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
||
1414 c
->protect_hostname
;
1419 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1421 if (is_seccomp_available())
1424 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1428 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1429 uint32_t negative_action
, default_action
, action
;
1435 if (!context_has_syscall_filters(c
))
1438 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1441 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1443 if (c
->syscall_whitelist
) {
1444 default_action
= negative_action
;
1445 action
= SCMP_ACT_ALLOW
;
1447 default_action
= SCMP_ACT_ALLOW
;
1448 action
= negative_action
;
1451 if (needs_ambient_hack
) {
1452 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1457 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1460 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1464 if (set_isempty(c
->syscall_archs
))
1467 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1470 return seccomp_restrict_archs(c
->syscall_archs
);
1473 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1477 if (!context_has_address_families(c
))
1480 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1483 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1486 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1490 if (!c
->memory_deny_write_execute
)
1493 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1496 return seccomp_memory_deny_write_execute();
1499 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1503 if (!c
->restrict_realtime
)
1506 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1509 return seccomp_restrict_realtime();
1512 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1516 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1517 * let's protect even those systems where this is left on in the kernel. */
1519 if (!c
->protect_kernel_tunables
)
1522 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1525 return seccomp_protect_sysctl();
1528 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1532 /* Turn off module syscalls on ProtectKernelModules=yes */
1534 if (!c
->protect_kernel_modules
)
1537 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1540 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1543 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1547 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1549 if (!c
->private_devices
)
1552 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1555 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1558 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1562 if (!exec_context_restrict_namespaces_set(c
))
1565 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1568 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1571 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1572 unsigned long personality
;
1578 if (!c
->lock_personality
)
1581 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1584 personality
= c
->personality
;
1586 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1587 if (personality
== PERSONALITY_INVALID
) {
1589 r
= opinionated_personality(&personality
);
1594 return seccomp_lock_personality(personality
);
1599 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1602 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1603 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1605 if (idle_pipe
[0] >= 0) {
1608 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1610 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1613 /* Signal systemd that we are bored and want to continue. */
1614 n
= write(idle_pipe
[3], "x", 1);
1616 /* Wait for systemd to react to the signal above. */
1617 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1620 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1624 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1627 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1629 static int build_environment(
1631 const ExecContext
*c
,
1632 const ExecParameters
*p
,
1635 const char *username
,
1637 dev_t journal_stream_dev
,
1638 ino_t journal_stream_ino
,
1641 _cleanup_strv_free_
char **our_env
= NULL
;
1642 ExecDirectoryType t
;
1651 our_env
= new0(char*, 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1656 _cleanup_free_
char *joined
= NULL
;
1658 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1660 our_env
[n_env
++] = x
;
1662 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1664 our_env
[n_env
++] = x
;
1666 joined
= strv_join(p
->fd_names
, ":");
1670 x
= strjoin("LISTEN_FDNAMES=", joined
);
1673 our_env
[n_env
++] = x
;
1676 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1677 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1679 our_env
[n_env
++] = x
;
1681 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1683 our_env
[n_env
++] = x
;
1686 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1687 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1688 * check the database directly. */
1689 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1690 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1693 our_env
[n_env
++] = x
;
1697 x
= strappend("HOME=", home
);
1700 our_env
[n_env
++] = x
;
1704 x
= strappend("LOGNAME=", username
);
1707 our_env
[n_env
++] = x
;
1709 x
= strappend("USER=", username
);
1712 our_env
[n_env
++] = x
;
1716 x
= strappend("SHELL=", shell
);
1719 our_env
[n_env
++] = x
;
1722 if (!sd_id128_is_null(u
->invocation_id
)) {
1723 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1726 our_env
[n_env
++] = x
;
1729 if (exec_context_needs_term(c
)) {
1730 const char *tty_path
, *term
= NULL
;
1732 tty_path
= exec_context_tty_path(c
);
1734 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1735 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1736 * passes to PID 1 ends up all the way in the console login shown. */
1738 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1739 term
= getenv("TERM");
1741 term
= default_term_for_tty(tty_path
);
1743 x
= strappend("TERM=", term
);
1746 our_env
[n_env
++] = x
;
1749 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1750 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1753 our_env
[n_env
++] = x
;
1756 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1757 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1763 if (strv_isempty(c
->directories
[t
].paths
))
1766 n
= exec_directory_env_name_to_string(t
);
1770 pre
= strjoin(p
->prefix
[t
], "/");
1774 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1778 x
= strjoin(n
, "=", joined
);
1782 our_env
[n_env
++] = x
;
1785 our_env
[n_env
++] = NULL
;
1786 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1788 *ret
= TAKE_PTR(our_env
);
1793 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1794 _cleanup_strv_free_
char **pass_env
= NULL
;
1795 size_t n_env
= 0, n_bufsize
= 0;
1798 STRV_FOREACH(i
, c
->pass_environment
) {
1799 _cleanup_free_
char *x
= NULL
;
1805 x
= strjoin(*i
, "=", v
);
1809 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1812 pass_env
[n_env
++] = TAKE_PTR(x
);
1813 pass_env
[n_env
] = NULL
;
1816 *ret
= TAKE_PTR(pass_env
);
1821 static bool exec_needs_mount_namespace(
1822 const ExecContext
*context
,
1823 const ExecParameters
*params
,
1824 const ExecRuntime
*runtime
) {
1829 if (context
->root_image
)
1832 if (!strv_isempty(context
->read_write_paths
) ||
1833 !strv_isempty(context
->read_only_paths
) ||
1834 !strv_isempty(context
->inaccessible_paths
))
1837 if (context
->n_bind_mounts
> 0)
1840 if (context
->n_temporary_filesystems
> 0)
1843 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
1846 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1849 if (context
->private_devices
||
1850 context
->private_mounts
||
1851 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1852 context
->protect_home
!= PROTECT_HOME_NO
||
1853 context
->protect_kernel_tunables
||
1854 context
->protect_kernel_modules
||
1855 context
->protect_control_groups
)
1858 if (context
->root_directory
) {
1859 ExecDirectoryType t
;
1861 if (context
->mount_apivfs
)
1864 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1865 if (!params
->prefix
[t
])
1868 if (!strv_isempty(context
->directories
[t
].paths
))
1873 if (context
->dynamic_user
&&
1874 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1875 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1876 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1882 static int setup_private_users(uid_t uid
, gid_t gid
) {
1883 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1884 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1885 _cleanup_close_
int unshare_ready_fd
= -1;
1886 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1891 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1892 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1893 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1894 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1895 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1896 * continues execution normally. */
1898 if (uid
!= 0 && uid_is_valid(uid
)) {
1899 r
= asprintf(&uid_map
,
1900 "0 0 1\n" /* Map root → root */
1901 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1906 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1911 if (gid
!= 0 && gid_is_valid(gid
)) {
1912 r
= asprintf(&gid_map
,
1913 "0 0 1\n" /* Map root → root */
1914 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1919 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1924 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1926 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1927 if (unshare_ready_fd
< 0)
1930 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1932 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1935 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1939 _cleanup_close_
int fd
= -1;
1943 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1944 * here, after the parent opened its own user namespace. */
1947 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1949 /* Wait until the parent unshared the user namespace */
1950 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1955 /* Disable the setgroups() system call in the child user namespace, for good. */
1956 a
= procfs_file_alloca(ppid
, "setgroups");
1957 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1959 if (errno
!= ENOENT
) {
1964 /* If the file is missing the kernel is too old, let's continue anyway. */
1966 if (write(fd
, "deny\n", 5) < 0) {
1971 fd
= safe_close(fd
);
1974 /* First write the GID map */
1975 a
= procfs_file_alloca(ppid
, "gid_map");
1976 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1981 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1985 fd
= safe_close(fd
);
1987 /* The write the UID map */
1988 a
= procfs_file_alloca(ppid
, "uid_map");
1989 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1994 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1999 _exit(EXIT_SUCCESS
);
2002 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2003 _exit(EXIT_FAILURE
);
2006 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2008 if (unshare(CLONE_NEWUSER
) < 0)
2011 /* Let the child know that the namespace is ready now */
2012 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2015 /* Try to read an error code from the child */
2016 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2019 if (n
== sizeof(r
)) { /* an error code was sent to us */
2024 if (n
!= 0) /* on success we should have read 0 bytes */
2027 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2031 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2037 static int setup_exec_directory(
2038 const ExecContext
*context
,
2039 const ExecParameters
*params
,
2042 ExecDirectoryType type
,
2045 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2046 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2047 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2048 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2049 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2050 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2057 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2058 assert(exit_status
);
2060 if (!params
->prefix
[type
])
2063 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2064 if (!uid_is_valid(uid
))
2066 if (!gid_is_valid(gid
))
2070 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2071 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2073 p
= strjoin(params
->prefix
[type
], "/", *rt
);
2079 r
= mkdir_parents_label(p
, 0755);
2083 if (context
->dynamic_user
&&
2084 !IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2085 _cleanup_free_
char *private_root
= NULL
;
2087 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2088 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2089 * whose UID is later on reused. To lock this down we use the same trick used by container
2090 * managers to prohibit host users to get access to files of the same UID in containers: we
2091 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2092 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2093 * to make this directory permeable for the service itself.
