read_write_paths
)) {
4391 fprintf(f
, "%sReadWritePaths:", prefix
);
4392 strv_fprintf(f
, c
->read_write_paths
);
4396 if (!strv_isempty(c
->read_only_paths
)) {
4397 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4398 strv_fprintf(f
, c
->read_only_paths
);
4402 if (!strv_isempty(c
->inaccessible_paths
)) {
4403 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4404 strv_fprintf(f
, c
->inaccessible_paths
);
4408 if (c
->n_bind_mounts
> 0)
4409 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4410 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4411 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4412 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4413 c
->bind_mounts
[i
].source
,
4414 c
->bind_mounts
[i
].destination
,
4415 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4417 if (c
->n_temporary_filesystems
> 0)
4418 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4419 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4421 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4423 isempty(t
->options
) ? "" : ":",
4424 strempty(t
->options
));
4429 "%sUtmpIdentifier: %s\n",
4430 prefix
, c
->utmp_id
);
4432 if (c
->selinux_context
)
4434 "%sSELinuxContext: %s%s\n",
4435 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4437 if (c
->apparmor_profile
)
4439 "%sAppArmorProfile: %s%s\n",
4440 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4442 if (c
->smack_process_label
)
4444 "%sSmackProcessLabel: %s%s\n",
4445 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4447 if (c
->personality
!= PERSONALITY_INVALID
)
4449 "%sPersonality: %s\n",
4450 prefix
, strna(personality_to_string(c
->personality
)));
4453 "%sLockPersonality: %s\n",
4454 prefix
, yes_no(c
->lock_personality
));
4456 if (c
->syscall_filter
) {
4464 "%sSystemCallFilter: ",
4467 if (!c
->syscall_whitelist
)
4471 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4472 _cleanup_free_
char *name
= NULL
;
4473 const char *errno_name
= NULL
;
4474 int num
= PTR_TO_INT(val
);
4481 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4482 fputs(strna(name
), f
);
4485 errno_name
= errno_to_name(num
);
4487 fprintf(f
, ":%s", errno_name
);
4489 fprintf(f
, ":%d", num
);
4497 if (c
->syscall_archs
) {
4504 "%sSystemCallArchitectures:",
4508 SET_FOREACH(id
, c
->syscall_archs
, j
)
4509 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4514 if (exec_context_restrict_namespaces_set(c
)) {
4515 _cleanup_free_
char *s
= NULL
;
4517 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4519 fprintf(f
, "%sRestrictNamespaces: %s\n",
4523 if (c
->syscall_errno
> 0) {
4524 const char *errno_name
;
4526 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4528 errno_name
= errno_to_name(c
->syscall_errno
);
4530 fprintf(f
, "%s\n", errno_name
);
4532 fprintf(f
, "%d\n", c
->syscall_errno
);
4535 if (c
->apparmor_profile
)
4537 "%sAppArmorProfile: %s%s\n",
4538 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4541 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4544 /* Returns true if the process forked off would run under
4545 * an unchanged UID or as root. */
4550 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4556 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4564 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4566 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4571 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4576 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4577 free(c
->log_extra_fields
[l
].iov_base
);
4578 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4579 c
->n_log_extra_fields
= 0;
4582 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4589 dual_timestamp_get(&s
->start_timestamp
);
4592 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4595 if (s
->pid
!= pid
) {
4601 dual_timestamp_get(&s
->exit_timestamp
);
4607 if (context
->utmp_id
)
4608 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4610 exec_context_tty_reset(context
, NULL
);
4614 void exec_status_reset(ExecStatus
*s
) {
4617 *s
= (ExecStatus
) {};
4620 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4621 char buf
[FORMAT_TIMESTAMP_MAX
];
4629 prefix
= strempty(prefix
);
4632 "%sPID: "PID_FMT
"\n",
4635 if (dual_timestamp_is_set(&s
->start_timestamp
))
4637 "%sStart Timestamp: %s\n",
4638 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4640 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4642 "%sExit Timestamp: %s\n"
4644 "%sExit Status: %i\n",
4645 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4646 prefix
, sigchld_code_to_string(s
->code
),
4650 static char *exec_command_line(char **argv
) {
4658 STRV_FOREACH(a
, argv
)
4666 STRV_FOREACH(a
, argv
) {
4673 if (strpbrk(*a
, WHITESPACE
)) {
4684 /* FIXME: this doesn't really handle arguments that have
4685 * spaces and ticks in them */
4690 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4691 _cleanup_free_
char *cmd
= NULL
;
4692 const char *prefix2
;
4697 prefix
= strempty(prefix
);
4698 prefix2
= strjoina(prefix
, "\t");
4700 cmd
= exec_command_line(c
->argv
);
4702 "%sCommand Line: %s\n",