2095 * Specifically: for a service which wants a special directory "foo/" we first create a
2096 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2097 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2098 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2099 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2100 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2101 * disabling the access boundary for the service and making sure it only gets access to the
2102 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2104 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2105 * owned by the service itself.
2106 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2107 * files or sockets with other services. */
2109 private_root
= strjoin(params
->prefix
[type
], "/private");
2110 if (!private_root
) {
2115 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2116 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2120 pp
= strjoin(private_root
, "/", *rt
);
2126 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2127 r
= mkdir_parents_label(pp
, 0755);
2131 if (is_dir(p
, false) > 0 &&
2132 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2134 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2135 * it over. Most likely the service has been upgraded from one that didn't use
2136 * DynamicUser=1, to one that does. */
2138 if (rename(p
, pp
) < 0) {
2143 /* Otherwise, create the actual directory for the service */
2145 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2146 if (r
< 0 && r
!= -EEXIST
)
2150 /* And link it up from the original place */
2151 r
= symlink_idempotent(pp
, p
, true);
2155 /* Lock down the access mode */
2156 if (chmod(pp
, context
->directories
[type
].mode
) < 0) {
2161 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2162 if (r
< 0 && r
!= -EEXIST
)
2167 if (stat(p
, &st
) < 0) {
2171 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2172 log_warning("%s \'%s\' already exists but the mode is different. "
2173 "(filesystem: %o %sMode: %o)",
2174 exec_directory_type_to_string(type
), *rt
,
2175 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2176 if (!context
->dynamic_user
)
2181 /* Don't change the owner of the configuration directory, as in the common case it is not written to by
2182 * a service, and shall not be writable. */
2183 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2186 /* Then, change the ownership of the whole tree, if necessary */
2187 r
= path_chown_recursive(pp
?: p
, uid
, gid
);
2195 *exit_status
= exit_status_table
[type
];
2200 static int setup_smack(
2201 const ExecContext
*context
,
2202 const ExecCommand
*command
) {
2209 if (context
->smack_process_label
) {
2210 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2214 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2216 _cleanup_free_
char *exec_label
= NULL
;
2218 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2219 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2222 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2232 static int compile_bind_mounts(
2233 const ExecContext
*context
,
2234 const ExecParameters
*params
,
2235 BindMount
**ret_bind_mounts
,
2236 size_t *ret_n_bind_mounts
,
2237 char ***ret_empty_directories
) {
2239 _cleanup_strv_free_
char **empty_directories
= NULL
;
2240 BindMount
*bind_mounts
;
2242 ExecDirectoryType t
;
2247 assert(ret_bind_mounts
);
2248 assert(ret_n_bind_mounts
);
2249 assert(ret_empty_directories
);
2251 n
= context
->n_bind_mounts
;
2252 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2253 if (!params
->prefix
[t
])
2256 n
+= strv_length(context
->directories
[t
].paths
);
2260 *ret_bind_mounts
= NULL
;
2261 *ret_n_bind_mounts
= 0;
2262 *ret_empty_directories
= NULL
;
2266 bind_mounts
= new(BindMount
, n
);
2270 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2271 BindMount
*item
= context
->bind_mounts
+ i
;
2274 s
= strdup(item
->source
);
2280 d
= strdup(item
->destination
);
2287 bind_mounts
[h
++] = (BindMount
) {
2290 .read_only
= item
->read_only
,
2291 .recursive
= item
->recursive
,
2292 .ignore_enoent
= item
->ignore_enoent
,
2296 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2299 if (!params
->prefix
[t
])
2302 if (strv_isempty(context
->directories
[t
].paths
))
2305 if (context
->dynamic_user
&&
2306 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2307 !(context
->root_directory
|| context
->root_image
)) {
2310 /* So this is for a dynamic user, and we need to make sure the process can access its own
2311 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2312 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2314 private_root
= strjoin(params
->prefix
[t
], "/private");
2315 if (!private_root
) {
2320 r
= strv_consume(&empty_directories
, private_root
);
2325 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2328 if (context
->dynamic_user
&&
2329 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2330 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2332 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2338 if (context
->dynamic_user
&&
2339 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2340 (context
->root_directory
|| context
->root_image
))
2341 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2342 * directory is not created on the root directory. So, let's bind-mount the directory
2343 * on the 'non-private' place. */
2344 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2353 bind_mounts
[h
++] = (BindMount
) {
2358 .ignore_enoent
= false,
2365 *ret_bind_mounts
= bind_mounts
;
2366 *ret_n_bind_mounts
= n
;
2367 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2372 bind_mount_free_many(bind_mounts
, h
);
2376 static int apply_mount_namespace(
2378 const ExecCommand
*command
,
2379 const ExecContext
*context
,
2380 const ExecParameters
*params
,
2381 const ExecRuntime
*runtime
) {
2383 _cleanup_strv_free_
char **empty_directories
= NULL
;
2384 char *tmp
= NULL
, *var
= NULL
;
2385 const char *root_dir
= NULL
, *root_image
= NULL
;
2386 NamespaceInfo ns_info
;
2387 bool needs_sandboxing
;
2388 BindMount
*bind_mounts
= NULL
;
2389 size_t n_bind_mounts
= 0;
2394 /* The runtime struct only contains the parent of the private /tmp,
2395 * which is non-accessible to world users. Inside of it there's a /tmp
2396 * that is sticky, and that's the one we want to use here. */
2398 if (context
->private_tmp
&& runtime
) {
2399 if (runtime
->tmp_dir
)
2400 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2401 if (runtime
->var_tmp_dir
)
2402 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2405 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2406 root_image
= context
->root_image
;
2409 root_dir
= context
->root_directory
;
2412 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2416 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2417 if (needs_sandboxing
)
2418 ns_info
= (NamespaceInfo
) {
2419 .ignore_protect_paths
= false,
2420 .private_dev
= context
->private_devices
,
2421 .protect_control_groups
= context
->protect_control_groups
,
2422 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2423 .protect_kernel_modules
= context
->protect_kernel_modules
,
2424 .protect_hostname
= context
->protect_hostname
,
2425 .mount_apivfs
= context
->mount_apivfs
,
2426 .private_mounts
= context
->private_mounts
,
2428 else if (!context
->dynamic_user
&& root_dir
)
2430 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2431 * sandbox info, otherwise enforce it, don't ignore protected paths and
2432 * fail if we are enable to apply the sandbox inside the mount namespace.
2434 ns_info
= (NamespaceInfo
) {
2435 .ignore_protect_paths
= true,
2438 ns_info
= (NamespaceInfo
) {};
2440 if (context
->mount_flags
== MS_SHARED
)
2441 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2443 r
= setup_namespace(root_dir
, root_image
,
2444 &ns_info
, context
->read_write_paths
,
2445 needs_sandboxing
? context
->read_only_paths
: NULL
,
2446 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2450 context
->temporary_filesystems
,
2451 context
->n_temporary_filesystems
,
2454 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2455 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2456 context
->mount_flags
,
2457 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2459 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2461 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2462 * that with a special, recognizable error ENOANO. In this case, silently proceeed, but only if exclusively
2463 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2464 * completely different execution environment. */
2466 if (n_bind_mounts
== 0 &&
2467 context
->n_temporary_filesystems
== 0 &&
2468 !root_dir
&& !root_image
&&
2469 !context
->dynamic_user
) {
2470 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2474 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2475 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2476 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2484 static int apply_working_directory(
2485 const ExecContext
*context
,
2486 const ExecParameters
*params
,
2488 const bool needs_mount_ns
,
2494 assert(exit_status
);
2496 if (context
->working_directory_home
) {
2499 *exit_status
= EXIT_CHDIR
;
2505 } else if (context
->working_directory
)
2506 wd
= context
->working_directory
;
2510 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2511 if (!needs_mount_ns
&& context
->root_directory
)
2512 if (chroot(context
->root_directory
) < 0) {
2513 *exit_status
= EXIT_CHROOT
;
2519 d
= prefix_roota(context
->root_directory
, wd
);
2521 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2522 *exit_status
= EXIT_CHDIR
;
2529 static int setup_keyring(
2531 const ExecContext
*context
,
2532 const ExecParameters
*p
,
2533 uid_t uid
, gid_t gid
) {
2535 key_serial_t keyring
;
2544 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2545 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2546 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2547 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2548 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2549 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2551 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2554 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2555 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2556 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2557 * & group is just as nasty as acquiring a reference to the user keyring. */
2559 saved_uid
= getuid();
2560 saved_gid
= getgid();
2562 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2563 if (setregid(gid
, -1) < 0)
2564 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2567 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2568 if (setreuid(uid
, -1) < 0) {
2569 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2574 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2575 if (keyring
== -1) {
2576 if (errno
== ENOSYS
)
2577 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2578 else if (IN_SET(errno
, EACCES
, EPERM
))
2579 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2580 else if (errno
== EDQUOT
)
2581 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2583 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2588 /* When requested link the user keyring into the session keyring. */
2589 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2591 if (keyctl(KEYCTL_LINK
,
2592 KEY_SPEC_USER_KEYRING
,
2593 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2594 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2599 /* Restore uid/gid back */
2600 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2601 if (setreuid(saved_uid
, -1) < 0) {
2602 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2607 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2608 if (setregid(saved_gid
, -1) < 0)
2609 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2612 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2613 if (!sd_id128_is_null(u
->invocation_id
)) {
2616 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2618 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2620 if (keyctl(KEYCTL_SETPERM
, key
,
2621 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2622 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2623 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2628 /* Revert back uid & gid for the the last time, and exit */
2629 /* no extra logging, as only the first already reported error matters */
2630 if (getuid() != saved_uid
)
2631 (void) setreuid(saved_uid
, -1);
2633 if (getgid() != saved_gid
)
2634 (void) setregid(saved_gid
, -1);
2639 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2647 array
[(*n
)++] = pair
[0];
2649 array
[(*n
)++] = pair
[1];
2652 static int close_remaining_fds(
2653 const ExecParameters
*params
,
2654 const ExecRuntime
*runtime
,
2655 const DynamicCreds
*dcreds
,
2659 int *fds
, size_t n_fds
) {
2661 size_t n_dont_close
= 0;
2662 int dont_close
[n_fds
+ 12];
2666 if (params
->stdin_fd
>= 0)
2667 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2668 if (params
->stdout_fd
>= 0)
2669 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2670 if (params
->stderr_fd
>= 0)
2671 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2674 dont_close
[n_dont_close
++] = socket_fd
;
2676 dont_close
[n_dont_close
++] = exec_fd
;
2678 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2679 n_dont_close
+= n_fds
;
2683 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2687 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2689 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2692 if (user_lookup_fd
>= 0)
2693 dont_close
[n_dont_close
++] = user_lookup_fd
;
2695 return close_all_fds(dont_close
, n_dont_close
);
2698 static int send_user_lookup(
2706 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2707 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2710 if (user_lookup_fd
< 0)
2713 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2716 if (writev(user_lookup_fd
,
2718 IOVEC_INIT(&uid
, sizeof(uid
)),
2719 IOVEC_INIT(&gid
, sizeof(gid
)),
2720 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2726 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2733 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2738 if (!c
->working_directory_home
)
2742 /* Hardcode /root as home directory for UID 0 */
2747 r
= get_home_dir(buf
);
2755 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2756 _cleanup_strv_free_
char ** list
= NULL
;
2757 ExecDirectoryType t
;
2764 assert(c
->dynamic_user
);
2766 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2767 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2770 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2773 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2779 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2782 if (t
== EXEC_DIRECTORY_RUNTIME
)
2783 e
= strjoin(p
->prefix
[t
], "/", *i
);
2785 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2789 r
= strv_consume(&list
, e
);
2795 *ret
= TAKE_PTR(list
);
2800 static char *exec_command_line(char **argv
);
2802 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2803 bool using_subcgroup
;
2809 if (!params
->cgroup_path
)
2812 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2813 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2814 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2815 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2816 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2817 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2818 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2819 * flag, which is only passed for the former statements, not for the latter. */
2821 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2822 if (using_subcgroup
)
2823 p
= strjoin(params
->cgroup_path
, "/.control");
2825 p
= strdup(params
->cgroup_path
);
2830 return using_subcgroup
;
2833 static int exec_child(
2835 const ExecCommand
*command
,
2836 const ExecContext
*context
,
2837 const ExecParameters
*params
,
2838 ExecRuntime
*runtime
,
2839 DynamicCreds
*dcreds
,
2843 size_t n_socket_fds
,
2844 size_t n_storage_fds
,
2849 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
2850 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2851 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2852 const char *username
= NULL
, *groupname
= NULL
;
2853 _cleanup_free_
char *home_buffer
= NULL
;
2854 const char *home
= NULL
, *shell
= NULL
;
2855 char **final_argv
= NULL
;
2856 dev_t journal_stream_dev
= 0;
2857 ino_t journal_stream_ino
= 0;
2858 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2859 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2860 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2861 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2863 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2864 bool use_selinux
= false;
2867 bool use_smack
= false;
2870 bool use_apparmor
= false;
2872 uid_t uid
= UID_INVALID
;
2873 gid_t gid
= GID_INVALID
;
2875 ExecDirectoryType dt
;
2882 assert(exit_status
);
2884 rename_process_from_path(command
->path
);
2886 /* We reset exactly these signals, since they are the
2887 * only ones we set to SIG_IGN in the main daemon. All
2888 * others we leave untouched because we set them to
2889 * SIG_DFL or a valid handler initially, both of which
2890 * will be demoted to SIG_DFL. */
2891 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2892 SIGNALS_IGNORE
, -1);
2894 if (context
->ignore_sigpipe
)
2895 (void) ignore_signals(SIGPIPE
, -1);
2897 r
= reset_signal_mask();
2899 *exit_status
= EXIT_SIGNAL_MASK
;
2900 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2903 if (params
->idle_pipe
)
2904 do_idle_pipe_dance(params
->idle_pipe
);
2906 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2907 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2908 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2909 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2912 log_set_open_when_needed(true);
2914 /* In case anything used libc syslog(), close this here, too */
2917 n_fds
= n_socket_fds
+ n_storage_fds
;