4703 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4705 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4708 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4711 prefix
= strempty(prefix
);
4713 LIST_FOREACH(command
, c
, c
)
4714 exec_command_dump(c
, f
, prefix
);
4717 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4724 /* It's kind of important, that we keep the order here */
4725 LIST_FIND_TAIL(command
, *l
, end
);
4726 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4731 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4739 l
= strv_new_ap(path
, ap
);
4751 free_and_replace(c
->path
, p
);
4753 return strv_free_and_replace(c
->argv
, l
);
4756 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4757 _cleanup_strv_free_
char **l
= NULL
;
4765 l
= strv_new_ap(path
, ap
);
4771 r
= strv_extend_strv(&c
->argv
, l
, false);
4778 static void *remove_tmpdir_thread(void *p
) {
4779 _cleanup_free_
char *path
= p
;
4781 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4785 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4792 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4794 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4795 if (destroy
&& rt
->tmp_dir
) {
4796 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4798 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4800 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4807 if (destroy
&& rt
->var_tmp_dir
) {
4808 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4810 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4812 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4813 free(rt
->var_tmp_dir
);
4816 rt
->var_tmp_dir
= NULL
;
4819 rt
->id
= mfree(rt
->id
);
4820 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4821 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4822 safe_close_pair(rt
->netns_storage_socket
);
4826 static void exec_runtime_freep(ExecRuntime
**rt
) {
4828 (void) exec_runtime_free(*rt
, false);
4831 static int exec_runtime_allocate(ExecRuntime
**rt
) {
4834 *rt
= new0(ExecRuntime
, 1);
4838 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
4842 static int exec_runtime_add(
4845 const char *tmp_dir
,
4846 const char *var_tmp_dir
,
4847 const int netns_storage_socket
[2],
4848 ExecRuntime
**ret
) {
4850 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4856 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4860 r
= exec_runtime_allocate(&rt
);
4864 rt
->id
= strdup(id
);
4869 rt
->tmp_dir
= strdup(tmp_dir
);
4873 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4874 assert(var_tmp_dir
);
4875 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4876 if (!rt
->var_tmp_dir
)
4880 if (netns_storage_socket
) {
4881 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4882 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4885 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4894 /* do not remove created ExecRuntime object when the operation succeeds. */
4899 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4900 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4901 _cleanup_close_pair_
int netns_storage_socket
[2] = {-1, -1};
4908 /* It is not necessary to create ExecRuntime object. */
4909 if (!c
->private_network
&& !c
->private_tmp
)
4912 if (c
->private_tmp
) {
4913 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
4918 if (c
->private_network
) {
4919 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
4923 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
4928 netns_storage_socket
[0] = -1;
4929 netns_storage_socket
[1] = -1;
4933 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
4941 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
4943 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
4949 /* If not found, then create a new object. */
4950 r
= exec_runtime_make(m
, c
, id
, &rt
);
4952 /* When r == 0, it is not necessary to create ExecRuntime object. */
4956 /* increment reference counter. */
4962 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
4966 assert(rt
->n_ref
> 0);
4972 return exec_runtime_free(rt
, destroy
);
4975 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
4983 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
4984 fprintf(f
, "exec-runtime=%s", rt
->id
);
4987 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
4989 if (rt
->var_tmp_dir
)
4990 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
4992 if (rt
->netns_storage_socket
[0] >= 0) {
4995 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4999 fprintf(f
, " netns-socket-0=%i", copy
);
5002 if (rt
->netns_storage_socket
[1] >= 0) {
5005 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5009 fprintf(f
, " netns-socket-1=%i", copy
);
5018 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5019 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5023 /* This is for the migration from old (v237 or earlier) deserialization text.
5024 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5025 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5026 * so or not from the serialized text, then we always creates a new object owned by this. */
5032 /* Manager manages ExecRuntime objects by the unit id.