2918 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2920 *exit_status
= EXIT_FDS
;
2921 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2924 if (!context
->same_pgrp
)
2926 *exit_status
= EXIT_SETSID
;
2927 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2930 exec_context_tty_reset(context
, params
);
2932 if (unit_shall_confirm_spawn(unit
)) {
2933 const char *vc
= params
->confirm_spawn
;
2934 _cleanup_free_
char *cmdline
= NULL
;
2936 cmdline
= exec_command_line(command
->argv
);
2938 *exit_status
= EXIT_MEMORY
;
2942 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2943 if (r
!= CONFIRM_EXECUTE
) {
2944 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2945 *exit_status
= EXIT_SUCCESS
;
2948 *exit_status
= EXIT_CONFIRM
;
2949 log_unit_error(unit
, "Execution cancelled by the user");
2954 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2955 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2956 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2957 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2958 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2959 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2960 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2961 *exit_status
= EXIT_MEMORY
;
2962 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2965 if (context
->dynamic_user
&& dcreds
) {
2966 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2968 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2969 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2970 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2971 *exit_status
= EXIT_USER
;
2972 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2975 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2977 *exit_status
= EXIT_MEMORY
;
2981 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2983 *exit_status
= EXIT_USER
;
2985 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2988 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2991 if (!uid_is_valid(uid
)) {
2992 *exit_status
= EXIT_USER
;
2993 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
2997 if (!gid_is_valid(gid
)) {
2998 *exit_status
= EXIT_USER
;
2999 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3004 username
= dcreds
->user
->name
;
3007 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3009 *exit_status
= EXIT_USER
;
3010 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3013 r
= get_fixed_group(context
, &groupname
, &gid
);
3015 *exit_status
= EXIT_GROUP
;
3016 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3020 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3021 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3022 &supplementary_gids
, &ngids
);
3024 *exit_status
= EXIT_GROUP
;
3025 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3028 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3030 *exit_status
= EXIT_USER
;
3031 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3034 user_lookup_fd
= safe_close(user_lookup_fd
);
3036 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3038 *exit_status
= EXIT_CHDIR
;
3039 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3042 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3043 * must sure to drop O_NONBLOCK */
3045 (void) fd_nonblock(socket_fd
, false);
3047 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3048 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3049 if (params
->cgroup_path
) {
3050 _cleanup_free_
char *p
= NULL
;
3052 r
= exec_parameters_get_cgroup_path(params
, &p
);
3054 *exit_status
= EXIT_CGROUP
;
3055 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3058 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3060 *exit_status
= EXIT_CGROUP
;
3061 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3065 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3066 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3068 *exit_status
= EXIT_NETWORK
;
3069 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3073 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3075 *exit_status
= EXIT_STDIN
;
3076 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3079 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3081 *exit_status
= EXIT_STDOUT
;
3082 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3085 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3087 *exit_status
= EXIT_STDERR
;
3088 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3091 if (context
->oom_score_adjust_set
) {
3092 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3093 * prohibit write access to this file, and we shouldn't trip up over that. */
3094 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3095 if (IN_SET(r
, -EPERM
, -EACCES
))
3096 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3098 *exit_status
= EXIT_OOM_ADJUST
;
3099 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3103 if (context
->nice_set
)
3104 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3105 *exit_status
= EXIT_NICE
;
3106 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3109 if (context
->cpu_sched_set
) {
3110 struct sched_param param
= {
3111 .sched_priority
= context
->cpu_sched_priority
,
3114 r
= sched_setscheduler(0,
3115 context
->cpu_sched_policy
|
3116 (context
->cpu_sched_reset_on_fork
?
3117 SCHED_RESET_ON_FORK
: 0),
3120 *exit_status
= EXIT_SETSCHEDULER
;
3121 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3125 if (context
->cpuset
)
3126 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3127 *exit_status
= EXIT_CPUAFFINITY
;
3128 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3131 if (context
->ioprio_set
)
3132 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3133 *exit_status
= EXIT_IOPRIO
;
3134 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3137 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3138 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3139 *exit_status
= EXIT_TIMERSLACK
;
3140 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3143 if (context
->personality
!= PERSONALITY_INVALID
) {
3144 r
= safe_personality(context
->personality
);
3146 *exit_status
= EXIT_PERSONALITY
;
3147 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3151 if (context
->utmp_id
)
3152 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3154 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3155 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3159 if (context
->user
) {
3160 r
= chown_terminal(STDIN_FILENO
, uid
);
3162 *exit_status
= EXIT_STDIN
;
3163 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3167 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3168 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3169 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3170 * touch a single hierarchy too. */
3171 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3172 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3174 *exit_status
= EXIT_CGROUP
;
3175 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3179 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3180 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3182 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3185 r
= build_environment(
3197 *exit_status
= EXIT_MEMORY
;
3201 r
= build_pass_environment(context
, &pass_env
);
3203 *exit_status
= EXIT_MEMORY
;
3207 accum_env
= strv_env_merge(5,
3208 params
->environment
,
3211 context
->environment
,
3215 *exit_status
= EXIT_MEMORY
;
3218 accum_env
= strv_env_clean(accum_env
);
3220 (void) umask(context
->umask
);
3222 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3224 *exit_status
= EXIT_KEYRING
;
3225 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3228 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3229 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3231 /* 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 */
3232 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3234 /* 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 */
3235 if (needs_ambient_hack
)
3236 needs_setuid
= false;
3238 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3240 if (needs_sandboxing
) {
3241 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3242 * present. The actual MAC context application will happen later, as late as possible, to avoid
3243 * impacting our own code paths. */
3246 use_selinux
= mac_selinux_use();
3249 use_smack
= mac_smack_use();
3252 use_apparmor
= mac_apparmor_use();
3256 if (needs_sandboxing
) {
3259 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3260 * is set here. (See below.) */
3262 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3264 *exit_status
= EXIT_LIMITS
;
3265 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3271 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3272 * wins here. (See above.) */
3274 if (context
->pam_name
&& username
) {
3275 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3277 *exit_status
= EXIT_PAM
;
3278 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3283 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3285 if (ns_type_supported(NAMESPACE_NET
)) {
3286 r
= setup_netns(runtime
->netns_storage_socket
);
3288 *exit_status
= EXIT_NETWORK
;
3289 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3291 } else if (context
->network_namespace_path
) {
3292 *exit_status
= EXIT_NETWORK
;
3293 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
), "NetworkNamespacePath= is not supported, refusing.");
3295 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3298 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3299 if (needs_mount_namespace
) {
3300 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3302 *exit_status
= EXIT_NAMESPACE
;
3303 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3307 if (context
->protect_hostname
) {
3308 if (ns_type_supported(NAMESPACE_UTS
)) {
3309 if (unshare(CLONE_NEWUTS
) < 0) {
3310 *exit_status
= EXIT_NAMESPACE
;
3311 return log_unit_error_errno(unit
, errno
, "Failed to set up UTS namespacing: %m");
3314 log_unit_warning(unit
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
3316 r
= seccomp_protect_hostname();
3318 *exit_status
= EXIT_SECCOMP
;
3319 return log_unit_error_errno(unit
, r
, "Failed to apply hostname restrictions: %m");
3324 /* Drop groups as early as possbile */
3326 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3328 *exit_status
= EXIT_GROUP
;
3329 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3333 if (needs_sandboxing
) {
3335 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3336 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3338 *exit_status
= EXIT_SELINUX_CONTEXT
;
3339 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3344 if (context
->private_users
) {
3345 r
= setup_private_users(uid
, gid
);
3347 *exit_status
= EXIT_USER
;
3348 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3353 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3354 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3355 * however if we have it as we want to keep it open until the final execve(). */
3357 if (params
->exec_fd
>= 0) {
3358 exec_fd
= params
->exec_fd
;
3360 if (exec_fd
< 3 + (int) n_fds
) {
3363 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3364 * process we are about to execute. */
3366 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3368 *exit_status
= EXIT_FDS
;
3369 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3372 safe_close(exec_fd
);
3375 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3376 r
= fd_cloexec(exec_fd
, true);
3378 *exit_status
= EXIT_FDS
;
3379 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3383 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3384 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3385 fds_with_exec_fd
[n_fds
] = exec_fd
;
3386 n_fds_with_exec_fd
= n_fds
+ 1;
3388 fds_with_exec_fd
= fds
;
3389 n_fds_with_exec_fd
= n_fds
;
3392 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3394 r
= shift_fds(fds
, n_fds
);
3396 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3398 *exit_status
= EXIT_FDS
;
3399 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3402 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3403 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3404 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3407 secure_bits
= context
->secure_bits
;
3409 if (needs_sandboxing
) {
3412 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3413 * requested. (Note this is placed after the general resource limit initialization, see
3414 * above, in order to take precedence.) */
3415 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3416 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3417 *exit_status
= EXIT_LIMITS
;
3418 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3423 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3424 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3426 r
= setup_smack(context
, command
);
3428 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3429 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3434 bset
= context
->capability_bounding_set
;
3435 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3436 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3437 * instead of us doing that */
3438 if (needs_ambient_hack
)
3439 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3440 (UINT64_C(1) << CAP_SETUID
) |
3441 (UINT64_C(1) << CAP_SETGID
);
3443 if (!cap_test_all(bset
)) {
3444 r
= capability_bounding_set_drop(bset
, false);
3446 *exit_status
= EXIT_CAPABILITIES
;
3447 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3451 /* This is done before enforce_user, but ambient set
3452 * does not survive over setresuid() if keep_caps is not set. */
3453 if (!needs_ambient_hack
&&
3454 context
->capability_ambient_set
!= 0) {
3455 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3457 *exit_status
= EXIT_CAPABILITIES
;
3458 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3464 if (context
->user
) {
3465 r
= enforce_user(context
, uid
);
3467 *exit_status
= EXIT_USER
;
3468 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3471 if (!needs_ambient_hack
&&
3472 context
->capability_ambient_set
!= 0) {
3474 /* Fix the ambient capabilities after user change. */
3475 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3477 *exit_status
= EXIT_CAPABILITIES
;
3478 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3481 /* If we were asked to change user and ambient capabilities
3482 * were requested, we had to add keep-caps to the securebits
3483 * so that we would maintain the inherited capability set
3484 * through the setresuid(). Make sure that the bit is added
3485 * also to the context secure_bits so that we don't try to
3486 * drop the bit away next. */
3488 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3493 /* Apply working directory here, because the working directory might be on NFS and only the user running
3494 * this service might have the correct privilege to change to the working directory */
3495 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3497 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3499 if (needs_sandboxing
) {
3500 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3501 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3502 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3503 * are restricted. */
3507 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3510 r
= setexeccon(exec_context
);
3512 *exit_status
= EXIT_SELINUX_CONTEXT
;
3513 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3520 if (use_apparmor
&& context
->apparmor_profile
) {
3521 r
= aa_change_onexec(context
->apparmor_profile
);
3522 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3523 *exit_status
= EXIT_APPARMOR_PROFILE
;
3524 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3529 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3530 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3531 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3532 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3533 *exit_status
= EXIT_SECUREBITS
;
3534 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3537 if (context_has_no_new_privileges(context
))
3538 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3539 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3540 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3544 r
= apply_address_families(unit
, context
);
3546 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3547 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3550 r
= apply_memory_deny_write_execute(unit
, context
);
3552 *exit_status
= EXIT_SECCOMP
;
3553 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3556 r
= apply_restrict_realtime(unit
, context
);
3558 *exit_status
= EXIT_SECCOMP
;
3559 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3562 r
= apply_restrict_namespaces(unit
, context
);
3564 *exit_status
= EXIT_SECCOMP
;
3565 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3568 r
= apply_protect_sysctl(unit
, context
);
3570 *exit_status
= EXIT_SECCOMP
;
3571 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3574 r
= apply_protect_kernel_modules(unit
, context
);
3576 *exit_status
= EXIT_SECCOMP
;
3577 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3580 r
= apply_private_devices(unit
, context
);
3582 *exit_status
= EXIT_SECCOMP
;
3583 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3586 r
= apply_syscall_archs(unit
, context
);
3588 *exit_status
= EXIT_SECCOMP
;
3589 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3592 r
= apply_lock_personality(unit
, context
);
3594 *exit_status
= EXIT_SECCOMP
;
3595 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3598 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3599 * by the filter as little as possible. */
3600 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3602 *exit_status
= EXIT_SECCOMP
;
3603 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3608 if (!strv_isempty(context
->unset_environment
)) {
3611 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3613 *exit_status
= EXIT_MEMORY
;
3617 strv_free_and_replace(accum_env
, ee
);
3620 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3621 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3622 if (!replaced_argv
) {
3623 *exit_status
= EXIT_MEMORY
;
3626 final_argv
= replaced_argv
;
3628 final_argv
= command
->argv
;
3630 if (DEBUG_LOGGING
) {
3631 _cleanup_free_
char *line
;
3633 line
= exec_command_line(final_argv
);
3635 log_struct(LOG_DEBUG
,
3636 "EXECUTABLE=%s", command
->path
,
3637 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3639 LOG_UNIT_INVOCATION_ID(unit
));
3645 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3646 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3648 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3649 *exit_status
= EXIT_EXEC
;
3650 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3654 execve(command
->path
, final_argv
, accum_env
);
3660 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3661 * that POLLHUP on it no longer means execve() succeeded. */
3663 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3664 *exit_status
= EXIT_EXEC
;
3665 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3669 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3670 log_struct_errno(LOG_INFO
, r
,
3671 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3673 LOG_UNIT_INVOCATION_ID(unit
),
3674 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3676 "EXECUTABLE=%s", command
->path
);
3680 *exit_status
= EXIT_EXEC
;
3681 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3684 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3685 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3687 int exec_spawn(Unit
*unit
,
3688 ExecCommand
*command
,
3689 const ExecContext
*context
,
3690 const ExecParameters
*params
,
3691 ExecRuntime
*runtime
,
3692 DynamicCreds
*dcreds
,
3695 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3696 _cleanup_free_
char *subcgroup_path
= NULL
;
3697 _cleanup_strv_free_
char **files_env
= NULL
;
3698 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3699 _cleanup_free_
char *line
= NULL
;
3707 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3709 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3710 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3711 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3713 if (params
->n_socket_fds
> 1) {
3714 log_unit_error(unit
, "Got more than one socket.");
3718 if (params
->n_socket_fds
== 0) {
3719 log_unit_error(unit
, "Got no socket.");
3723 socket_fd
= params
->fds
[0];
3727 n_socket_fds
= params
->n_socket_fds
;
3728 n_storage_fds
= params
->n_storage_fds
;
3731 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3733 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3735 r
= exec_context_load_environment(unit
, context
, &files_env
);
3737 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3739 line
= exec_command_line(command
->argv
);
3743 log_struct(LOG_DEBUG
,
3744 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3745 "EXECUTABLE=%s", command
->path
,
3747 LOG_UNIT_INVOCATION_ID(unit
));
3749 if (params
->cgroup_path
) {