5033 * So, we omit the serialized text when the unit does not have id (yet?)... */
5034 if (isempty(u
->id
)) {
5035 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5039 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5041 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5045 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5047 r
= exec_runtime_allocate(&rt_create
);
5051 rt_create
->id
= strdup(u
->id
);
5058 if (streq(key
, "tmp-dir")) {
5061 copy
= strdup(value
);
5065 free_and_replace(rt
->tmp_dir
, copy
);
5067 } else if (streq(key
, "var-tmp-dir")) {
5070 copy
= strdup(value
);
5074 free_and_replace(rt
->var_tmp_dir
, copy
);
5076 } else if (streq(key
, "netns-socket-0")) {
5079 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5080 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5084 safe_close(rt
->netns_storage_socket
[0]);
5085 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5087 } else if (streq(key
, "netns-socket-1")) {
5090 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5091 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5095 safe_close(rt
->netns_storage_socket
[1]);
5096 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5100 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5102 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5104 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5108 rt_create
->manager
= u
->manager
;
5117 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5118 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5119 int r
, fd0
= -1, fd1
= -1;
5120 const char *p
, *v
= value
;
5127 n
= strcspn(v
, " ");
5128 id
= strndupa(v
, n
);
5133 v
= startswith(p
, "tmp-dir=");
5135 n
= strcspn(v
, " ");
5136 tmp_dir
= strndupa(v
, n
);
5142 v
= startswith(p
, "var-tmp-dir=");
5144 n
= strcspn(v
, " ");
5145 var_tmp_dir
= strndupa(v
, n
);
5151 v
= startswith(p
, "netns-socket-0=");
5155 n
= strcspn(v
, " ");
5156 buf
= strndupa(v
, n
);
5157 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5158 log_debug("Unable to process exec-runtime netns fd specification.");
5161 fd0
= fdset_remove(fds
, fd0
);
5167 v
= startswith(p
, "netns-socket-1=");
5171 n
= strcspn(v
, " ");
5172 buf
= strndupa(v
, n
);
5173 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5174 log_debug("Unable to process exec-runtime netns fd specification.");
5177 fd1
= fdset_remove(fds
, fd1
);
5182 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5184 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5187 void exec_runtime_vacuum(Manager
*m
) {
5193 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5195 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5199 (void) exec_runtime_free(rt
, false);
5203 void exec_params_clear(ExecParameters
*p
) {
5207 strv_free(p
->environment
);
5210 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5211 [EXEC_INPUT_NULL
] = "null",
5212 [EXEC_INPUT_TTY
] = "tty",
5213 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5214 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5215 [EXEC_INPUT_SOCKET
] = "socket",
5216 [EXEC_INPUT_NAMED_FD
] = "fd",
5217 [EXEC_INPUT_DATA
] = "data",
5218 [EXEC_INPUT_FILE
] = "file",
5221 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5223 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5224 [EXEC_OUTPUT_INHERIT
] = "inherit",
5225 [EXEC_OUTPUT_NULL
] = "null",
5226 [EXEC_OUTPUT_TTY
] = "tty",
5227 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5228 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5229 [EXEC_OUTPUT_KMSG
] = "kmsg",
5230 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5231 [EXEC_OUTPUT_JOURNAL
] = "journal",
5232 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5233 [EXEC_OUTPUT_SOCKET
] = "socket",
5234 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5235 [EXEC_OUTPUT_FILE
] = "file",
5236 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5239 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5241 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5242 [EXEC_UTMP_INIT
] = "init",
5243 [EXEC_UTMP_LOGIN
] = "login",
5244 [EXEC_UTMP_USER
] = "user",
5247 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5249 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5250 [EXEC_PRESERVE_NO
] = "no",
5251 [EXEC_PRESERVE_YES
] = "yes",
5252 [EXEC_PRESERVE_RESTART
] = "restart",
5255 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5257 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5258 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5259 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5260 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5261 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5262 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5265 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5267 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5268 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5269 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5270 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5271 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5272 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5275 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5277 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5278 [EXEC_KEYRING_INHERIT
] = "inherit",
5279 [EXEC_KEYRING_PRIVATE
] = "private",
5280 [EXEC_KEYRING_SHARED
] = "shared",
5283 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);