3750 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3752 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3753 if (r
> 0) { /* We are using a child cgroup */
3754 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3756 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3762 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3765 int exit_status
= EXIT_SUCCESS
;
3767 r
= exec_child(unit
,
3779 unit
->manager
->user_lookup_fds
[1],
3783 log_struct_errno(LOG_ERR
, r
,
3784 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3786 LOG_UNIT_INVOCATION_ID(unit
),
3787 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3788 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3790 "EXECUTABLE=%s", command
->path
);
3795 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3797 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3798 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3799 * process will be killed too). */
3801 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3803 exec_status_start(&command
->exec_status
, pid
);
3809 void exec_context_init(ExecContext
*c
) {
3810 ExecDirectoryType i
;
3815 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3816 c
->cpu_sched_policy
= SCHED_OTHER
;
3817 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3818 c
->syslog_level_prefix
= true;
3819 c
->ignore_sigpipe
= true;
3820 c
->timer_slack_nsec
= NSEC_INFINITY
;
3821 c
->personality
= PERSONALITY_INVALID
;
3822 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3823 c
->directories
[i
].mode
= 0755;
3824 c
->capability_bounding_set
= CAP_ALL
;
3825 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3826 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3827 c
->log_level_max
= -1;
3830 void exec_context_done(ExecContext
*c
) {
3831 ExecDirectoryType i
;
3836 c
->environment
= strv_free(c
->environment
);
3837 c
->environment_files
= strv_free(c
->environment_files
);
3838 c
->pass_environment
= strv_free(c
->pass_environment
);
3839 c
->unset_environment
= strv_free(c
->unset_environment
);
3841 rlimit_free_all(c
->rlimit
);
3843 for (l
= 0; l
< 3; l
++) {
3844 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3845 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3848 c
->working_directory
= mfree(c
->working_directory
);
3849 c
->root_directory
= mfree(c
->root_directory
);
3850 c
->root_image
= mfree(c
->root_image
);
3851 c
->tty_path
= mfree(c
->tty_path
);
3852 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3853 c
->user
= mfree(c
->user
);
3854 c
->group
= mfree(c
->group
);
3856 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3858 c
->pam_name
= mfree(c
->pam_name
);
3860 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3861 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3862 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3864 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3865 c
->bind_mounts
= NULL
;
3866 c
->n_bind_mounts
= 0;
3867 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3868 c
->temporary_filesystems
= NULL
;
3869 c
->n_temporary_filesystems
= 0;
3871 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3873 c
->utmp_id
= mfree(c
->utmp_id
);
3874 c
->selinux_context
= mfree(c
->selinux_context
);
3875 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3876 c
->smack_process_label
= mfree(c
->smack_process_label
);
3878 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3879 c
->syscall_archs
= set_free(c
->syscall_archs
);
3880 c
->address_families
= set_free(c
->address_families
);
3882 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3883 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3885 c
->log_level_max
= -1;
3887 exec_context_free_log_extra_fields(c
);
3889 c
->log_rate_limit_interval_usec
= 0;
3890 c
->log_rate_limit_burst
= 0;
3892 c
->stdin_data
= mfree(c
->stdin_data
);
3893 c
->stdin_data_size
= 0;
3895 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
3898 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3903 if (!runtime_prefix
)
3906 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3907 _cleanup_free_
char *p
;
3909 p
= strjoin(runtime_prefix
, "/", *i
);
3913 /* We execute this synchronously, since we need to be sure this is gone when we start the service
3915 (void) rm_rf(p
, REMOVE_ROOT
);
3921 static void exec_command_done(ExecCommand
*c
) {
3924 c
->path
= mfree(c
->path
);
3925 c
->argv
= strv_free(c
->argv
);
3928 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3931 for (i
= 0; i
< n
; i
++)
3932 exec_command_done(c
+i
);
3935 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3939 LIST_REMOVE(command
, c
, i
);
3940 exec_command_done(i
);
3947 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3950 for (i
= 0; i
< n
; i
++)
3951 c
[i
] = exec_command_free_list(c
[i
]);
3954 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3957 for (i
= 0; i
< n
; i
++)
3958 exec_status_reset(&c
[i
].exec_status
);
3961 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3964 for (i
= 0; i
< n
; i
++) {
3967 LIST_FOREACH(command
, z
, c
[i
])
3968 exec_status_reset(&z
->exec_status
);
3972 typedef struct InvalidEnvInfo
{
3977 static void invalid_env(const char *p
, void *userdata
) {
3978 InvalidEnvInfo
*info
= userdata
;
3980 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3983 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3989 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3992 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3995 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3998 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4001 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4004 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4011 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]) {
4013 const char* stdio_fdname
[3];
4019 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4020 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4021 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4023 for (i
= 0; i
< 3; i
++)
4024 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4026 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4028 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4029 if (named_iofds
[STDIN_FILENO
] < 0 &&
4030 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4031 stdio_fdname
[STDIN_FILENO
] &&
4032 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4034 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4037 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4038 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4039 stdio_fdname
[STDOUT_FILENO
] &&
4040 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4042 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4045 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4046 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4047 stdio_fdname
[STDERR_FILENO
] &&
4048 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4050 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4054 return targets
== 0 ? 0 : -ENOENT
;
4057 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4058 char **i
, **r
= NULL
;
4063 STRV_FOREACH(i
, c
->environment_files
) {
4067 bool ignore
= false;
4069 _cleanup_globfree_ glob_t pglob
= {};
4078 if (!path_is_absolute(fn
)) {
4086 /* Filename supports globbing, take all matching files */
4087 k
= safe_glob(fn
, 0, &pglob
);
4096 /* When we don't match anything, -ENOENT should be returned */
4097 assert(pglob
.gl_pathc
> 0);
4099 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4100 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4108 /* Log invalid environment variables with filename */
4110 InvalidEnvInfo info
= {
4112 .path
= pglob
.gl_pathv
[n
]
4115 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4123 m
= strv_env_merge(2, r
, p
);
4139 static bool tty_may_match_dev_console(const char *tty
) {
4140 _cleanup_free_
char *resolved
= NULL
;
4145 tty
= skip_dev_prefix(tty
);
4147 /* trivial identity? */
4148 if (streq(tty
, "console"))
4151 if (resolve_dev_console(&resolved
) < 0)
4152 return true; /* if we could not resolve, assume it may */
4154 /* "tty0" means the active VC, so it may be the same sometimes */
4155 return streq(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4158 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4160 return (ec
->tty_reset
||
4162 ec
->tty_vt_disallocate
||
4163 is_terminal_input(ec
->std_input
) ||
4164 is_terminal_output(ec
->std_output
) ||
4165 is_terminal_output(ec
->std_error
)) &&
4166 tty_may_match_dev_console(exec_context_tty_path(ec
));
4169 static void strv_fprintf(FILE *f
, char **l
) {
4175 fprintf(f
, " %s", *g
);
4178 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4179 ExecDirectoryType dt
;
4187 prefix
= strempty(prefix
);
4191 "%sWorkingDirectory: %s\n"
4192 "%sRootDirectory: %s\n"
4193 "%sNonBlocking: %s\n"
4194 "%sPrivateTmp: %s\n"
4195 "%sPrivateDevices: %s\n"
4196 "%sProtectKernelTunables: %s\n"
4197 "%sProtectKernelModules: %s\n"
4198 "%sProtectControlGroups: %s\n"
4199 "%sPrivateNetwork: %s\n"
4200 "%sPrivateUsers: %s\n"
4201 "%sProtectHome: %s\n"
4202 "%sProtectSystem: %s\n"
4203 "%sMountAPIVFS: %s\n"
4204 "%sIgnoreSIGPIPE: %s\n"
4205 "%sMemoryDenyWriteExecute: %s\n"
4206 "%sRestrictRealtime: %s\n"
4207 "%sKeyringMode: %s\n"
4208 "%sProtectHostname: %s\n",
4210 prefix
, c
->working_directory
? c
->working_directory
: "/",
4211 prefix
, c
->root_directory
? c
->root_directory
: "/",
4212 prefix
, yes_no(c
->non_blocking
),
4213 prefix
, yes_no(c
->private_tmp
),
4214 prefix
, yes_no(c
->private_devices
),
4215 prefix
, yes_no(c
->protect_kernel_tunables
),
4216 prefix
, yes_no(c
->protect_kernel_modules
),
4217 prefix
, yes_no(c
->protect_control_groups
),
4218 prefix
, yes_no(c
->private_network
),
4219 prefix
, yes_no(c
->private_users
),
4220 prefix
, protect_home_to_string(c
->protect_home
),
4221 prefix
, protect_system_to_string(c
->protect_system
),
4222 prefix
, yes_no(c
->mount_apivfs
),
4223 prefix
, yes_no(c
->ignore_sigpipe
),
4224 prefix
, yes_no(c
->memory_deny_write_execute
),
4225 prefix
, yes_no(c
->restrict_realtime
),
4226 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4227 prefix
, yes_no(c
->protect_hostname
));
4230 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4232 STRV_FOREACH(e
, c
->environment
)
4233 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4235 STRV_FOREACH(e
, c
->environment_files
)
4236 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4238 STRV_FOREACH(e
, c
->pass_environment
)
4239 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4241 STRV_FOREACH(e
, c
->unset_environment
)
4242 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4244 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4246 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4247 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4249 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4250 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4258 if (c
->oom_score_adjust_set
)
4260 "%sOOMScoreAdjust: %i\n",
4261 prefix
, c
->oom_score_adjust
);
4263 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4265 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4266 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4267 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4268 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4271 if (c
->ioprio_set
) {
4272 _cleanup_free_
char *class_str
= NULL
;
4274 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4276 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4278 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4281 if (c
->cpu_sched_set
) {
4282 _cleanup_free_
char *policy_str
= NULL
;
4284 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4286 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4289 "%sCPUSchedulingPriority: %i\n"
4290 "%sCPUSchedulingResetOnFork: %s\n",
4291 prefix
, c
->cpu_sched_priority
,
4292 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4296 fprintf(f
, "%sCPUAffinity:", prefix
);
4297 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4298 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4299 fprintf(f
, " %u", i
);
4303 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4304 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4307 "%sStandardInput: %s\n"
4308 "%sStandardOutput: %s\n"
4309 "%sStandardError: %s\n",
4310 prefix
, exec_input_to_string(c
->std_input
),
4311 prefix
, exec_output_to_string(c
->std_output
),
4312 prefix
, exec_output_to_string(c
->std_error
));
4314 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4315 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4316 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4317 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4318 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4319 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4321 if (c
->std_input
== EXEC_INPUT_FILE
)
4322 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4323 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4324 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4325 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4326 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4327 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4328 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4329 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4330 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4336 "%sTTYVHangup: %s\n"
4337 "%sTTYVTDisallocate: %s\n",
4338 prefix
, c
->tty_path
,
4339 prefix
, yes_no(c
->tty_reset
),
4340 prefix
, yes_no(c
->tty_vhangup
),
4341 prefix
, yes_no(c
->tty_vt_disallocate
));
4343 if (IN_SET(c
->std_output
,
4346 EXEC_OUTPUT_JOURNAL
,
4347 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4348 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4349 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4350 IN_SET(c
->std_error
,
4353 EXEC_OUTPUT_JOURNAL
,
4354 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4355 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4356 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4358 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4360 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4362 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4364 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4366 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4369 if (c
->log_level_max
>= 0) {
4370 _cleanup_free_
char *t
= NULL
;
4372 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4374 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4377 if (c
->log_rate_limit_interval_usec
> 0) {
4378 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4381 "%sLogRateLimitIntervalSec: %s\n",
4382 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4385 if (c
->log_rate_limit_burst
> 0)
4386 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4388 if (c
->n_log_extra_fields
> 0) {
4391 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4392 fprintf(f
, "%sLogExtraFields: ", prefix
);
4393 fwrite(c
->log_extra_fields
[j
].iov_base
,
4394 1, c
->log_extra_fields
[j
].iov_len
,
4400 if (c
->secure_bits
) {
4401 _cleanup_free_
char *str
= NULL
;
4403 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4405 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4408 if (c
->capability_bounding_set
!= CAP_ALL
) {
4409 _cleanup_free_
char *str
= NULL
;
4411 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4413 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4416 if (c
->capability_ambient_set
!= 0) {
4417 _cleanup_free_
char *str
= NULL
;
4419 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4421 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4425 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4427 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4429 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4431 if (!strv_isempty(c
->supplementary_groups
)) {
4432 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4433 strv_fprintf(f
, c
->supplementary_groups
);
4438 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4440 if (!strv_isempty(c
->read_write_paths
)) {
4441 fprintf(f
, "%sReadWritePaths:", prefix
);
4442 strv_fprintf(f
, c
->read_write_paths
);
4446 if (!strv_isempty(c
->read_only_paths
)) {
4447 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4448 strv_fprintf(f
, c
->read_only_paths
);
4452 if (!strv_isempty(c
->inaccessible_paths
)) {
4453 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4454 strv_fprintf(f
, c
->inaccessible_paths
);
4458 if (c
->n_bind_mounts
> 0)
4459 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4460 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4461 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4462 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4463 c
->bind_mounts
[i
].source
,
4464 c
->bind_mounts
[i
].destination
,
4465 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4467 if (c
->n_temporary_filesystems
> 0)
4468 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4469 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4471 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4473 isempty(t
->options
) ? "" : ":",
4474 strempty(t
->options
));
4479 "%sUtmpIdentifier: %s\n",
4480 prefix
, c
->utmp_id
);
4482 if (c
->selinux_context
)
4484 "%sSELinuxContext: %s%s\n",
4485 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4487 if (c
->apparmor_profile
)
4489 "%sAppArmorProfile: %s%s\n",
4490 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4492 if (c
->smack_process_label
)
4494 "%sSmackProcessLabel: %s%s\n",
4495 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4497 if (c
->personality
!= PERSONALITY_INVALID
)
4499 "%sPersonality: %s\n",
4500 prefix
, strna(personality_to_string(c
->personality
)));
4503 "%sLockPersonality: %s\n",
4504 prefix
, yes_no(c
->lock_personality
));
4506 if (c
->syscall_filter
) {
4514 "%sSystemCallFilter: ",
4517 if (!c
->syscall_whitelist
)
4521 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4522 _cleanup_free_
char *name
= NULL
;
4523 const char *errno_name
= NULL
;
4524 int num
= PTR_TO_INT(val
);
4531 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4532 fputs(strna(name
), f
);
4535 errno_name
= errno_to_name(num
);
4537 fprintf(f
, ":%s", errno_name
);
4539 fprintf(f
, ":%d", num
);
4547 if (c
->syscall_archs
) {
4554 "%sSystemCallArchitectures:",
4558 SET_FOREACH(id
, c
->syscall_archs
, j
)
4559 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4564 if (exec_context_restrict_namespaces_set(c
)) {
4565 _cleanup_free_
char *s
= NULL
;
4567 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4569 fprintf(f
, "%sRestrictNamespaces: %s\n",
4573 if (c
->network_namespace_path
)
4575 "%sNetworkNamespacePath: %s\n",
4576 prefix
, c
->network_namespace_path
);
4578 if (c
->syscall_errno
> 0) {
4579 const char *errno_name
;
4581 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4583 errno_name
= errno_to_name(c
->syscall_errno
);
4585 fprintf(f
, "%s\n", errno_name
);
4587 fprintf(f
, "%d\n", c
->syscall_errno
);
4591 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4594 /* Returns true if the process forked off would run under
4595 * an unchanged UID or as root. */
4600 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4606 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4614 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4616 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4621 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4626 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4627 free(c
->log_extra_fields
[l
].iov_base
);
4628 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4629 c
->n_log_extra_fields
= 0;
4632 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4639 dual_timestamp_get(&s
->start_timestamp
);
4642 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4645 if (s
->pid
!= pid
) {
4651 dual_timestamp_get(&s
->exit_timestamp
);
4657 if (context
->utmp_id
)
4658 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4660 exec_context_tty_reset(context
, NULL
);
4664 void exec_status_reset(ExecStatus
*s
) {
4667 *s
= (ExecStatus
) {};
4670 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4671 char buf
[FORMAT_TIMESTAMP_MAX
];
4679 prefix
= strempty(prefix
);
4682 "%sPID: "PID_FMT
"\n",
4685 if (dual_timestamp_is_set(&s
->start_timestamp
))
4687 "%sStart Timestamp: %s\n",
4688 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4690 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4692 "%sExit Timestamp: %s\n"
4694 "%sExit Status: %i\n",
4695 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4696 prefix
, sigchld_code_to_string(s
->code
),
4700 static char *exec_command_line(char **argv
) {
4708 STRV_FOREACH(a
, argv
)
4716 STRV_FOREACH(a
, argv
) {
4723 if (strpbrk(*a
, WHITESPACE
)) {
4734 /* FIXME: this doesn't really handle arguments that have
4735 * spaces and ticks in them */
4740 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4741 _cleanup_free_
char *cmd
= NULL
;
4742 const char *prefix2
;
4747 prefix
= strempty(prefix
);
4748 prefix2
= strjoina(prefix
, "\t");
4750 cmd
= exec_command_line(c
->argv
);
4752 "%sCommand Line: %s\n",
4753 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4755 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4758 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4761 prefix
= strempty(prefix
);
4763 LIST_FOREACH(command
, c
, c
)
4764 exec_command_dump(c
, f
, prefix
);
4767 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4774 /* It's kind of important, that we keep the order here */
4775 LIST_FIND_TAIL(command
, *l
, end
);
4776 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4781 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4789 l
= strv_new_ap(path
, ap
);
4801 free_and_replace(c
->path
, p
);
4803 return strv_free_and_replace(c
->argv
, l
);
4806 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4807 _cleanup_strv_free_
char **l
= NULL
;
4815 l
= strv_new_ap(path
, ap
);
4821 r
= strv_extend_strv(&c
->argv
, l
, false);
4828 static void *remove_tmpdir_thread(void *p
) {
4829 _cleanup_free_
char *path
= p
;
4831 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4835 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4842 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4844 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4845 if (destroy
&& rt
->tmp_dir
) {
4846 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4848 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4850 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4857 if (destroy
&& rt
->var_tmp_dir
) {
4858 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4860 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4862 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4863 free(rt
->var_tmp_dir
);
4866 rt
->var_tmp_dir
= NULL
;
4869 rt
->id
= mfree(rt
->id
);
4870 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4871 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4872 safe_close_pair(rt
->netns_storage_socket
);
4876 static void exec_runtime_freep(ExecRuntime
**rt
) {
4877 (void) exec_runtime_free(*rt
, false);
4880 static int exec_runtime_allocate(ExecRuntime
**ret
) {
4885 n
= new(ExecRuntime
, 1);
4889 *n
= (ExecRuntime
) {
4890 .netns_storage_socket
= { -1, -1 },
4897 static int exec_runtime_add(
4900 const char *tmp_dir
,
4901 const char *var_tmp_dir
,
4902 const int netns_storage_socket
[2],
4903 ExecRuntime
**ret
) {
4905 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4911 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4915 r
= exec_runtime_allocate(&rt
);
4919 rt
->id
= strdup(id
);
4924 rt
->tmp_dir
= strdup(tmp_dir
);
4928 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4929 assert(var_tmp_dir
);
4930 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4931 if (!rt
->var_tmp_dir
)
4935 if (netns_storage_socket
) {
4936 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4937 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4940 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4949 /* do not remove created ExecRuntime object when the operation succeeds. */
4954 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4955 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4956 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
4963 /* It is not necessary to create ExecRuntime object. */
4964 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
4967 if (c
->private_tmp
) {
4968 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
4973 if (c
->private_network
|| c
->network_namespace_path
) {
4974 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
4978 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
4983 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
4987 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
4995 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
4997 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5003 /* If not found, then create a new object. */
5004 r
= exec_runtime_make(m
, c
, id
, &rt
);
5006 /* When r == 0, it is not necessary to create ExecRuntime object. */
5010 /* increment reference counter. */
5016 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5020 assert(rt
->n_ref
> 0);
5026 return exec_runtime_free(rt
, destroy
);
5029 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5037 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5038 fprintf(f
, "exec-runtime=%s", rt
->id
);
5041 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5043 if (rt
->var_tmp_dir
)
5044 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5046 if (rt
->netns_storage_socket
[0] >= 0) {
5049 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5053 fprintf(f
, " netns-socket-0=%i", copy
);
5056 if (rt
->netns_storage_socket
[1] >= 0) {
5059 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5063 fprintf(f
, " netns-socket-1=%i", copy
);
5072 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5073 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5077 /* This is for the migration from old (v237 or earlier) deserialization text.
5078 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5079 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5080 * so or not from the serialized text, then we always creates a new object owned by this. */
5086 /* Manager manages ExecRuntime objects by the unit id.
5087 * So, we omit the serialized text when the unit does not have id (yet?)... */
5088 if (isempty(u
->id
)) {
5089 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5093 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5095 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5099 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5101 r
= exec_runtime_allocate(&rt_create
);
5105 rt_create
->id
= strdup(u
->id
);
5112 if (streq(key
, "tmp-dir")) {
5115 copy
= strdup(value
);
5119 free_and_replace(rt
->tmp_dir
, copy
);
5121 } else if (streq(key
, "var-tmp-dir")) {
5124 copy
= strdup(value
);
5128 free_and_replace(rt
->var_tmp_dir
, copy
);
5130 } else if (streq(key
, "netns-socket-0")) {
5133 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5134 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5138 safe_close(rt
->netns_storage_socket
[0]);
5139 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5141 } else if (streq(key
, "netns-socket-1")) {
5144 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5145 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5149 safe_close(rt
->netns_storage_socket
[1]);
5150 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5154 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5156 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5158 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5162 rt_create
->manager
= u
->manager
;
5171 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5172 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5173 int r
, fd0
= -1, fd1
= -1;
5174 const char *p
, *v
= value
;
5181 n
= strcspn(v
, " ");
5182 id
= strndupa(v
, n
);
5187 v
= startswith(p
, "tmp-dir=");
5189 n
= strcspn(v
, " ");
5190 tmp_dir
= strndupa(v
, n
);
5196 v
= startswith(p
, "var-tmp-dir=");
5198 n
= strcspn(v
, " ");
5199 var_tmp_dir
= strndupa(v
, n
);
5205 v
= startswith(p
, "netns-socket-0=");
5209 n
= strcspn(v
, " ");
5210 buf
= strndupa(v
, n
);
5211 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5212 log_debug("Unable to process exec-runtime netns fd specification.");
5215 fd0
= fdset_remove(fds
, fd0
);
5221 v
= startswith(p
, "netns-socket-1=");
5225 n
= strcspn(v
, " ");
5226 buf
= strndupa(v
, n
);
5227 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5228 log_debug("Unable to process exec-runtime netns fd specification.");
5231 fd1
= fdset_remove(fds
, fd1
);
5236 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5238 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5241 void exec_runtime_vacuum(Manager
*m
) {
5247 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5249 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5253 (void) exec_runtime_free(rt
, false);
5257 void exec_params_clear(ExecParameters
*p
) {
5261 strv_free(p
->environment
);
5264 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5265 [EXEC_INPUT_NULL
] = "null",
5266 [EXEC_INPUT_TTY
] = "tty",
5267 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5268 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5269 [EXEC_INPUT_SOCKET
] = "socket",
5270 [EXEC_INPUT_NAMED_FD
] = "fd",
5271 [EXEC_INPUT_DATA
] = "data",
5272 [EXEC_INPUT_FILE
] = "file",
5275 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5277 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5278 [EXEC_OUTPUT_INHERIT
] = "inherit",
5279 [EXEC_OUTPUT_NULL
] = "null",
5280 [EXEC_OUTPUT_TTY
] = "tty",
5281 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5282 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5283 [EXEC_OUTPUT_KMSG
] = "kmsg",
5284 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5285 [EXEC_OUTPUT_JOURNAL
] = "journal",
5286 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5287 [EXEC_OUTPUT_SOCKET
] = "socket",
5288 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5289 [EXEC_OUTPUT_FILE
] = "file",
5290 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5293 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5295 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5296 [EXEC_UTMP_INIT
] = "init",
5297 [EXEC_UTMP_LOGIN
] = "login",
5298 [EXEC_UTMP_USER
] = "user",
5301 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5303 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5304 [EXEC_PRESERVE_NO
] = "no",
5305 [EXEC_PRESERVE_YES
] = "yes",
5306 [EXEC_PRESERVE_RESTART
] = "restart",
5309 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5311 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5312 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5313 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5314 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5315 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5316 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5319 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5321 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5322 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5323 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5324 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5325 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5326 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5329 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5331 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5332 [EXEC_KEYRING_INHERIT
] = "inherit",
5333 [EXEC_KEYRING_PRIVATE
] = "private",
5334 [EXEC_KEYRING_SHARED
] = "shared",
5337 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);